Discussions around future employability tend to highlight the unique qualities that we, as humans, value. While we might pride ourselves on our emotional intelligence, communication skills and creativity, it leaves a set of skills that would have our secondary school careers advisors directing us all off to retrain in nursing and the creative arts. And, quite honestly, if I have a tricky email to send, Chat GPT does a much better job at writing with immense tact than I do.

Fortunately for us all, these simplifications of such a complex issue overlook some reassuring limitations built into the Transformers architecture, the technology that the latest and most impressive generation of AI is built on. 

The limits of modern AI

These tools have learnt to be literate in the most basic sense. They can predict the next, most logical, token that will please their human audience. The human audience can then connect that representation to something in the real world. There is nothing in the transformers architecture to help answer questions like ‘Where am I right now?’ or ‘What is happening around me?’ 

In business these are often crucial questions. The architecture can’t just be tweaked to add that as an upgrade. Unless someone has already built an alternative architecture in secret somewhere in Silicon Valley, we won’t see a machine that combines Chat GPT with contextual understanding any time soon. 

Where transformers have been revolutionary, it tends to be areas where humans had almost given up the job. Medical research, for example, is a terrifically expensive and failure-ridden process. But using a well-trained transformer to sift through millions of potential substances to identify candidates for human development and testing is making success a more familiar sensation for our medical researchers. But that kind of success can’t be replicated everywhere.

Joining it all up

We, of course, have some wonderful examples of technologies that can actually answer questions like ‘Where am I and what’s going?’ Your satnav, for one, has some idea where you are and of some hazards ahead. More traditional neural networks can look at images of construction sites and spot risk hazards before they become an accident. Machines can look at medical scans and see if cancer is or is not present. 

But these machines are highly specialised. The same AI can’t spot hazards around my home, or in a school. The machine that can spot bowel cancer can’t be used to detect lung cancer. This lack of interaction between highly specialised algorithms means that, for now, AI still needs a human running the show. They must choose which machine to use, and whether to override the suggestions that the machine makes.

AI: Confidently wrong

And that is the other crucial point. Many of the algorithms that are being embedded into our workplace have very poor understanding of their own capabilities. They’re like the teenager who thinks they’re invincible because they haven’t experienced failure and disappointment often enough yet. 

If you train a machine to recognise road signs, it will function very well at recognising clean, clear road signs. We would expect it to struggle more with ‘edge’ cases. Images of dirty, mud-splattered road signs taken at night during a storm, for example, trip up AI where humans succeed. But what if you show it something completely different, like images of foods? 

Unless it has also been taught that images of food are not road signs and need a completely different classification, the machine may well look at a hamburger and come to the conclusion that – of all the labels it can apply – it most clearly represents a stop sign. The machine might make that choice with great confidence – a circle and a line across the middle – it’s obviously not a give way sign! So human oversight to be able to say, ‘Silly machine, that’s a hamburger!’ is essential. 

What does this mean for the next 10 years of your career?

It does not mean the end of your career, unless you are in a very small and unfortunate category of professions. But it does mean that the most complex decisions you have to take today are soon going to become the norm. The ability to make consistent, adaptable, high quality decisions is vital to helping your career to flourish. 

Fortunately for our careers, the world is unlikely to run out of problems to solve any time soon. 

With complex chains of dependencies and huge volatility in world markets, it’s not enough to evolve your intelligence to make more rational decisions (although that will always help – we are, by default, highly emotional decision makers). 

To make great decisions, you need to know what you can’t compute, and what the machines can’t compute. There will be times when external insights from data can support you in decision making. But there will also be intermediaries to coordinate, errors to identify, and competing views on solutions to weigh up. 

All machine intelligence requires compromise, and fortunately, that limitation leaves space for us, but only if we train ourselves to work in this new professional environment. At the Institute of Analytics, we work with professionals to support them in this journey. 

Dr Clare Walsh is a leading academic in the world of  data and AI, advising governments worldwide on ethical AI strategies. The IoA is a global, not-for-profit professional body for analytics and data professionals. It promotes the ethical use of data-driven decision making and offers membership services to individuals and businesses, helping them stay at the cutting edge of analytics and AI technology.

It’s getting to the time of year when priorities suddenly come into sharp focus. Just a few months ago, 2024 was fresh and getting started. Now, the days and weeks are being ticked off the calendar at breakneck speed, and with 2025 within touching distance, many CIOs will be under pressure to deliver before the year is out. 

This isn’t about juggling one or two priorities. Most CIOs are stretched across multiple projects on top of keeping their organisations’ IT systems on track; from delivering large digital transformation projects and fending off cyber attacks, to introducing AI and other innovative tech.

So, where should CIOs put their focus in the last months of 2024, when they face competing priorities and time is tight? How do they strike the right balance between innovation and overall performance? 

We’ve asked a panel of experts to share what they think will make the most impact, when it comes to data.

Get your data in order

Building a strong foundation for current and future projects is a great place to start, according to our specialists. First stop, managing data. Specifically data quality.

“Without the right, accurate data, the rest of your initiatives will be challenging: whether that’s a complex migration, AI innovation or simply operating business as usual,” Syniti MD and SVP EMEA Chris Gorton explains. “Start by getting to know your data, understanding the data that’s business critical and linked to your organisational objectives. Next, set meaningful objectives around accuracy and availability, track your progress and be ready to adjust your approach if needed. Then introduce robust governance your organisation can follow to make sure your data quality remains on track. 

“By putting data first over the next few months, you’ll be in a great position to move forward with those big projects in 2025.”

As well as giving a good base to build from, getting to grips with data governance can also help to protect valuable data. 

Keepit CISO Kim Larsen points out: “When organisations don’t have a clear understanding and mapping of their data and its importance, they cannot protect it or determine which technologies to implement, and therefore preserve that data and determine who has access to it.

“When disaster strikes and they lose access to their data, whether because of cyberattacks, human error or system outages, it’s too late to identify and prioritise which data sets they need to recover to ensure business continuity. Good data governance equals control. In a constantly evolving cyber threat landscape, control is essential.”

Understand the infrastructure you need behind the scenes

Once CIOs are confident of their data quality, infrastructure may well be the next focus: particularly if AI, Machine Learning or other innovative technologies are on the cards for next year. Understanding the infrastructure needed for optimum performance is key, otherwise new tools may fail to deliver the results they promise.

Xinnor CRO Davide Villa explains: “As CIOs implement innovative solutions to drive their businesses forward, it’s crucial to consider the foundation that supports them. Modern workloads like AI, Machine Learning, and Big Data analytics all require rapid data access. In recent years, fast storage has become an integral part of IT strategy, with technologies like NVMe SSDs emerging as powerful tools for high-performance storage.

“However, it’s important to think holistically about how these technologies integrate with existing infrastructures and data protection methods. As you plan for the future, take time to assess your storage needs and explore various solutions. Determine whether traditional storage solutions best suit your workload or if more modern approaches, such as software-based versions of RAID, could enhance flexibility and performance. The goal is to create an infrastructure that not only meets your current demands efficiently but also remains adaptable to future requirements, ensuring your systems can handle evolving workloads’ speed and capacity needs while optimising resource utilisation.”

Protect against cyber attacks…

With threats from AI-powered cyber crime and ransomware increasing, data protection is high on our experts’ priorities.

As a first step, Scality CMO Paul Speciale says “CIOs should assess their existing storage backup solutions to make sure they are truly immutable to provide a baseline of defence against ransomware that threatens to overwrite or delete data. Not all so-called immutable storage is actually safe at all times, so inherently immutable object storage is a must-have.

“Then look beyond immutable storage to stop exfiltration attacks. Mitigating the threat of data exfiltration requires a multi-layered approach for a more comprehensive standard of end-to-end cyber resilience. This builds safeguards at every level of the system – from API to architecture – and closes the door on as many threat vectors as possible.”

Piql founder and MD, Rune Bjerkestrand, agrees: “We rely on trusted digital solutions in almost every aspect of our lives, and business is no exception. And although this offers us many opportunities to innovate, it also makes us vulnerable. Whether those threats are physical, from climate change, terrorism, and war, or virtual, think cyber attack, data manipulation and ransomware, CIOs need to ensure guaranteed, continuous access to authentic data.

“As the year comes to an end, prioritise your critical data and make sure you have the right protection in place to guarantee access to it.”

Understanding the wider cyber crime landscape can also help to identify the most vulnerable parts of an infrastructure, says iTernity CEO Ralf Steinemann. “In these next few months, prioritise business continuity. Strengthen your ransomware protection and focus on the security of your backup data. Given the increasing sophistication and frequency of ransomware attacks, which often target backups, look for solutions that ensure data remains unaltered and recoverable. And consider how you’ll further enhance security by minimising vulnerabilities and reducing the risk of human error.”

Remember edge data

Central storage and infrastructure is a high priority for CIOs. But with the majority of data often created, managed and stored at the edge, it’s incredibly important to get to grips with this critical data.

StorMagic CTO Julian Chesterfield explains: “Often businesses do not apply the same rigorous process for providing high availability and redundancy at the edge as they do in the core datacentre or in the cloud. Plus, with a larger distributed edge infrastructure comes a larger attack surface and increased vulnerabilities. CIOs need to think about how they mitigate that risk and how they deploy trusted and secure infrastructure at their edge locations without compromising integrity of overall IT services.”

Think long term

With all these competing challenges, CIOs must make sure whatever they prioritise supports the wider data strategy, so that the work put in now has long-term benefits, say Pure Storage Field CTO EMEA Patrick Smith. 

“CIO focus should be on a long term strategy to meet these multiple pressures. Don’t fall into the trap of listening to hype and making decisions based on FOMO,” he warns. “Given the uncertainty associated with some new initiatives, consuming infrastructure through an as-a-Service model provides a flexible way to approach these goals. The ability to scale up and down as needed, only pay for what’s being used, and have guarantees baked into the contract should be an appealing proposition.”

Where will you focus?

As we enter the final stretch of 2024, it’s crucial to prioritise and take action. With the right strategies in place focusing on data quality, governance, infrastructure, and security, CIOs will be set up to meet current demands, and build a solid foundation for their organisations in 2025 and beyond. 

Don’t wait for the pressures to mount. The experts agree: start prioritising now, and get ready to thrive in the year ahead.

Interest in implementing AI is truly global and industry-agnostic. However, few companies have established the foundational building blocks that enable AI to generate value at scale. While each organisation and industry will have their own specific challenges that may impact AI adoption, there are four common barriers that all companies tend to encounter: People, Control of AI models, Quality, and Cost. To implement AI successfully and ensure long-term value creation, it’s critical that organisations take steps to address these challenges.

Accessible upskilling 

At the forefront of these challenges is the impending AI skills gap. The speed at which the technology has developed demands attention, with executives estimating that 40% of their workforce will need to re-skill in the next three years as a result of implementing AI – outlying that this is a challenge that requires immediate attention.

To tackle this hurdle, organisations must provide training that is relevant to their needs, while also establishing a culture of continuous learning in their workforce. As the technology continues to evolve and new iterations of tools are introduced, it’s vital that workforces stay up to date on their skills.

Equally important is democratising AI upskilling across the entire organisation – not just focusing on tech roles. Everyone within an organisation, from HR and administrative roles to analysts and data scientists, can benefit from using AI. It’s up to the organisation to ensure learning materials and upskilling initiatives are as widely accessible as possible. However, democratising access to AI shouldn’t be seen as a radical move that instantly prepares a workforce to use AI. Instead, it’s crucial to establish not just what is rolled out, but how this will be done. Organisations should consider their level of AI maturity, making strategic choices about which teams have the right skills for AI and where the greatest need lies. 

Consider AI models

As organisations embrace AI, protecting data and intellectual property becomes paramount. One effective strategy is to shift focus from larger, generic models (LLMs) to smaller, customised language models and move toward agentic or compound AI systems. These purpose-built models offer numerous advantages, including improved accuracy, relevance to specific business needs, and better alignment with industry-specific requirements.

Custom-built models also address efficiency concerns. Training a generalised LLM requires significant resources, including expensive Graphics Processing Units (GPUs). Smaller models require fewer GPUs for training and inference, benefiting businesses aiming to keep costs and energy consumption low.

When building these customised models, organisations should use an open, unified foundation for all their data and governance. A data intelligence platform ensures the quality, accuracy, and accessibility of the data behind language models. This approach democratises data access, enabling employees across the enterprise to query corporate data using natural language, freeing up in-house experts to focus on higher-level, innovative tasks.

The importance of data quality 

Data quality forms the foundation of successful AI implementation. As organisations rush to adopt AI, they must recognise that data serves as the fuel for these systems, directly impacting their accuracy, reliability, and trustworthiness. By leveraging high-quality, organisation-specific data to train smaller, customised models, companies ensure AI outputs are contextually relevant and aligned with their unique needs. This approach not only enhances security and regulatory compliance but also allows for confident AI experimentation while maintaining robust data governance.

Implementing AI hastily without proper data quality assurance can lead to significant challenges. AI hallucinations – instances where models generate false or misleading information – pose a real threat to businesses, potentially resulting in legal issues, reputational damage, or loss of trust. 

By prioritising data quality, organisations can mitigate risks associated with AI adoption while maximising its potential benefits. This approach not only ensures more reliable AI outputs but also builds trust in AI systems among employees, stakeholders, and customers alike, paving the way for successful long-term AI integration.

Managing expenses in AI deployment

For C-suite executives under pressure to reduce spending, data architectures are a key area to examine. While a recent survey found that Generative AI has skyrocketed to the #2 priority for enterprise tech buyers, and 84% of CIOs plan to increase AI/ML budgets, 92% noted they don’t have a budget increase over 10%. This indicates that executives need to plan strategically about how to integrate AI while remaining within cost constraints.

Legacy architectures like data lakes and data warehouses can be cumbersome to operate, leading to information silos and inaccurate, duplicated datasets, ultimately impacting businesses’ bottom lines. While migrating to a scalable data architecture, such as a data lakehouse, comes with an initial cost, it’s an investment in the future. Lakehouses are easier to operate, saving crucial time, and are open platforms, freeing organisations from vendor lock-in. They also simplify the skills needed by data teams as they rationalise their data architecture.

With the right architecture underpinning an AI strategy, organisations should also consider data intelligence platforms to leverage data and AI by being tailored to its specific needs and industry jargon, resulting in more accurate responses. This customisation allows users at all levels to effectively navigate and analyse their enterprise’s data.

Consider the costs, pump the brakes, and take a holistic approach

Before investing in any AI systems, businesses should consider the costs of the data platform on which they will perform their AI use cases. Cloud-based enterprise data platforms are not a one-off expense but form part of a business’ ongoing operational expenditure. The total cost of ownership (TCO) includes various regular costs, such as cloud computing, unplanned downtime, training, and maintenance.

Mitigating these costs isn’t about putting the brakes on AI investment, but rather consolidating and standardising AI systems into one enterprise data platform. This approach brings AI models closer to the data that trains and drives them, removing overheads from operating across multiple systems and platforms.

As organisations navigate the complexities of AI adoption, addressing these four main barriers is crucial. By taking a holistic approach that focuses on upskilling, data governance, customisation, and cost management, companies will be better placed for successful AI integration.  

With the launch of the Labour Government’s Autumn Budget, Sir Kier Starmer’s government and Chancellor Rachel Reeves seem determined to convince Labour voters that the adults are back in charge of the UK’s finances, and convince conservatives that nothing all that fundamental will change. Popular policies like renationalising infrastructure are absent. Some commenters worry that Reeves’ £40 billion tax increase will affect workers in the form of lower wages and slimmer pay rises. 

Nevertheless, tech industry experts have hailed more borrowing, investment, and productivity savings targets across government departments as positive signs for the UK economy. In the wake of the budget’s release, we heard from three leaders in the UK tech sector about their expectations and hopes for the future. 

Growth driven by AI 

Damian Stirrett, Group Vice President & General Manager UK & Ireland at ServiceNow 

“As expected, growth and investment is the underlying message behind the UK Government’s Autumn Budget. When we talk about economic growth, we cannot leave technology out of the equation. We are at an interesting point in time for the UK, where business leaders recognise the great potential of technology as a growth driver leading to impactful business transformation.   

AI is, and will increasingly be, one of the biggest technological drivers behind economic growth in the UK. In fact, recent research from ServiceNow, has found that while the UK’s AI-powered business transformation is in its early days, British businesses are among Europe’s leaders when it comes to AI optimism and maturity, with 85% of those planning to increase investment in AI in the next year. It is clear that appetite for AI continues to grow- from manufacturing to healthcare, and education. Furthermore, with the government setting a 2% productivity savings target for government departments, AI has the potential to play a significant role here, not only by boosting productivity, but driving innovation, reducing operational costs, as well as creating new job opportunities.   

To remain competitive as a country, we must not forget to also invest in education, upskilling initiatives, and partnerships between the public and private sectors, fostering AI innovation to drive transformative change for all.” 

Investing in the industries of the future

By James Hall, Vice President and Country Manager UK&I at Snowflake

“Given the Autumn budget’s focus on investing in industries of the future, AI must be at the forefront of this innovation. This follows the new AI Opportunities Action Plan earlier this year, looking to identify ways to accelerate the use of AI to better people’s lives by improving services and developing new products. Yet, to truly capitalise on AI’s potential, the UK Government must prioritise investments in data infrastructure.

AI systems are only as powerful as the data they’re trained on; making high-quality, accessible data essential for innovation. Robust data-sharing frameworks and platforms enable more accurate AI insights and drive efficiency, which will help the UK remain globally competitive. With the right resources, the UK can lead in offering responsible and effective AI applications. This will benefit both public services and the wider economy, helping to fuel smart industries and meet the growth goals set out by the Chancellor.” 

Growth, stability, and a careful, considered approach 

By Rupal Karia, VP & Country Leader UK&I at Celonis

“Hearing the UK Government’s autumn budget, it’s clear that growth and stability are the biggest messages. With the Chancellor outlining a 2% productivity savings target for government departments, it is crucial the public sector takes heed of the role of technology which cannot be understated as we look to the future. Artificial intelligence is being heralded by businesses, across multiple sectors, as a game-changing phenomenon. Yet for all of the hype, UK businesses must take a step back and consider how to make the most of their AI investments to maximise ROI. 

The UK must complement investments in AI with a strong commitment to process intelligence technology. AI holds transformative potential for both the public and private sectors, but without the relevant context being provided by process intelligence, organisations risk failing to achieve ROI. Process intelligence empowers businesses with full visibility into how internal processes are operating, pinpointing where there are bottlenecks, and then remediates these issues. It is the connective tissue that gives organisations the insight and context they need to drive impactful AI use cases which will help businesses achieve return on AI investment. 

Celonis’ research reveals that UK business leaders believe that getting support with AI implementation would be more important for their businesses than reducing red tape or cutting business rates. This is a clear guideline for the UK government to consider when looking to fuel growth.” 

Prior to the mainstream AI revolution, which started with the public launch of ChatGPT, organisations were already embedding AI in one form or another into security controls for some time. Historically, security product developers have favoured using Machine Learning (ML) in rheir products, dating back to the millennium when intrusion detection systems began to use complex models to identify unusual network traffic.  

Machine learning and security 

Since then, developers have employed ML in many categories of security products, as it excels in organising large data sets. 

If you show a machine learning model a million pictures of a dog, followed by a million pictures of a cat, it can determine with pretty good accuracy whether a new, unseen image is of a dog or a cat. 

This works the same way with ‘legitimate’ and ‘malicious’ data. Today, it would be unusual to find an antivirus product for sale that does not incorporate ML functionality. It works well, and it isn’t easily fooled by slight changes to a virus, for example. This is important with the speed of change in today’s threat landscape. 

LLM security applications 

ChatGPT is a type of Artificial Intelligence that falls under the category of a ‘Large Language Model’ (LLM). LLMs are relatively new to the security market, and there is a rush from vendors to jump on the bandwagon and incorporate this type of AI into their products. 

Two areas of greatest value so far include the ability to summarise complex technical information – such as ingesting the technical details about a security incident and describing it – and how to remediate it, in an easy-to-understand way. 

The reverse is also true, many complex security products which previously required the administrator to learn a complex scripting language to interact with it, can now ask it simple questions in their native language. 

The LLM will ‘translate’ these queries into the specific syntax required by the tool. 

This is enabling organisations to get more value from their junior team members, and reducing the time-to-value for new employees. We’re likely to see some of the ‘heavy lifting’ of repetitive tasks offloaded to AI models.  

LLM AI integration requires organisations to keep both eyes open 

When integrating AI security tools, businesses must establish policies and training to ensure staff can leverage these tools effectively. Protecting sensitive training data and understanding privacy policies are crucial to mitigating data privacy risks. 

Additionally, businesses should keep informed about the latest developments and updates so they can ensure continuous improvement of their AI tools. This approach ensures AI tools augment security while aligning with ethical standards and organisational policies, maintaining the balance between technology and human expertise.  

Finally, organisations must remain vigilant when it comes to developments in regulation. For instance, the EU Artificial Intelligence Act, which will start to take effect over the next 12 months, requires organisations to ensure that their AI systems comply with stringent requirements regarding safety, transparency, and accountability. 

This includes conducting risk assessments, ensuring data quality and robustness, providing clear and understandable information to users, and establishing mechanisms for human oversight and control. Businesses must use documentation AI system activity logging Prior to the mainstream AI revolution, which started with the public launch of ChatGPT, AI in some form had been embedded into security controls for some time. Historically, Machine Learning (ML) has been the category of AI used in security products, dating back to the millennium when intrusion detection systems began to use complex models to identify unusual network traffic.  

Machine learning and security 

Since then, organisations have used ML in many categories of security products, as it excels in organising large data sets. 

If you show a machine learning model a million pictures of a dog, followed by a million pictures of a cat, it can determine with pretty good accuracy whether a new, unseen image is of a dog or a cat. 

This works the same way with ‘legitimate’ and ‘malicious’ data. Today, it would be unusual to find an antivirus product for sale that does not incorporate ML functionality. It works well, and it isn’t easily fooled by slight changes to a virus, for example. This is important with the speed of change in today’s threat landscape. 

LLM security applications 

ChatGPT is a type of Artificial Intelligence that falls under the category of a ‘Large Language Model’ (LLM). LLMs are relatively new to the security market, and there is a rush from vendors to jump on the bandwagon and incorporate this type of AI into their products. 

Two areas of greatest value so far include the ability to summarise complex technical information – such as ingesting the technical details about a security incident and describing it – and how to remediate it, in an easy-to-understand way. 

The reverse is also true, many complex security products which previously required the administrator to learn a complex scripting language to interact with it, can now ask it simple questions in their native language. 

The LLM will ‘translate’ these queries into the specific syntax required by the tool. 

This is enabling organisations to get more value from their junior team members, and reducing the time-to-value for new employees. We’re likely to see companies offload some of the ‘heavy lifting’ of repetitive tasks to AI models. This in turn will free up more time for humans to use their expertise for more complex and interesting tasks that aid staff retention.

These models are also prone to ‘hallucinate’. Whn this happens, AI models make up information that is completely incorrect. Because of this, it’s important not to become overly reliant on AI – using it as an assistant rather than a replacement for expertise, and to avoid becoming exclusively dependent on it.  

LLM AI integration requires organisations to keep both eyes open 

When integrating AI security tools, businesses must establish policies and training to ensure staff can leverage these tools effectively. Protecting sensitive training data and understanding privacy policies are crucial to mitigating data privacy risks. 

Additionally, businesses should keep informed about the latest developments and updates so they can ensure continuous improvement of their AI tools. This approach ensures AI tools augment security while aligning with ethical standards and organisational policies, maintaining the balance between technology and human expertise.  

Finally, organisations must remain vigilant when it comes to developments in regulation. For instance, the EU Artificial Intelligence Act, which will start to take effect over the next 12 months, requires organisations to ensure that their AI systems comply with stringent requirements regarding safety, transparency, and accountability. 

This includes conducting risk assessments, ensuring data quality and robustness, providing clear and understandable information to users, and establishing mechanisms for human oversight and control. Businesses must also maintain thorough documentation and logging of AI system activities to prepare for regular audits and inspections by regulatory authorities.

Do you remember, a few years ago, when all the talk was about us increasingly living in the virtual world? Where mixed reality living, powered by technology such as virtual reality (VR), was going to define how people lived, worked and played? So much so that fashion houses started selling in the virtual world. Estate agents started selling property in the virtual world and virtual conference centres were built so you could attend business events and network from the comfort of your office swivel chair. Futurists were predicting we were going to be living semi-Matrix-style in the near future.

Has it turned out like that? No… or certainly not yet anyway.

VR is just one example of how business is uniquely adept at propagating hype, particularly when it comes to emerging technologies. And you can probably guess where I am heading with this argument… AI.

The AI overhype cycle 

Since ChatGPT exploded into the public consciousness in 2022, I have spoken to scores of business leaders who feel like they need to jump on the AI bandwagon. It’s reflected by the last quarterly results announcements by the S&P 500, with over 40% of companies mentioning AI.  

They are understandably caught in the hype and buzz AI has created, and often think their businesses need to integrate this technology or face being left behind. This is reinforced by a recent BSI survey of over 900 leaders which found 76% believe they will be at a competitive disadvantage unless they invest in AI.

But is that true? The answer may be more nuanced than a simple yes or no.

To be clear, I am not saying the development of AI is anything but seismic. It is recognised by many leading academics as a general purpose technology (GTP). That is to say, it will be a game changer for humanity.

However, at an enterprise level, AI has been overhyped in many quarters, creating a disconnect between reality and expectations. 

Too much money for too little return 

This overhype is leading to two outcomes.

First, leaders feel pressured to be seen using it and heard talking about it. So they dabble with it, often without being certain how it will benefit their business, and how to effectively measure those benefits.

Second, the lack of a proper strategy and metrics is leading to time and resources being wasted. Just 44% of businesses globally have an AI strategy, according to the BSI survey. 

And importantly, if a user has a bad initial experience with a technology, it will often lead to mistrust and plummeting confidence in its future potential. This means it will take even more resources at a future date to effectively leverage the same technology. 

Recent media reporting has provided cases in point. There was the story of a chief marketing officer who abandoned one of Google’s AI tools because they disrupted the company’s advertising strategy so much, while another tool performed no better than a human. Then there was the tale of a chief information officer who dropped Microsoft’s Copilot tool after it created “middle school presentations”.

This disconnect is nothing new. As a consultant, what I often see is a detachment between a company’s business goals and how their technology is set up and operated. Or as in this case, a delta between expectations and delivery capability.

“Keep it simple” and focus on the business basics 

So amid all this noise around AI, my advice to clients is simple: keep in mind it is just another tool, and that the fundamentals of business haven’t changed.

You still need to provide a product or service that someone else wants to buy at a price point that is higher than what it costs to manufacture.

You still need to make a profit.

AI as a business tool may change the process by which we create and deliver value, but those business fundamentals haven’t changed and never will.

So if we recognise AI is just a tool, albeit one with the potential to accelerate the transformation of enterprises, what can leaders do to avoid landing in the gap between the hype and reality? Here are six suggestions:

1. Education

Invest in learning about the technology, its capabilities, the pros and cons, its roadmap and what dependencies AI has for it to be successful. Share this knowledge across the enterprise, so you start to take everyone on a collective journey

2. Build ethical AI policies and governance framework

Ethical AI policy is more than just guardrails to protect your business. It is also the north star that gives your employees, clients, partners, suppliers and investors confidence in what you will do with AI

3. Adopt a strategic approach

Focus on identifying key business problems where AI can be part of the solution. Put in place the appropriate metrics. This will help to prioritise investment and resource allocation

4. Develop your data strategy

AI success is intrinsically linked to data, so build your data strategy. Focus on building a solid data infrastructure and ensuring the quality of your data. This will lay the groundwork for successful AI implementation

5. Foster collaboration 

Consider collaborating with external partners, such as vendors or even universities and research institutions. This collective solving of problems will help provide deep insights into the latest AI developments and best practices

6. Communicate

Given the pace of business evolution nowadays, for most enterprises change management has become a core operational competency. So start your communication and change management early with AI. With its high public profile and fears persisting about AI replacing workers, you want to fill the knowledge gap in your team members so they understand how AI will be used to empower, not replace them. Taking employees on this journey will massively help the chances of success of future AI programmes.

Overall, unless leaders know how to integrate AI in a way that provides business benefits, they are just throwing mud at a wall and hoping some will stick… and all the while the cost base is rapidly increasing as a result of adopting this hugely expensive technology.

So to answer the big question, will a business be at a competitive disadvantage if it doesn’t invest in AI?

Typically, yes it will. But invest in a plan focused on how AI can help achieve longer-term business goals. Its capabilities will continue to emerge and evolve over the coming years, so building the right foundations will help effectively leverage AI both today and tomorrow.  

And ultimately remember that like all technology, AI is just one tool in the business kitbag.

Nigel O’Neill is founder and CEO of Tarralugo.

Since the launch of ChatGPT captured the global imagination, the technology has attreacted questions regarding its workings. Some of these questions stem from a growing interest in the field of AI design. Others are the result of suspicion as to whether AI models are being trained ethically.

Indeed, there’s good reason to have some level of skepticism towards generative AI. After all, current iterations of Large Language Models use underlying technology that’s extremely data-hungry. Even a cursory glance at the amount of information needed to train models like GPT-4 indicates that documents in the public domain were never going to be enough.

But I’m going to leave the ethical and legal questions for better-trained specialists in those specific fields and look at the technical side of AI. The development of generative AI is a fascinating occurence, as several distinct yet closely related disciplines had to progress to the point where such an achievement became possible.

While there are numerous different AI models, each accomplishing a separate goal, most of the current underlying technologies and requirements have many similarities. So, I’ll be focusing on Large Language Models as they’re likely the most familiar version of an AI model to most people.

How do LLMs work?

There are a few key concepts everyone should understand about AI models as I see many of them being conflated into one:

Large Language Model (LLM) is a broad term that describes any language model that uses a large amount of (usually) human-written text and is primarily used to understand and generate human-like language. Every LLM is part of the Natural Language Processing (NLP) field.

A Generative Pre-trained Transformer (GPT) is a type of LLM that was introduced by OpenAI. Unlike some other LLMs, the primary goal was to specifically generate human-like text (hence, “generative”). Pre-trained simply means that the model requires lots of labeled data to function.

Transformer is another part of GPT that people are often confused by. While GPTs were introduced by OpenAI, Transformers were initially developed by Google researchers in a breakthrough paper called “Attention is All You Need”.

One of the major breakthroughs was the implementation of self-attention. This allows a model that uses such a transformer to evaluate all words within it at once. Previous iterations of language models had numerous issues such as putting more emphasis on recent words.

While the underlying technology of a transformer is extremely complex, the basics are that they convert words (for language models) into mathematical vectors of three-dimensional space. Earlier iterations would only convert single words and place them in a three-dimensional space with some prediction if the words are related (such as “king” and “queen” being closer to each other than “cat” and “king”). A transformer is able to evaluate an entire sentence, allowing better contextual understanding.

Almost all current LLMs use transformers as their underlying technology. Some refer to non-OpenAI models as “GPT-like.” However, that may be a bit of an oversimplification. Nevertheless, it’s a handy umbrella term.

Scaling and data

Anyone who has spent some time analysing natural human language will quickly realize that language, as a concept or technology, is one of the most complicated things ever created. In fact, philosophers and linguists still spend decades trying to decipher even small aspects of natural language.

Computers have another problem – they don’t get to experience language as it is. So, like the aforementioned transformers, language has to be converted into a mathematical representation, which poses significant challenges by itself. Couple that with the enormous amount of complexities that our daily use of language has. From humor to ambiguity to domain-specific language – all of that adds to largely unspoken rules most of us understand intuitively.

Intuitive understanding, however, isn’t all that useful when you need to convert those rules into mathematical representations. So, instead of attempting to input rules to machines themselves, the idea was to give them enough data to glean out the intricacies of language. Unavoidably, however, that means that machine learning models have to acquire lots of different expressions, uses, applications, and other aspects of language. There’s simply no way to provide all of these within a single text or even a corpus of texts.

Finally, most machine learning models face scaling law problems. Most business-folk will be familiar with diminishing returns – at some point, each invested dollar into an aspect of business will start generating fewer returns. Machine learning models, GPTs included, face exactly the same issue. To get from 50% accuracy to 60% accuracy, you may need twice as much data and computing power than before. Getting from 90% to 95% may require hundreds of times more data and computing power than before.

Currently, the challenge seems largely unavoidable as it’s simply part of the technology, it can only be optimised.

Web scraping and AI

It should be clear by now that no matter how many books were written before the invention of copyright, there wouldn’t nearly be enough data for models like GPT-4 to exist. The enormous requirements of data, and the existence of an OpenAI web crawler, outside of publicly available datasets, OpenAI (and likely many of their competitors) likely used web scraping to gather the information they needed to build their LLMs.

Web scraping is the process of creating automated scripts that visit websites, download the HTML file, and store it internally. HTML files are intended for browser rendering, not data analysis, so the downloaded information is largely gibberish. Web scraping systems also have a parsing aspect that fixes the HTML file so that only the valuable data remains. Many companies use already use these tools to extract information such as product pricing or descriptions. LLM companies parse and format content in such a way that it resembles regular text like a blog post. Once a website has been parsed, it’s ready to be fed into the LLM.

All of this is used to acquire the contents of blog posts, articles, and other textual content. It’s being done at a remarkable scale.

Problems with web scraping

However, web scraping runs into two issues. One, websites aren’t usually all that happy about a legion of bots sending thousands of requests per second. Second, there is the question of copyright. Most web scraping companies use proxies, intermediary servers, that make changing IP addresses easy, which circumvents blocks, intentional or not. Additionally, it allows companies to acquire localised data – extremely important to some business models such as travel fare aggregation.

Copyright is a burning question in both the data acquisition and AI model industry. While the current stance is that publicly available data, in most cases, is alright to scrape, there’s questions about basing an entire business model that, in some sense, uses the data to replicate the text through an AI model.

Conclusion

There are a few key technologies that have collided to create the current iteration of AI models. Most of the familiar ones are based on machine learning, particularly the transformer invention.

Transformers can take textual data and convert it into vectors, however, their key advantage is the ability to take larger pieces of text (such as sentences) and look at them in their entirety. Previous technologies usually were only capable of evaluating words themselves.

Machine learning, however, has the problem of being data-hungry and exponentially-so. Web scraping was utilized in many cases to acquire terabytes of information from publicly available sources.

All of that data, in OpenAI’s case, was cleaned up and fed into a GPT. They are then often fine-tuned through human intervention to get better results out of the same corpus of data.

Inventions like ChatGPT (or chatbots with LLMs in general) are simply wrappers that make interacting with GPTs a lot easier. In fact, the chatbot part of the model might just be the simplest part of it.

In an era which demands flexibility and fast-paced responses to cyber threats and sudden market shifts, risk management has never been in more need for tools to support its ever-evolving transformation. 

AI is the key player which can keep up and perform beyond expectations. 

This isn’t about flashy tech for tech’s sake; rather, it’s about harnessing tools that can make businesses more resilient and agile. Sounds complicated? It’s not.  Here’s how your company can manage risk with ease and let your business grow with AI. 

Why should I care?

Put simply, AI-driven automation involves using technology to perform tasks that were traditionally done by humans, but with added intelligence. 

Unlike basic automation that follows set instructions, AI systems learn from data, recognise patterns, and even make decisions. In risk management, this means AI can help identify potential risks, assess their impact, and even respond in real time—often faster and more accurately than human teams.

Think of it like this: In finance, AI can monitor market fluctuations and automatically adjust portfolios to reduce exposure to risk. In operations, it can predict supply chain disruptions and recommend alternative strategies to keep production on track. AI helps by doing the heavy lifting, leaving leaders with clearer insights and the ability to make more informed decisions.

The insurance industry is a stand-out example of how AI-powered risk management can be done. It is transforming the sector by streamlining underwriting and claims processing, making confusing paperwork a thing of the past and loyal customers a thing of the future.

The Potential

Risk is part of doing business. We all know that, but the nature of risk has evolved, calling into question just how much companies can tolerate. Thanks to the interconnectedness of our digital and global economies, we can make fewer compromises and implement effective coping strategies to mitigate potential disruption which can ripple within minutes. 

For example, if you are a large international organisation, AI-driven automation can prove to be a valuable assistant when dealing with regulatory changes. JP Morgan jumped at the chance to incorporate AI’s uses. It has integrated AI into its risk management processes for fraud detection and credit risk analysis. The bank uses machine learning algorithms to analyse vast amounts of transaction data, detecting unusual patterns and flagging potentially fraudulent activities in real time. This has helped them significantly reduce fraud losses and improve the efficiency of their internal audit processes.

Additionally, the pace at which data is generated has exploded, making it nearly impossible for traditional risk management processes to keep up. 

This is where AI’s ability to process vast amounts of data quickly and accurately comes in handy. It offers predictive power that helps leaders anticipate risks instead of reacting to them. AI doesn’t get overwhelmed by the volume of information or distracted by the noise of the day; it consistently analyses data to identify potential threats and opportunities.

The automation aspect ensures that once risks are identified, responses can be triggered automatically. This reduces the chance of human error, speeds up reaction times, and allows teams to focus on strategic tasks rather than manual monitoring and troubleshooting.

The limitations

While a powerful tool, it doesn’t make it invincible or infallible. 

To ensure proper implementation, leaders must take note of its limitations. This means rolling out training across company departments to educate and upskill staff. This can involve conducting workshops, recruiting AI experts to the team, and setting realistic expectations from day one about what AI can and can’t do.

By teaming up with AI, company leaders can create a sandbox environment where you interact with AI using your own data. This practical approach simplifies the transition more than a lecture in a seminar room and can be tried and tested without full commitment or investment.

How AI Automation Can Make an Impact

There are several critical areas where AI-driven automation is already making a significant impact in risk management:

Cybersecurity is a sector that has huge potential for growth. As cyber threats become more sophisticated, AI systems are helping companies defend themselves. These systems can identify patterns of malicious behaviour, recognise the latest attack methods, and automate responses to neutralise threats quickly. 

This reduces downtime and limits damage, allowing companies to stay one step ahead of hackers. AXA has developed AI-powered tools to manage and mitigate cyber risks for both its operations and its customers. By leveraging AI, AXA analyses vast amounts of network data to detect and predict cyber threats. This helps businesses proactively manage vulnerabilities and minimise cyberattacks. 

The regulatory landscape is constantly shifting, and keeping up with these changes can be overwhelming. AI can automate the process of monitoring new regulations, assess their impact on the business, and ensure compliance by flagging potential issues before they become problems. This is especially critical for industries like finance and healthcare, where non-compliance can result in heavy fines or legal trouble.

Supply Chain Management also benefits from its implementation. Walmart uses AI to monitor risks in its vast network of suppliers. The company has developed machine learning models that analyse data from its suppliers, including financial stability, production capabilities, and past performance. AI also evaluates external data sources such as economic indicators, political risks, and natural disasters to identify potential threats to supply chain continuity.

How Leaders Can Implement AI-Driven Automation in Risk Management

How to embrace its innovation:

Identify Key Risk Areas: Start by mapping out the areas of your business most susceptible to risk. Whether it’s cybersecurity, regulatory compliance, financial instability, or operational inefficiencies, knowing where the biggest vulnerabilities lie will help you focus your AI efforts.

Assess Current Capabilities: Look at your current risk management processes and assess where automation could provide the most value. Are your teams spending too much time monitoring data? Are there manual tasks that could be streamlined? AI can enhance these processes by improving speed and accuracy.

Choose the Right Tools: Not all AI solutions are created equal, and it’s essential to choose tools that fit your specific needs. Work with trusted vendors who understand your industry and can offer customised solutions. Look for AI systems that are transparent, explainable, and adaptable to evolving risks.

Monitor and Adapt: AI systems need regular updates and monitoring to remain effective. Make sure you have a plan in place to review performance, adjust algorithms, and update data sets. This will ensure your AI tools continue to provide relevant, actionable insights as risks evolve.

If you don’t have the right talent, or capacity, or you’re unsure where to start, choose a reliable partner to help accelerate your use case and really get the best out of it. 

AI-driven automation is reshaping the future of risk management by making it more proactive, predictive, and efficient. Company leaders who embrace these technologies will not only be better equipped to navigate today’s complex risk landscape but will also position their businesses for long-term success. 

According to Forbes Advisor, 56% of businesses are using AI to improve and perfect business operations. Don’t risk falling behind and discover the wonders of AI today.

In this day and age, it’s safe to say we’re drowning in data. Every second, staggering amounts of information are generated across the globe—from social media posts and news articles to market transactions and sensor readings. This deluge of data presents both a challenge and an opportunity for businesses and organisations. The question is: how can we effectively harness this wealth of information to drive better decision-making?

As the founder of Permutable AI, I’ve been at the forefront of developing solutions to this very problem. It all started with a simple observation: traditional data analysis methods were buckling under the sheer volume, velocity, and variety of modern data streams. The truth is, a new approach was needed—one that could not only process vast amounts of information but also extract meaningful insights in real-time.

Enter AI 

Artificial Intelligence, particularly ML and NLP, has emerged as the key to unlocking the potential of big data. At Permutable AI, we’ve witnessed firsthand how AI can transform data overload from a burden into a strategic asset.

Consider the financial sector, where we’ve focused much of our efforts. There was a time when traders and analysts would spend hours poring over news reports, economic indicators, and market data to make informed decisions. In stark contrast, our AI-powered tools can now process millions of data points in seconds, identifying patterns and correlations that would be impossible for human analysts to spot.

But this isn’t just because of speed. The real power of AI lies in its ability to understand context and nuance. And this isn’t just about systems that can count keywords; they can also comprehend the sentiment behind news articles, social media chatter, and financial reports. This nuanced understanding allows for a more holistic view of market dynamics, leading to more accurate predictions and better-informed strategies.

AI’s Impact across industries

Needless to say, this transformation isn’t just limited to the financial sector, because the reality is AI is transforming how data is gathered, processed and used  across various sectors. Think of the potential for AI algorithms in analysing patient data, research papers, and clinical trials to assist in diagnosis and treatment planning. 

During the COVID-19 pandemic, while we were all happily – or perhaps not so happily, cooped up indoors, we saw how AI could be used to predict outbreak hotspots and optimise resource allocation. Meanwhile, the retail sector is already benefiting from AI’s ability to analyse customer behaviour, purchase history, and market trends, providing personalised product recommendations that are far too tempting, as well as optimising inventory management.

The list goes on, but in every sector, and in every use case, there is the potential here to not replace human expertise, but augment it. The goal should be to empower decision-makers with timely, accurate, and actionable insights, because in my personal opinion, a safe pair of human hands is needed to truly get the best out of these kinds of deep insights. 

Overcoming challenges in AI implementation

Despite its potential, implementing AI for data analysis is not without challenges. In my experience, three key hurdles often arise. Firstly, data quality is crucial, as AI models are only as good as the data they’re trained on. Ensuring data accuracy, consistency, and relevance is paramount. Secondly, as AI models become more complex, explaining their decisions becomes more challenging. 

This means investing heavily in developing explainable AI techniques to maintain transparency and build trust – and the importance of this can not be understated. AI plays an increasingly significant role in decision-making, addressing issues of bias, privacy, and accountability will become ever more crucial. With that said, overcoming these challenges requires a multidisciplinary approach, combining expertise in data science, domain knowledge, and ethical considerations.

The Future of AI-Driven Data Analysis

Looking ahead, I see several exciting developments on the horizon. Federated learning is a technique that allows AI models to be trained across multiple decentralised datasets without compromising data privacy. 

It could unlock new possibilities for collaboration and insight generation. Then, as quantum computers become more accessible, they could dramatically accelerate certain types of data analysis and AI model training. Automated machine learning tools will almost certainly democratise AI, allowing smaller organisations to benefit from advanced data analysis techniques rather than it just being the playground of the big boys.

 Finally, Edge AI, which processes data closer to its source, will enable faster, more efficient analysis, particularly crucial for IoT applications.

Navigating the AI future 

One thing if for certain, the data deluge shows no signs of slowing down. But with AI, what once seemed like an insurmountable challenge is now an unprecedented opportunity. By harnessing the power of AI, organisations can turn data overload into a wellspring of strategic insights.

It’s important to remember that the future of business intelligence is not just about having more data; it’s about having the right tools to make that data meaningful. In this data-rich world, those who can effectively harness AI to cut through the noise and extract valuable insights will have a decisive advantage. The question is no longer whether to embrace AI-driven data analysis, but how quickly and effectively we can implement it to drive our organisations forward.

To be clear, the competition is fierce in this rapidly evolving field. But while challenges remain, the potential rewards are immense. The reality is that AI-driven data analysis is becoming increasingly important across all sectors. For now, we’re just scratching the surface of what’s possible. As so often happens with transformative technologies, we’re likely to see even more remarkable insights emerge as AI continues to evolve. But it’s important to remember that AI is a tool, not a magic solution. 

Embracing the AI-driven future

As it stands, nearly every industry is grappling with how to make the most of their data. As for the future, it’s hard to predict exactly where we’ll be in five or ten years. Today, we’re seeing AI make a big splash in fields from finance to healthcare. The concern for people often centres around job displacement. However, all this means is that we need to focus on upskilling and retraining to work alongside AI systems.

And that’s before we address the potential of AI in tackling global challenges like climate change or pandemics. It’s the same story on a smaller scale in businesses around the world. AI is helping to solve problems and create opportunities like never before.

Ultimately, we must remember that the goal of all this technology is to enhance human decision-making, not replace it. It’s no secret that the world is becoming more complex and interconnected. In large part, our ability to navigate this complexity will depend on how well we can harness the power of AI to make sense of the vast amounts of data at our fingertips.

At the end of the day, AI-driven data analysis is not just about technology—it’s about unlocking human potential. And that, to me, is the most exciting prospect of all.

Data analytics has truly gone mainstream. Organisations across the world, in nearly every industry, are embracing the practice. Despite this, however, the execution of data analytics remains varied – and not all data analytics approaches are made equal.

For most organisations, the most advanced data analytics team is  the centralised Business Intelligence (BI) team. This isn’t necessarily inferior to having a specialist data science team in place. However, the world’s most successful BI teams do embrace data science principles. Comparatively, this isn’t something that all ‘classic BI teams’ nail. 

With more and more mature organisations benefiting from best practice data analytics – competitors that haven’t adapted risk getting left in the dust. The charter and organisation of typical BI need to be set up correctly for data analytics to address increasingly complicated challenges and drive transformational change across the business in a holistic manner.

Where is classic BI lacking?

BI’s primary focus is descriptive analytics. This means summarising what has happened and providing visualisation of data through dashboards and reports to establish trends and patterns. Visualisation is foundational in data analytics. The problem lies in how this visualisation is being carried out by BI teams. It’s often the case that BI teams are following an IT project model. They churn out specific reports like a factory production line based on requirements set by another part of the business. Too often, the goal is to deliver outputs quickly in a visually appealing way. However, this approach has several key deficiencies.

Firstly, it’s reactive rather than proactive. It is rooted in delivering reports or visualisations that answer predefined questions framed by the business. This is opposed to exploring data to uncover new insights or solve open-ended problems. This limits the potential of analytics to drive new innovative solutions.

Secondly, when BI teams follow an IT project model, they typically report to central IT teams rather than business leads. They lack the authority to influence broader business strategy or transformation. Therefore, their work remains siloed and disconnected from the core strategic objectives of the organisation. For too many companies, BI has remained a tool for looking backwards, rather than a driver of forward-thinking, data-driven decision-making. The IT model of collecting requirements and building to specification is not the transformational process used by world-class data science teams. Instead, understanding the business and driving change is a central theme seen within the world’s leading analytic organisations. 

The case for centralisation

To unlock the full potential of data analytics, organisations must centralise their data functions. They need a simple chain of command that feeds directly into the C-Suite. Doing so aligns data science with the business’s strategic direction. Doing so successfully creates several advantages that set companies with world-class data analytics practices apart from their peers.

Solving multi-domain problems with analytics

A compelling argument for centralising data science is the cross-functional nature of many analytical challenges. For example, an organisation might be trying to understand why its product is experiencing quality issues. The solution might involve exploring climatic conditions causing product failure, identifying plant processes or considering customer demographic data. These are not isolated problems confined to a single department. The solution therefore spans multiple domains, from manufacturing to product development to customer service.

A centralised data science function is ideally positioned to tackle such complex problems. It can draw insights from various domains as an integrated team to create holistic solutions without different parts of the organisation working at odds with each other. In contrast, where data scientists report to individual departments (centralisation isn’t happening) there’s a big risk of duplicating efforts and developing siloed solutions that miss the bigger picture.

Creating career pathways and developing talent

It should be obvious to state – data scientists need career paths too. The most important asset of any data science domain is the people. But despite this, where teams are decentralised, data scientists tend to work in small, isolated teams within specific departments. This limits their exposure to a broader range of problems and stifling career advancement opportunities. 

For example, a data scientist in a three-person marketing analytics team has fewer opportunities and less interaction with the overall business than a member of a 50-person corporate data science team reporting to the C-suite.

Centralising the data science team within a single organisational structure enables a more robust career path and fosters a culture of continuous learning and professional development. 

Data scientists can collaborate across domains, learn from each other and build a diverse skill set that enhances their ability to tackle complex problems. Moreover, it’s easier to provide consistent training, mentorship and development opportunities where data science is centralised, ensuring that teams are fully equipped with the latest tools and techniques.

Linking analytics across the business

A centralised data science function acts as a valuable bridge across different parts of the business. Let’s take an example. Two departments approach the data science team with seemingly conflicting requests. 

The supply chain team wants to minimise shipment costs and asks for an analytic that will identify opportunities to find new suppliers near existing manufacturing facilities. 

The purchasing team, separately, approaches the data science team to reduce the cost of each part. To do this, they want to identify where they have multiple suppliers, and move to a model with a single global supplier that has much larger volumes and will reduce costs. These competing philosophies will each optimise a piece of the business, but in reality, what should happen is a single optimised approach for the business.

Instead of developing competing solutions, a centralised data science team can balance competing objectives and deliver an optimal solution that’s aligned with overall strategy. Cast in this role, data science is the strategic partner contributing to the delivery of the best outcomes for the organisation.

Leveraging analytics methods across domains

The best breakthroughs in analytics come not from new algorithms, but from applying existing methods to innovate use cases. 

A centralised data science team, with its broad view of the organisation’s challenges, is more likely to recognise these opportunities and adapt solutions from one domain to another. For example, an algorithm that proves successful in optimising marketing campaigns could be adapted to improve inventory management or streamline production processes.

Driving organisational change and analytics maturity

Finally, a centralised data science function is best positioned to drive the overall analytic maturity of the organisation. 

This function can standardise governance, as well as best practices. In doing so, it can drive the change management processes, ensuring that data-driven decision-making becomes ingrained in company culture. 

The way forward

The shift from classic BI to a centralised data science function is not just a structural change; it is a crucial strategy for companies looking to stay ahead in a competitive, data-driven landscape. By centralising data science and enforcing a charter for BI to solve key problems of the organisation rather than be dictated to, companies can solve complex, cross-functional problems more effectively, foster talent development, create inter-departmental synergies and drive a culture of continuous improvement and innovation. 

This evolution is what sets world-class companies apart from the rest. It might just be the transformation your company needs to unlock its full potential.

The widespread adoption of Generative AI has infiltrated nearly every business sector. While tools like transcription and content creation are readily accessible to all, AI’s transformative potential extends far deeper. Its influence on coding and software development raises profound questions about the future of mutliple industries.

Addressing how AI can be best adopted without hampering creativity or overstepping the line when it comes to copyright or licensing laws is one of the major challenges facing software developers today. For instance, the Intellectual Property Office (IPO), the Government body responsible for overseeing intellectual property rights in the UK, confirmed recently that it has been unable to facilitate an agreement for a voluntary code of practice which would govern the use of copyright works by AI developers. 

The perfect match of AI and OS

Today, most AIs are being trained on open source (OSS) projects. This is because they can be accessed without the restrictions associated with proprietary software. This is something of a perfect match. It provides AI with an ideal training environment. The models are given access to a huge amount of standard code bases running in infrastructures around the world. At the same time, OS software is exposed to the acceleration and improvement that running with AI can provide.

Developers, too, are massively benefiting from AI. For example, they can ask questions, get answers and, whether it’s right or wrong, use AI as a basis to create something to work with. This major productivity gain is helping to refine coding at a rapid rate. Developers are also using it to solve mundane tasks quickly, get inspiration or source alternative examples on something they thought was a perfect solution.

Total certainty and transparency

However, it’s not all upside. The integration of AI into OSS has complicated licensing. General Public Licenses (GPL) are a series of widely used free software licences (there are others too), or copyleft, that guarantee end users four freedoms; to run, study, share, and modify the software. Under these licences, any modification of software needs to be released within the same software licence. If a code is licensed under GPL, any modification to it also needs to be GPL licensed.

There lies  the issue. There must be total transparency with regard to how the software has been trained. Without it, it’s impossible to determine the appropriate licensing requirements or how to even licence it in the first place. This makes traceability paramount if copyright infringement and other legal complications are to be avoided. Additionally, there are ethical questions? For example, is a developer has taken a piece of code and modified it, is it still the same code?

So the pressing issue is this: What practical steps can developers take to safeguard themselves against the code they produce? Alspo what role can the rest of the software community – OSS platforms, regulators, enterprises and AI companies – play in helping them do that? 

Here is where foundations come to offer guidance

Integrity and confidence in traceability matters more when it comes to OSS because everything is out in the open. A mistake or oversight in proprietary software might still happen. But, because it happens in a closed system, the chances of exposure are practically zero. Developers working in OSS are operating in full view of a community of millions. They need certainty with regard to a source code’s origin – is it a human, or is it AI?

There are foundations in place. Apache Software Foundation has a directive that says developers shouldn’t take source code done by AI. They can be assisted by AI but the code they contribute is the responsibility of the developer. If it turns out that there is a problem then it’s the developers issue to resolve. We have a similar protocol at Aiven. Our guidelines state that our developers can make use only of the pre-approved constrained Generative AI tools, but in any case, developers are responsible for the outputs and need to be scrutinised and analysed, and not simply taken as they are. This way we can ensure we are complying with the highest standards.

Beyond this, there are ways organisations using OSS can also play a role, taking steps to safeguard their own risks in the process. This includes the establishment of an internal AI Tactical Discovery team – a team set-up specifically to focus on the challenges and opportunities created by AI. We wrote more about this in a recent blog but, in this case it would involve a project specifically designed to critique OSS code bases, using tools like Software Composition Analysis to analyse the AI-generated codebase, comparing it against known open source repositories and vulnerability databases.

Creating a root of trust in AI

While it is happening, creating new licensing and laws around the role of AI in software development will take time. Not least because consensus is required when it comes to the specifics of its role and the terminology used to describe it. This is made more challenging because the speed of AI development and how it is being applied in code bases moves at a much quicker pace than those trying to put parameters in place to control it. 

When it comes to assessing if AI has provided copied OSS code as part of its output, factors such as proper attribution, licence compatibility, and ensuring the availability of the corresponding open source code and modifications are absolutely necessary. It would also help if AI companies start adding traceability to their source code. This will create a root of trust that has the potential to unlock significant benefits in software development. 

With the UK general election taking place earlier this Summer – and the November US presidential election on the horizon – 2024 is shaping up to be a record breaking year for elections. Over 100 ballot votes are taking place this year across 64 countries. However, around the globe, the rising threat of misinformation and disinformation is putting both public confidence in, and the integrity of, these elections at risk. 

The 2020 US election and the 2019 UK election have vividly illustrated how misinformation can create a sharp divide public opinion and heighten social tensions. The elections in early 2024, including the Indian general election and the European Parliament election, demonstrate that misinformation remains a persistent issue. 

As countries around the world gear up for their upcoming elections, the risk of misinformation influencing outcomes is a key concern, emphasising the need for vigilance and proactive measures to safeguard the integrity of the electoral process.

Misinformation and disinformation in election history 

In order to properly protect the electoral process, it’s important to understand how intentional misinformation and disinformation have affected previous elections. 

UK general election (2019)

Misinformation and disinformation played pivotal roles in the 2019 UK general election, prompting action from fact checking organisations like Full Fact, which published 110+ fact checks to address the deluge of false claims during the campaign. The Conservative Party drew significant backlash for its tactics, which included a rebranding of its X account to ‘FactCheckUK’ during a live televised debate – an act that was widely condemned as both deceptive and deliberately misleading.

Brexit, already a contentious issue, was also the target of numerous misinformation and disinformation campaigns during the election. Unverified and often false claims about economic impacts, border control, the migrant crisis and trade agreements further complicated the Brexit discourse and contributed to a deeply divided electorate. The spread of misinformation biassed public perception and raised serious concerns about its lasting effects on democratic processes, with 77% of people stating that truthfulness in UK politics had declined since the 2017 general election, per Full Fact.

US presidential election (2020)

During the 2020 presidential elections, the US faced significant challenges in maintaining legitimacy and integrity due to widespread misinformation and disinformation campaigns. False claims regarding the origins and treatments of COVID-19, as well as the illegitimacy of mail-in ballots, impacted the election discourse heavily. Competing narratives arose, with some supporting mask-wearing and mail-in voting, while others arguing against masks and alleging voter fraud. Russia-affiliated actors were instrumental in spreading false information.

Reports indicated that the Wagner Group hired workers in Mexico to disseminate divisive messages and misinformation online ahead of the elections. Russia also targeted the US presidential elections using social media platforms such as Gettr, Parler and Truth Social to spread political messages, including voter fraud allegations. 

Aptly named ‘supersharers’ were pivotal in spreading misinformation and disinformation, with a sample of 2,107 supersharers found responsible for spreading 80% of content from fake news sites during the 2020 US presidential election, in a study by Science Magazine researchers.

2024 electoral disinformation campaigns

While many elections are still pending this year, it is important to acknowledge the influence of key electoral events that have already occurred, notably in India and the European Parliament. These concluded elections, tainted by substantial misinformation and disinformation campaigns, have significant repercussions on the political landscape. 

India general election

The widespread use of WhatsApp led to rampant misinformation and disinformation in India’s general elections in the second quarter of 2024. The Bharatiya Janata Party (BJP) managed an extensive network of WhatsApp groups to influence voters with campaign messaging and propaganda. 

Researchers from Rest of World estimate that the BJP controls at least 5 million WhatsApp groups across India, allowing rapid dissemination of information from Delhi to any location within 12 minutes. Specifically, the BJP used WhatsApp to amplify misinformation designed to inflame religious and ethnic tensions. Bad actors also disseminated incorrect information about election dates, polling locations and voter ID requirements to undermine participation by segments of the population. Independent hacktivists also targeted the elections, with Anonymous Bangladesh, Morocco Black Cyber Army and Anon Black Flag Indonesia among the groups seeking to exploit geopolitical narratives and tensions to influence the outcome.

European Parliamentary elections

The European Parliament elections were another key target of sophisticated misinformation and disinformation campaigns. Russia sought to sway public opinion and fuel discord among European Union (EU) countries. The Pravda Russian disinformation network, active since November 2023, targeted 19 EU countries, along with multiple non-EU nations and countries outside of Europe, including Norway, Moldova, Japan and Taiwan. 

Leveraging Russian state-owned or controlled media such as Lenta, Tass and Tsargrad, as well as Russian and pro-Russian Telegram accounts, Pravda websites disseminate pro-Russian content. 

Additionally, a related Russia-based disinformation network, named Portal Kombat – comprising 193 fake news websites targeting Ukraine, Poland, France and Germany among other countries – was uncovered by Vignium researchers. This campaign aimed to influence the European Parliament elections by spreading false information, including claims about French soldiers operating in Ukraine, pro-Ukraine German politicians being Nazis and Western elites supporting a global dictatorship intent on waging war with Russia. 

These efforts highlight the extensive and malicious strategies employed to manipulate public opinion and undermine democratic processes across multiple nations.

2024 emerging threats 

With a series of crucial elections set to unfold, past evidence suggests that misinformation and disinformation campaigns will again try to sway public opinion. Looking ahead, the 2024 US presidential elections are poised to face even more sophisticated disinformation tactics. The advent of deepfake technology and advanced AI-generated content poses new challenges for ensuring truthful political discourse.

United States presidential election

The 2024 US presidential election has already faced significant misinformation and disinformation, with thousands of accounts circulating various false claims about election fraud. 

Nearly one-third of US citizens believe the 2020 Presidential election was fraudulent, per research from Monmouth University – a narrative actively promoted by Donald Trump to support his candidacy. Unfounded allegations like these are dangerous as they legitimise conspiracy theories and false claims, establishing a foothold for these beliefs in mainstream politics.

AI tools are anticipated to intensify the spread of misinformation and disinformation in the upcoming elections, making it even more challenging to discern fact from fiction. In one instance, voters in New Hampshire were targeted by an audio deepfake impersonating Joe Biden during his campaign, urging them not to vote. 

Despite the ban on AI-generated robocalls by the Federal Communications Commission in February 2024, AI’s influence on misinformation remains formidable. Various accounts have circulated AI-generated images, such as those showing Joe Biden in a military uniform or Donald Trump being arrested, with minimal moderation by social media platforms. These developments underscore the growing challenge of combating AI-driven disinformation and its potential to mislead voters and distort democratic processes.

Geopolitical issues, and the misinformation and disinformation surrounding them, are also likely to affect upcoming elections significantly.

Mitigating misinformation and disinformation in elections

Misinformation and disinformation show no signs of abating anytime soon, but several countries, including Australia, Argentina and Canada are exploring new strategies to combat their effects. Argentina’s National Electoral Chamber (CNE) collaborated with Meta before the 2023 general elections to enhance transparency in political campaigns on their platforms. The CNE also partnered with WhatsApp to develop a chatbot that provided accurate election information, proactively countering misinformation by giving voters access to reliable information.

Ahead of the 2019 federal election, Canada put in place a Social Media Monitoring Unit, and in 2023, the Australian Electoral Commission ran its ‘Stop and Consider’ campaign to reduce election-related disinformation. Notably, the ‘Stop and Consider’ campaign used YouTube and other social media channels to address electoral information almost in real time.

Although recent election strategies in Australia, Canada and Argentina show potential in curbing the spread of misinformation and disinformation, it is clear from recent elections that  these issues continue to affect the electoral landscape. 

The rapid evolution of AI and the ongoing challenges faced by social media platforms in managing misinformation mean that current countermeasures often fall short. As a result, investing in media literacy education is an essential part of the equation. While it won’t stop the creation of false content, empowering the public with critical thinking skills is essential for challenging and resisting misinformation.

As regulatory control continues to play catch-up with technological innovation, the battle against misinformation in elections will continue, demanding ongoing watchfulness and an adaptive response. And at the end of the day, protecting electoral integrity relies on the public’s ability to critically analyse and question the information they encounter online.

Artificial intelligence (AI) is driving the next (much scarier) evolution of mass surveillance. The mass deployment of AI as a way to monitor average citizens and, supposedly, police body cam footage, is coming. And Oracle is going to power it, according to the cloud company’s cofounder and chairman, Larry Ellison, during an Oracle financial analyst meeting. 

AI — keeping all of us on our “best behaviour” 

While Elon Musk’s increasingly public courting of right wing extremists, misogynist grifters, prominent transphobes, and outright nazis is perhaps the loudest example of the ways in which big tech will full-throatedly throw in its lot with fascism rather than watch stock prices dip in any way, he has some stiff competition. 

Larry Ellison, in what was the most expansive and clearly unscripted section of Oracle’s hour-long public Q&A session last week, talked at some length about his vision for AI as a tool of mass surveillance. And, of course, he also suggested that, if one were to build an AI-powered surveillance state, Oracle (a company with a significant track record as a contractor for the US government) was the strategic partner best-suited to help realise that vision. 

Who watches the watchmen (when they shoot an unarmed black teenager)? 

Ellison’s first example how he’d deploy this technology, however, was police body cams. Designed to record officer interactions with members of the public, body cams supposedly increase accountability, transparency, and trust at a time when the public opinion of law enforcement has rarely been lower.  

Since body cams first started making their way into police forces in the US and UK, results have been mixed. On one hand, police in the UK objectively lie less when on camera. Researchers at Queen Mary University in London found that, not only were police reports from the recorded interactions significantly more accurate, but cameras reduced the negative interaction index significantly. 

However, another “shocking” report on policing in the UK by the BBC found that police were routinely switching off their body-worn cameras when using force, as well as deleting footage and sharing videos on WhatsApp. The BBC’s investigation from September 2023 found more than 150 reports of camera misuse by forces in England and Wales.

The situation isn’t much different in the US, where Eric Umansky and Umar Farooq of ProPublica noted in a (very good) article last December that, despite “hundreds of millions in taxpayer dollars” being spent on a supposed “revolution in transparency and accountability” has instead resulted in a situation where “police departments routinely refuse to release footage — even when officers kill.” And officers kill a lot in the US. Last year, American police used lethal force against 1,163 people, up 66 people from 2022, and continuing an upward trend from 2017. 

Policing the police with AI

Ellison’s argument that he wants to use AI to make police more accountable is, on the face of it, a potentially positive one.  

Lauding the potential of Oracle Cloud Infrastructure combined with advanced AI, Ellison painted a picture of a more “accountable” world.  He described AI as a constant overseer that would ensure “police will be on their best behaviour because we’re constantly watching and recording everything that’s going on.” 

His plan is for the police to use always-on body cams. These cameras will even keep recording when officers visit the restroom or eat a meal — although accessing sensitive footage requires a subpoena. Ellison’s plan is then to use AI trained to monitor officer feeds for anything untoward. This could, he theorised, prevent abuse of police power and save lives. “Every police officer is going to be supervised at all times,” he said. “If there’s a problem AI will report that problem to the appropriate person.” 

So far, so totally not something that police officers could get around with the same tactics (duct tape and tampering) police officers already use to disable body cams. 

However, police officers aren’t the only ones Ellison envisions under the watchful eye of artificial intelligence, observing us constantly like some sort of… Large sibling? Huge male relative? There has got to be a better phrase for that. Anyway—

Policing the rest of us with AI 

Ellison’s almost throwaway point at the end of the call is by far the most alarming part of his answer. “Citizens will be on their best behaviour because we’re constantly recording and reporting,” he said. “There are so many opportunities to exploit AI… The world is going to be a better place as we exploit these opportunities and take advantage of this great technology.” 

AI powered, cloud connected surveillance solutions are already big business, from hardware devices offering 24/7 protection to software-based business intelligence delivering new data-driven business insights. The hyper-invasive “supervision” that Ellison describes (drools over might be more accurate) is far from the pipe dream of one tech oligarch. It’s what they talk about openly, at dinner with each other (Ellison recently had a high profile dinner with Elon Musk, another government surveillance contract profiteer), in earnings calls; it’s what they’re going to sell to governments for billions of dollars to make their EBITDA go up at the expense of fundamental rights to privacy.

It’s already happening. In 2022, a class action lawsuit accused Oracle’s “worldwide surveillance machine” of amassing detailed dossiers on some five billion people. The suit accused the company and its adtech and advertising subsidiaries of violating the privacy of the majority of the people on Earth. 

The European Union’s (EU) new artificial intelligence act is the first piece of major AI regulation to affect the market. As part of its digital strategy, the EU has expressed a desire to AI as the technology develops. 

We spoke to Sasan Moaveni, Global Business Lead for AI & High-Performance Data Platforms at Hitachi Vantara, to learn more about the act and how it will affect AI in Europe, as well as the rest of the world. 

1. The EU has now finalised its AI Act. The legislation is officially in effect, four years after it was first proposed. As the first major AI law in the world, does this set a precedent for global AI regulation?

The Act marks a turning point in the provision of strong regulatory framework for AI, highlighting the growing awareness of the need for the safe and ethical development of AI technologies.

AI in general and ethical AI in particular are complex topics, so it is important that regulatory authorities such as the European Union (EU) clearly define the legal frameworks that organisations should adhere to. This helps them to avoid any potential grey areas in their development and use of AI.

Since the EU is a frontrunner in introducing a comprehensive set of AI regulations, it is likely to have a significant global impact and set a precedent for other countries, becoming an international benchmark. In any case, the Act will have an impact on all companies operating in, selling in, or offering services consumed in the EU.

2. The Act introduces a risk-based approach to AI regulation, categorising AI systems into minimal, specific transparency, high, and unacceptable risk levels. The Act’s high risk AI systems, which can include critical infrastructures, must implement requirements such as strong risk-mitigation strategies and high-quality data sets. Why is this so crucial, and how can organisations ensure they do this?

Broadly speaking, high risk AI systems are those that may pose a significant risk to the public’s health, safety, or fundamental rights. This explains why systems categorised as such must meet a much more stringent set of requirements.

The first step for organisations is to correctly identify if a given system falls within this category. The Act itself provides guidelines here, and it is also advisable to consider getting expert legal, ethical, and technical advice. If a system is identified as high risk, then one of the key considerations is around data quality and governance. To be clear – this consideration should apply to all AI systems, but in the case of high risk systems it is even more important given the potential consequences of something going wrong.

Crucially, organisations must ensure that data sets used to train high risk AI systems are accurate, complete, representative, and, most importantly, free from bias. In addition, ongoing policies need to maintain the data’s integrity – for example, policies around data protection and privacy. And as AI develops, so too do the challenges around data management, requiring increasingly intelligent risk mitigation and data protection strategies.

With an effective strategy in place, businesses can ensure that should a data-threatening event occur, not only are the Act’s requirements not breached, but operations can resume imminently with minimal downtime, cost, and interruption to critical services.

3. With AI developing at an exponential rate, many have expressed concerns that regulatory efforts will always be on the back foot and racing to catch up, with the EU AI Act itself going through extensive revisions before its launch. How can regulators tackle this challenge?

As the prevalence of AI continues to increase, considerations such as data privacy, which is regulated by GDPR in Europe, continue to gain importance.

The EU AI Act marks another key legal framework. Moving forward, we will see more and more legal restrictions like this come into play. For example, we may see developments in areas such as intellectual property ownership. Those areas that will need to be tackled will evolve and mature as the AI market continues to develop.

However, it is also important to realise that no regulatory framework can anticipate all the possible future developments in AI technology. It’s for this reason that striking a balance between legislation and innovation is so important and necessary.

4. The Act will significantly impact big tech firms like Microsoft, Google, Amazon, Apple, and Meta, who will face substantial fines for non-compliance. Does the Act also hinder innovation by creating red tape for start-up businesses and emerging industries?

We don’t know yet whether the Act will help or hinder innovation. However, it’s important to remember that it won’t cetegorise all AI systems as high risk. There are different system designations within the EU AI Act, and the most stringent regulations only apply to those systems designated as high risk.

We may see some teething pains as the industry begins to adapt and strike the right balance between innovation and regulation. Think back to when cloud computing hit the market. Enterprises planned to put all their workloads on the cloud before they recognised that public cloud was not suitable for all.

Over time, I think that we will reach a similar state of equilibrium with AI.

5. Overall, how can businesses ensure they remain compliant with the Act as they implement AI into their operations?

First and foremost, before implementing any AI projects, businesses need to ensure that they have a clear strategy, goals, and objectives around what it is they want to achieve.

Once that is in place, they should carefully select the right partner or partners who can not only ensure delivery of the business objectives, but also adherence to all relevant regulations, including the EU AI Act.

This approach will go a long way towards ensuring that they get the business benefits that they’re looking for, as well as remaining compliant with applicable regulations.

Two key problems for businesses hoping to reap the benefits of generative AI have remained the same over the last 12 months: hallucinations and trust. 

Business leaders need to build trustworthy applications in order to harvest the benefits of generative AI, which include gains in productivity and new ways to deliver customer service. To build trustworthy AI applications that don’t ‘hallucinate’ and offer inaccurate answers, it helps to look at internet search engines.

Internet search engines can offer important lessons in terms of what they currently do well, like sifting through vast amounts of data to find ‘good’ results, but also areas in which they struggle to deliver, such as letting less trustworthy sources appear ahead of reliable websites. Business leaders have complex requirements when it comes to the accuracy needed from generative AI. 

For instance, if an organisation is building an AI application which positions adverts on a web page, the occasional error isn’t too much of a problem. But if the AI is powering a chatbot which answers questions from a customer on the loan amount they are eligible to, for example, the chatbot must always get it right otherwise there could be damaging consequences. 

By learning from the successful aspects of search, business leaders can build new approaches for gen AI, empowering them to untangle trust issues, and reap the benefits of the technology in everything from customer service to content creation. 

Finding answers

One area where search engines perform well is sifting through large volumes of information and identifying the highest-quality sources. For example, by looking at the number and quality of links to a web page, search engines return the web pages that are most likely to be trustworthy. 

Search engines also favour domains that they know to be trustworthy, such as government websites, or established news sources. 

In business, generative AI apps can emulate these ranking techniques to return reliable results. 

They should favour the sources of company data that people access, search, and share most frequently. And they should strongly favour sources that are known to be trustworthy, such as corporate training manuals or a human resources database, while deprioritising less reliable sources. 

Building trust

Many foundational large language models (LLMs) have been trained on the wider Internet, which as we all know contains both reliable and unreliable information. 

This means that they’re able to address questions on a wide variety of topics, but they have yet to develop the more mature, sophisticated ranking methods that search engines use to refine their results. That’s one reason why many reputable LLMs can hallucinate and provide incorrect answers. 

One of the learnings here is that developers should think of LLMs as a language interlocutor, rather than a source of truth. In other words, LLMs are strong at understanding language and formulating responses, but they should not be used as a canonical source of knowledge. 

To address this problem, many businesses train their LLMs on their own corporate data and on vetted third-party data sets, minimising the presence of bad data. By adopting the ranking techniques of search engines and favouring high-quality data sources, AI-powered applications for businesses become far more reliable. 

A swift answer

Search has become quite accomplished at understanding context to resolve ambiguous queries. For example, a search term like “swift” can have multiple meanings – the author, the programming language, the banking system, the pop sensation, and so on. Search engines look at factors like geographic location and other terms in the search query to determine the user’s intent and provide the most relevant answer. 

However, when a search engine can’t provide the right answer, because it lacks sufficient context or a page with the answer doesn’t exist, it will try to do so anyway. For example, if you ask a search engine, “What will the economy be like 100 years from now?” there may be no reliable answer available. But search engines are based on a philosophy that they should provide an answer in almost all cases, even if they lack a high degree of confidence. 

This is unacceptable for many business use cases, and so generative AI applications need a layer between the search, or prompt, interface and the LLM that studies the possible contexts and determines if it can provide an accurate answer or not. 

If this layer finds that it cannot provide the answer with a high degree of confidence, it needs to disclose this to the user. This greatly reduces the likelihood of a wrong answer, helps to build trust with the user, and can provide them with an option to provide additional context so that the gen AI app can produce a confident result. 

Be open about your sources

Explainability is another weak area for search engines, but one that generative AI apps must employ to build greater trust. 

Just as secondary school teachers tell their students to show their work and cite sources, generative AI applications must do the same. By disclosing the sources of information, users can see where information came from and why they should trust it. 

Some of the public LLMs have started to provide this transparency and it should be a foundational element of generative AI-powered tools used in business. 

A more trustworthy approach

The benefits of generative AI are real and measurable, but so too are the challenges of creating AI applications which make few or no mistakes. The correct ethos is to approach AI tools with open eyes. 

All of us have learned from the internet to have a healthy scepticism when it comes to facts and sources. We should be levelling the same level of scepticism at AI and the companies pushing for its adoption. This involves always demanding transparency from AI applications where possible, seeking explainability at every stage of development, and remaining vigilant to the ever-present risk of bias creeping in. 

Building trustworthy AI applications this way could transform the world of business and the way we work. But reliability cannot be an afterthought if we want AI applications which can deliver on this promise. By taking the knowledge gleaned from search and adding new techniques, business leaders can find their way to generative AI apps which truly deliver on the potential of the technology. 

It goes without saying that businesses ignoring Artificial Intelligence (AI) are at risk of falling behind the curve. The game-changing tech has the potential to streamline operations, personalise customer experiences, and reveal critical business insights. The promise of AI and Machine Learning (ML) presents immense opportunities for business innovation. However, realising this potential requires an ethical and empathetic approach. 

Our research, is AI a craze or crucial: what are businesses really doing about AI? found that 99% of organisations are looking to use AI and ML to seize new opportunities. It also reported that 80% of organisations say they’ll commit 10% or more of their total AI budget to meeting regulatory requirements by the end of 2024. 

If this is the case, the questions businesses should be asking themselves are: How to implement AI ethically? What are the concerns they should be aware of? And is it a philosophical question to answer or a technological one? Or perhaps a social and organisational one?

Implementing ethical AI 

Businesses shoulder a significant responsibility in shaping the ethical development of AI. For AI to genuinely serve people’s interests, developing AI ethically must be a part of the process from the outset. It’s essential that those impacted by the transformative changes brought about by AI are involved from the very start. Ethics must central to the process from inception and ideation, to the design of AI-based solutions and products.  

Implementing AI ethically requires stringent data governance, making algorithms fair and unbiased. AI developers also need to ensure they build transparency into how AI systems make decisions that impact people’s lives. With that, addressing fairness and bias mitigation throughout the AI lifecycle is also vital. It involves identifying biases present in training data, algorithms, and outcomes, and then taking proactive measures to address them.  

One way in which organisations can ensure fairness and bias mitigation is by employing techniques such as fairness impact assessments. This assessment involves having a diverse team, consulting stakeholders, examining training data for biases, and ensuring the model and system are designed and function fairly to mitigate biases. 

Fostering transparency in AI systems 

Fostering transparency in AI systems isn’t just a nice-to-have; it’s imperative for ensuring ethical use and mitigating potential risks. This can be achieved through data transparency and governance. Users should feel like they’re in the driver’s seat, fully aware of what data is being collected, how it’s being collected, and what it’s being used for. It’s all about being upfront and honest.  

Developers must implement robust data governance frameworks to ensure the responsible handling of data including data minimisation, anonymisation and consent management practices. Transparent data governance isn’t just about ticking boxes; it’s about building trust, empowering users, and ensuring that AI systems operate with integrity. The more transparent this is, the more easily users will be able to understand how data is used. 

Aligning AI systems with human values 

Ensuring AI systems align with human values is a significant challenge. It’s a technological hurdle requiring significant work, but also a philosophical and ethical dilemma. We must put in the social, organisational and political work to define the human values for AI alignment, consider how differing interests influence that process, and account for the ecological context shaping human and AI interactions. 

Current AI systems learn by ingesting vast amounts of data from online sources. However, this data is often disconnected from real-world human experiences and factors. It may not represent nuances such as interpersonal interactions, cultural contexts, and practical life skills that humans rely on. As a result, the capabilities developed by these AI systems could be out of touch with authentic human needs and perspectives that the data fails to capture comprehensively. 

The values we are concerned with, such as respect for autonomy, fairness, transparency, explainability, and accountability, are embedded in this data. The best AI systems we have, and the ones that are successful, use humans and human judgements again as a source of data. These humans judgements guide these models in the right direction. 

Next steps 

The way that AI model developers architect and train their models can result in more than issues of data quality. They can also result in unintended biases. For example, users of chat systems may already be aware of the strange relationship of those systems to uncertainty. They don’t really know what they don’t know and therefore cannot act to fill in the gaps during conversation.

Businesses must audit algorithms, processes, and data to ensure fairness, or risk legal consequences and public backlash. Assumptions and biases embedded in these algorithms, process and data,  as well as their unpredicted emergent properties, potentially contribute to disparities and dehumanisation that conflict with a company’s ethical mission and values. Those who deploy AI solutions must constantly measure their performance against these values.

Without a doubt, businesses have a significant obligation to steer AI’s development ethically. Ongoing dialogues with stakeholders, coupled with a diligent governance approach centred on transparency, accountability, empathy and human welfare – including concern for people’s agency – will enable companies to deploy AI in a principled manner. This thoughtful leadership will allow businesses to unlock AI’s benefits while building public trust.

Alarm bells are ringing in the artificial intelligence (AI) sector. After almost two years of fervent excitement, controversy, and billions of dollars in capital expenditure, it seems as though investors may be turning against the all-consuming rise of generative AI. 

The market for artificial intelligence eclipsed $184 billion already this year, a considerable jump of nearly $50 billion compared with 2023. Now, however, as the panic spreads, it seems as though the AI bubble might be about to burst. 

NVIDIA’s stock price and the big AI wobble 

The stock market is currently having a bad time. All three US stock market indexes fell sharply on Monday after similar dips shook Europe and Asia. The dive has ostensibly been due to poor growth outlook in the US and a disappointing job market outlook, but, as Brian Merchant at Blood in the Machine points out, “a selloff of AI-invested tech companies is partly to blame.” 

Going back to the start of this month, you’ll find the biggest canary (a $3 trillion canary, to be specific) gasping for air at the bottom of the coal mine. US chipmaker Nvidia has ridden the AI demand wave to become the world’s most valuable company. However, it seems like the chip giant’s fortunes may be reversing as, once buoyed by the rising tide of AI excitement, the company lost around $900 billion in market value at the start of August.  

Sean Williams at the Motley Fool notes that “investors have, without fail, overestimated the adoption and utility of every perceived-to-be game-changing technology or trend for three decades.” Now, it seems as though reality has caught up with the “sensational bull market”, as the commercial value of AI is increasingly called into question. 

Too much speculation, not enough accumulation 

Despite publishing an article on the 1st of August predicting that AI investment will hit $200 billion globally by the start of next year (citing the fact that “innovations in electricity and personal computers unleashed investment booms of as much as 2% of US GDP), Goldman Sachs also (to less fanfare) released a report in June that calls into question whether investors should tolerate the worrying ratio between generative AI spending and the technology’s actual benefits. “Tech giants and beyond are set to spend over $1tn on AI capex in coming years, with so far little to show for it,” notes their report. 

Some of the experts Goldman Sachs spoke to criticised the timeline within which generative AI will deliver returns. “Given the focus and architecture of generative AI technology today… truly transformative changes won’t happen quickly and few—if any—will likely occur within the next 10 years,” said economist Daron Acemoglu. 

Others, including Global co-head of single stock research at Goldman Sachs itself, called into question generative AI’s fundamental capacity for solving problems big enough to justify the amount of money being spent to shove it all down our throats. “AI technology is exceptionally expensive, and to justify those costs, the technology must be able to solve complex problems, which it isn’t designed to do,” he said. 

As Merchant noted earlier this week, things are “starting to look bleak for the most-hyped Silicon Valley technology since the iPhone.” 

Cold feet on Wall Street

However, none of this really matters if tech giants can convince their investors that the upfront costs will be worth it. I mean, Uber has managed to convince venture capitalists to keep pouring money into a business model that’s basically “taxis but more exploitative” for over a decade with no sign that its model will ever be sustainable. And yet, the money keeps on coming. 

Surely, the wonders of AI can convince investors to keep investment chugging along in the vague hope that something good will come of it (or, more likely, a raging case of sunk cost fallacy)? 

The fact that the world’s biggest tech giants are struggling to do just that is probably the most damning evidence of just how cooked AI’s goose might be. 

According to an article in Bloomberg from the start of August, major tech firms, including Amazon., Microsoft, Meta, and Alphabet “had one job heading into this earnings season: show that the billions of dollars they’ve each sunk into the infrastructure propelling the artificial intelligence boom is translating into real sales. In the eyes of Wall Street, they disappointed.” 

Not in it for the long haul

Microsoft said that investors should expect AI monetization in “the next 15 years and beyond” — a tough pill to swallow given how much of a dent generative AI has been putting in Microsoft’s otherwise stellar sustainability efforts. Google CEO Sundar Pichai revealed that capital expenditure in Q2 grew from $6.9 billion to $13 billion year on year, then struggled to justify the expense to investors.  Meta CFO, Susan Li, warned that investors should expect “significant capex growth” this year. By the end of the year, the company expects to spend up to $40 billion on AI research and product development, according to Business Insider.

Essentially, AI is almost unfathomably expensive. The daily server costs for OpenAI are around $1 million. The technology consumes eye-watering amounts of electricity at a time when we need to be drawing down on our energy usage, not cranking it up to eleven. Training and developing new AI models also requires paying the most talented programmers in the world very large amounts of money. OpenAI could reportedly lose $5 billion this year alone. All for the promise that generative AI could, one day, be profitable. Personally, it doesn’t seem like sub-par email summaries and really weird porn are going to cut it. For once, the Wall Street guys and I seem to be in agreement.  

Shares in all major tech giants lurched downwards in the days following each one revealing the sheer scale of capital expenditure they had planned to support their continued generative AI efforts. However, it might not matter. As Merchant observes, “big tech has absolutely convinced itself that generative AI is the future, and thus far they’re apparently unwilling to listen to anyone else.” 

Artificial Intelligence (AI) and Machine Learning (ML) are more than just trending topics, they’ve been influencing our daily interactions for many years now. AI is already a fundamental part of our digital lives. These technologies are not about creating a futuristic world but enhancing our current one. When wielded correctly AI makes businesses more efficient, drives better decision making and creates more personalised customer experiences.

At the core of any AI system is data. This data trains AI, helping to make more informed decisions. However, as the saying goes, “garbage in, garbage out“, which is a good reminder of the implications of biassed data in general, and why it is important to recognise this from an AI and ML perspective.

Don’t get me wrong, using AI tools to process large amounts of data can uncover insights not immediately apparent, guiding decisions and identifying workflow inefficiencies or repetitive tasks, recommending automation where it is beneficial, resulting in better decisions and more streamlined operations.

But the consequences of data bias can have significant ramifications for any business that relies on data to inform decision making. These range from the ethical issues associated with perpetuating systemic inequalities to the cost and commercial risks of distorted business insights that could mislead decision-making.

Ethics

The most commonly discussed aspect of data bias pertains to its ethical and social implications. For instance, an AI hiring tool trained on historical data might perpetuate historical biases, favouring candidates from a specific gender, race, or socio-economic background.

Similarly, credit scoring algorithms that rely on biased datasets could unjustly favour or penalise certain demographic groups, leading to unfair practices and potential legal repercussions.

Impact on business decisions and profitability

From a business perspective, biassed data can lead to misguided strategies and financial losses. Consider a retail company that uses AI to analyse customer purchasing patterns.

If their dataset primarily includes transactions from urban, high-income areas, the AI model might inaccurately predict the preferences of customers in rural or lower-income regions. This misalignment can lead to poor inventory decisions, ineffective marketing strategies, and ultimately, lost sales and revenue.

Targeted advertising is another example. If the user interaction data used to train an AI model is skewed, the model might incorrectly conclude certain products are unpopular. This could then lead to reduced advertising efforts for those products. However, the lack of interaction could be due to the product being under-promoted initially, not a lack of interest. This cycle can cause potentially profitable products to be overlooked.

Accidental bias

Bias in datasets can often be accidental, stemming from seemingly innocuous decisions or oversights. For instance, a company developing a voice recognition system collects voice samples from its predominantly young, urban-based employees. While unintentional, this sampling method introduces a bias towards a specific age group and possibly a certain accent or speech pattern. When deployed, the system might struggle to accurately recognise voices from older demographics or different regions, limiting its effectiveness and market appeal.

Consider a business that collects customer feedback exclusively through its online platform. This method inadvertently biases the dataset towards a tech-savvy demographic, potentially one younger and more digitally inclined. Based on this feedback, the business might make decisions that cater predominantly to this group’s preferences.

This could prove to be acceptable if that is also the demographic that the business should be focusing on, but it could be the case that the demographics from which the data originated do not align with the overall demographic of the customer base. This skew in data can lead to misinformed product development, marketing strategies, and customer service improvements, ultimately impacting the business’s bottom line and restricting market reach.

Ultimately what matters is that organisations understand how their methods for collecting and using data can introduce bias, and that they know who their usage of that data will impact and act accordingly.

AI projects require robust and relevant data

Adequate time spent on data preparation ensures the efficiency and accuracy of AI models. By implementing robust measures to detect, mitigate, and prevent bias, businesses can enhance the reliability and fairness of their data-driven initiatives. In doing so, they not only fulfil their ethical responsibilities but they also unlock new opportunities for innovation, growth, and social impact in an increasingly data-driven world.

Today, many small and medium-sized enterprises (SMEs) are still dependent on paper-based and offline workflows, with data from Inside Government revealing that 55% of businesses across West Europe and North America are still completely reliant on paper. This means that without existing digital systems and a centralised database of historical data, the transition to AI-powered workflows can seem completely out of reach.

Balancing the integration of new technology while maintaining regular operations is the key to digital transformation. This has been a challenge for each transition period, but with the move to AI, the balance is even harder to find. Implementing AI solutions without consideration for existing systems and workflows can negatively impact employee experience, with employees needing to double check and correct inaccurate AI outcomes. That’s why companies must strategically plan for AI adoption, understanding where AI will be the most effective at improving workflows and how to unlock the greatest value for employees.

The data challenge: Preparation for the AI revolution

AI has the power to transform the way we work. Through the automation of routine tasks, such as searching and retrieving files or summarising large, complex documents, it can free up time for professionals to focus on creativity, and innovation. 

For SMEs to unlock the full potential of AI, they need AI systems fully tailored to their business, their operations, and their industry. They also need tools that become more specialised to their business with use. However, businesses achieve this level of personalisation by leveraging historical data. Doing this remains a key challenge for many smaller businesses. Research from the World Economic Forum (WEF) shows that 64% of SMEs find it challenging to effectively use the data from their systems and 74% struggle to maximise the value of their company’s data investments. This is where digital document management is key to making the most out of your company’s data.

Document management is the key to unlock the value of historical data

Proper documenting and labelling of historical data are critical. Doing so ensures AI tools have the right context when learning to automate workflows and provide insights optimised for the unique characteristics of the business. 

Without the right tools, translating paper-based records into a digital format that AI systems can read is slow and labour-intensive. This is especially true for SMEs that may lack the additional resources required to take on the mammoth task of digitising their entire operational history.

Cloud-based document management tools can help SMEs lay the groundwork for AI adoption through improved data capture and data management:

Data capture

Ensuring the quality of data captured is especially challenging with paper-based workflows. Paper documents require manual input from employees, which takes up valuable time as well as leaving the process open to the risk of human error and missing records, where data has not been recorded correctly or at all.

Employees need a system that simplifies the data input process and reduces the level of manual intervention required to accurately update records. Here, cloud-based document management tools can streamline the data capture process by automatically translating one form of data into another format. For example, the ability for document management tools to convert basic smartphone photos of documents into PDFs allows employees to record data in seconds and ensures data is captured and stored in one central database.

Taking automation one step further with the power of natural language processing, AI-powered transcription can now automatically generate transcripts from audio-visual content. This significantly streamlines the data capture process and even allows users to search audio and video files by phrases and quotes. 

Data management

Without a central source of truth, version control becomes a significant challenge for paper-based workflows. Gaps in records, as well as a lack of a standardised process and improper labelling significantly limit the value of historical data.

It’s essential to develop a streamlined and centralised database where all all digital content is stored. These datanbases boost the value of historical data, enabling users to easily search and retrieve that data across different document formats. 

For example, the ability to search within audio-visual documents, including object and optical character recognition inside images, means that as you search for images, you’ll not only search the image metadata that is included in each file, but also the contents of the images. Therefore, boosting the data accessible for analysis and business insights.

And with further developments in workflow-productivity AI tools, centralised cloud databases will be able to automatically sort and file documents based on the standard organisation practices set out by the business.

The benefits of a strategic approach to AI

Embracing AI technology shouldn’t just be about ticking a box and using the latest new tool. It’s about the impact it can have on the business and the value it brings for employees, not just in saved hours on a single task a week, but in the seconds saved in every action taken throughout the working day. 

In order to achieve these benefits, AI algorithms require quality data to optimise workflows to suit the unique characteristics of each business and their employees’ needs. Now is the time for businesses to start laying the groundwork for AI-powered digital transformation by setting up processes to effectively capture and manage their digital data.

Humanoid robots have been floating around for years. We’re all familiar with the experience of watching a new annual video from Boston Dynamics depicting increasingly Terminator-reminiscent robots doing assault courses and getting the snot kicked out of them like they’re on a $2,000 per day masculinity retreat.  However, until recently, even the excitement surrounding Boston Dynamics’ robot dog Spot seemed to have died down. The consensus, it seemed, was that the road to robots that walk, talk, and hopefully don’t enslave us all to work in their bitcoin mines (I still don’t know what Bitcoin is so I’m just going to assume it’s a scam that robots use for food) was going to be long and slow. 

Now, however, that might be changing. 

Around the world, the robotics arms race is picking up speed. This newly catalysed competition is centering on the potential for artificial intelligence (AI) to be the catalyst for the next phase in the evolution of robotics. 

This week, Pennsylvania-based tech startup Skild managed to secure $200 million in Series A funding led by Lightspeed Venture Partners, Coatue, SoftBank Group, and Jeff Bezos’ venture capital firm, among others. The intersection of AI and robotics is a sector of the tech industry that attracts big money. All in all, robotics startups secured over $4.2 billion in seed through growth-stage financing this year already. 

AI could give us a general purpose robot brain 

Skild, along with other startups like Figure (which completed a $675 million Series B round in February funded by Nvidia, Microsoft, and Amazon) and 1X (an American-Norwegian startup that secured a relatively modest $98 million in January), is focusing on using large AI models to make robots better at interacting with the physical world. 

“The large-scale model we are building demonstrates unparalleled generalisation and emergent capabilities across robots and tasks, providing significant potential for automation within real-world environments,” said Deepak Pathak, CEO and Co-Founder of Skild AI. 

What this means is that, rather than designing software to make each individual robot move, perform tasks, and interact with the world around it, Skild AI’s model will serve as a shared, general-purpose brain for a diverse embodiment of robots, scenarios and tasks, including manipulation, locomotion and navigation. 

From “resilient quadrupeds mastering adverse physical conditions, to vision-based humanoids performing dexterous manipulation of objects for complex household and industrial tasks,” Skild AI plans for its model to make the production of robotics cheaper, enabling the use of low-cost robots across a broad range of industries and applications.

Pathak added that he believes his company represents “a step change” in how robotics will scale in the future. He adds that, if their scalable general purpose robot brain works, it “has the potential to change the entire physical economy.”

Experts are inclined to agree, with Henrik Christensen, professor of computer science and engineering at University of California at San Diego, telling CNBC that “Robotics is where AI meets reality.”

Okay, now the robots are coming for your jobs

Despite a national unemployment rate that remains hovering around 4%, US companies and media outlets continue to parrot the talking point that there is a massive skills shortage in the country. The solution, according to companies that make AI-powered robots is, unsurprisingly, AI-powered robots. 

According to the US Chamber of Commerce, there are currently more than 1.7 million jobs available than there are unemployed workers, especially in the manufacturing sector, where Goldman estimates there’s a shortage of around half a million skilled workers. 

Skild claims that its model enables robots to adapt and perform novel tasks alongside humans, or in dangerous settings, instead of humans.

“With general purpose robots that can safely perform any automated task, in any environment, and with any type of embodiment, we can expand the capabilities of robots, democratise their cost, and support the severely understaffed labour market,” said Abhinav Gupta, President and Co-Founder of Skild AI.

However, Andersson told CNBC that “When it comes to mass adoption or even something closely resembling mass adoption, I think we’ll have to wait quite a few years. Probably a decade at least.” 

Nevertheless, companies across the world are fighting to leverage the power of large AI models to spur the next generation of robots. “A GPT-3 moment is coming to the world of robotics,” said Stephanie Zhan, Partner, Sequoia Capital, one of the companies that led Skild AI’s funding round. “It will spark a monumental shift that brings advancements similar to what we’ve seen in the world of digital intelligence, to the physical world.”

Harness today warns that the exponential growth of AI-generated code could triple developer toil within the next 12 months, and leave organisations exposed to a bigger “blast radius” from software flaws that escape to production. Nine-in-ten developers are already using AI-assisted coding tools to accelerate software delivery. As this continues, the volume of code shipped to the business is increasing by an order of magnitude. It is therefore becoming difficult for developers to keep up with the need to test, secure, and remediate issues in every line of code they deliver. If they don’t find a way to reduce developer toil in these stages of the software delivery lifecycle (SDLC) it will soon become impossible to prevent flaws and vulnerabilities from reaching production. As a result, organisations will face an increased risk of downtime and security breaches. 

“Generative AI has been a gamechanger for developers. Now, they can suddenly complete eight-week projects in four,” said Martin Reynolds, Field CTO at Harness. “However, as the volume of code developers ship to the business increases, so does the ‘blast radius’ if developers don’t rigorously test for flaws and vulnerabilities. AI might not introduce new security gaps to the delivery pipeline, but it does mean there’s more code being funnelled through existing ones. That creates a much higher chance of vulnerabilities or bugs being introduced unless developers spend significantly more time on testing and security. When developers discovered the Log4J vulnerability, they spent months finding affected components to remediate the threat. In the world of generative AI, they’d have to find the same needle in a much larger haystack.” 

Fighting fire with fire

Harness advises that the only way to contain the AI-generated code boom is to fight fire with fire. This means using AI to automatically analyse code changes, test for flaws and vulnerabilities, identify the risk impact, and ensure developers can roll back deployment issues in an instant. To reduce the risk of AI-generated code while minimising developer toil, organisations should:

• Integrate security into every phase of the SDLC – developers should build secure and governed pipelines to automate every single test, check, and verification required to drive efficiency and reduce risk. Applying a policy-as-code approach to the software delivery process will prevent new code making its way to production if it fails to meet strict requirements for availability, performance, and security.
• Conduct rigorous code attestation – The Solarwinds and MoveIT incidents highlighted the importance of extending secure delivery practices beyond an organisation’s own four walls. To minimise toil, IT leaders must ensure their teams can automate the processes needed to monitor and control open source software components and third-party artifacts, such as generating a Software Bill of Materials (SBOM) and conducting SLSA attestation.
• Use Generative AI to instantly remediate security issues – As well as enabling development teams to create code faster, generative AI can also help them to quickly triage and analyse vulnerabilities and secure their applications. These capabilities enable developers and security personnel to manage security issue backlogs and address critical risks promptly with significantly reduced toil.

Where to go from here

“The whole point of AI is to make things easier, but without the right quality assurance and security measures, developers could lose all the time they have saved,” argues Reynolds. “Enterprises must consider the developer experience in every measure or new technology they implement to accelerate innovation. By putting robust guardrails in place and using AI to enforce them, developers can more freely leverage automation to supercharge software delivery. At the same time, teams will spend less time on remediation and other workloads that increase toil. Ultimately, this reduces operational overheads while increasing security and compliance, creating a win-win scenario.”

BT Group and ServiceNow are expanding a long term strategic partnership into a multi-year agreement centred on generative artificial intelligence (AI). The move will, according to the group’s press release, “drive savings, efficiency, and improved customer experiences”. 

Following a successful digital transformation project to update BT’s legacy systems in 2022, ServiceNow will now extend its service management capabilities to the entire BT Group. The group will also adopt several of ServiceNow’s products, including Now Assist for Telecom Service Management (TSM) to power generative AI capabilities for internal and customer-facing teams.  

Now Assist generative AI supposedly helps agents write case summaries and review complex notes faster. According to BT, the initial roll out to 300 agents saw Now Assist demonstrate “meaningful results” by improving agent responsiveness and driving better experiences for employees and customers. Case summarization supposedly reduced the time it took agents to generate case activity summaries by 55%. This, BT says, created a better agent handoff experience by reducing the time it takes to review complex case notes, also by 55%. By reducing overall handling time, Now Assist is helping BT Group improve its mean time to resolve by a third. 

Hena Jalil, Managing Director and Business CIO at BT Group said that reimagining how BT delivers its service management “requires a platform first approach” and that the new AI-powered approach would “transform customer experience at BT Group, unlocking value at every stage of the journey.”

“In this new era of intelligent automation, ServiceNow puts AI to work for our customers – with speed, trust, and security,” said Paul Smith, Chief Commercial Officer at ServiceNow. “By leveraging the speed and scale of the Now Platform, we’re creating a competitive advantage for BT, driving enterprise-wide transformation, and helping them achieve new levels of productivity, innovation, and business impact.” 

Does “unlocking value” mean layoffs for BT? 

The company’s push towards generative AI faced criticism last year when the company announced plans to reduce its overall workforce by more than 40% by 2030. In May, BT revealed plans to cut 55,000 jobs. The majority of the expected layoffs will stem from the winding down of BT’s full fibre and 5G rollout in the UK. 

However, BT chief executive Philip Jansen said he expects 10,000 jobs to be automated away by artificial intelligence and that BT would “be a huge beneficiary of AI.”

In general, the threat that generative AI poses to existing jobs has been mounting since the technology’s explosion into the mainstream. Results of a survey published in April found that C-Suite executives expect generative AI to reduce the number of jobs at thousands of US companies. Almost half of the execs surveyed (41%) expected to employ fewer people because of the technology in the near future.

Despite the fact this figure has more to do with the opinion executives have of AI than whether or not the technology is actually ready to start replacing jobs (it’s not — except maybe executive roles). What it means is that the people who decide whether or not to hire more staff, maintain their headcount, or gut their departments and replace human beings with AI think AI is ready to take on the challenge.

Since ChatGPT captured the public consciousness in early 2023, generative artificial intelligence (AI) has attracted three things. Vast amounts of media attention, controversy and, of course, capital. 

The Generative AI investment frenzy 

Funding for generative AI companies quintupled year-over-year in 2023. The number of deals increased by 66%, that year. And, as of February 2024, 36 generative AI startups had achieved unicorn status with $1 billion-plus valuations. In March of 2023, chatbot builder Character.ai raised $150 million in a single funding round. They did this without a single dollar of reported revenue. They weren’t the only ones. A year later, the company is currently at the centre of a bidding war between Meta and Elon Musk’s xAI. Unsurprisingly, they also aren’t the only ones. Tech giants with near-infinitely deep pockets are fighting to capture top AI talent and technology.  

The frenzied investment and industry-wide rush to invest is understandable. Since the launch of Chat GPT (and the flurry of image generators, chat bots, and other generative AI tools that quickly followed) industry experts have been hammering home the same point again and again. They say that generative AI will change everything. 

Experts from McKinsey said in June 2023 that “Generative AI is poised to unleash the next wave of productivity.” They predicted the technology could add between $2.6 trillion to $4.4 trillion to the global economy every year. A Google blog post called generative AI “one of the rare technologies powerful enough to accelerate overall economic growth”. It went on to effusively compare its inevitable economic impact to that of the steam engine or electricity. 

According to just about every company pouring billions of dollars into AI projects, this technology is the future. AI adoption sounds like an irresistible rising tide. It sounds as though it’s already transforming the business landscape and dividing companies into leaders and laggards. If you believe the hype.

Increasingly, however, a disconnect is emerging between tech industry enthusiasm for generative AI and the technology’s real world usefulness. 

Building the generative AI future is harder than it sounds 

In October, people using Microsoft’s generative AI imager creator found that they could easily generate forbidden imagery. Hackers forced the model, powered by OpenAI’s DALL-E, to create a vast array of compromising images. These included from Mario and Goofy participating in the January 6th insurrection. They also management to generate Spongebob flying a plane into the World Trade Center in 9/11. Vice’s tech brand Motherboard was able to “generate images including Mickey Mouse holding an AR-15, Disney characters as Abu Ghraib guards, and Lego characters plotting a murder while holding weapons without issue.” 

Microsoft is far from the only company whose eye-wateringly expensive image generator has experienced serious issues. A study by researchers at Johns Hopkins in November found that “while [AI image generators are] supposed to make only G-rated pictures, they can be hacked to create content that’s not suitable for work,” including violent and pornographic imagery. “With the right code, the researchers said anyone, from casual users to people with malicious intent, could bypass the systems’ safety filters and use them to create inappropriate and potentially harmful content,” said researcher Roberto Molar Candanosa. 

Beyond image generation, virtually all generative AI applications, from Google’s malfunctioning replacement for search to dozens of examples of chatbots going rogue, have problems. 

Is generative AI a solution in search of a problem? 

As the technology struggles to bridge the gap between the billions upon billions of dollars spent to bring it to market and the reality that generative AI may not be the no-brainer-game-changer on which companies are already spending billions of dollars. In truth, it may be a very expensive, complicated, ethically flawed, and environmentally disastrous solution in desperate search of a problem.

“Much of the history of workplace technologies is thus: high-tech programs designed to squeeze workers, handed down by management to graft onto a problem created by an earlier one,” Brian Merchant, author of Blood in the Machine.  

“I have not lost a single wink of sleep over the notion that ChatGPT will become SkyNet, but I do worry that it, along with Copilot, Gemini, Cohere, and Anthropic, is being used by millions of managers around the world to cut the same sort of corners that the call centre companies have been cutting for decades. That the result will be lost and degraded jobs, worse customer service, hollowed out institutions, and all kinds of poor simulacra for what used to stand in its stead—all so a handful of Silicon Valley giants and its client companies might one day profit from the saved labour costs.” 

“AI chatbots and image generators are making headlines and fortunes, but a year and a half into their revolution, it remains tough to say exactly why we should all start using them,” observed Scott Rosenberg, managing editor of technology at Axios, in April. 

Nevertheless, the Generative AI genie is out of the bottle. The budgets have been spent. The partnerships have been announced. Now, both the companies building generative AI and the companies paying for it are desperately seeing a way to justify the expense. 

AI in search of an easy win  

It’s likely that AI will have applications that are worth the price of admission. One day. 

Its problems will be resolved in time. They have to be; the world’s biggest tech companies have spent too much money for it not to work. Nevertheless, using “AI” as a magic password to unlock unlimited portions of the budget feels like asking for trouble. 

As Mehul Nagrani, managing director for North America at InMoment, notes in a recent op-ed, “the technology of the moment is AI and anything remotely associated with it. Large language models (LLMs): They are AI. Machine learning (ML): That’s AI. That project you’re told there’s no funding for every year — call it AI and try again.” Nagrani warns that “Billions of dollars will be wasted on AI over the next decade,” and applying AI to any process without more than the general notion that it will magically create efficiencies and unlock new capabilities carries significant risk. 

As a result, many companies with significant dollar amounts earmarked for AI are reaching for “the absolute lowest hanging fruit for deploying generative AI: Helpdesks.”

The problem with AI chatbots and other “low hanging fruit” 

“Helpdesks are a pain for most companies because 90% of customer pain points can typically be answered by content that has already been generated and is available on the knowledge base, website, forums, or other knowledge sources (like Slack),” writes CustomGPT CEO Alden Do Rosario. “They are a pain for customers because customers don’t have the luxury of navigating your website and going through a needle in a haystack to find the answers they want.” He argues that, rather than navigate a maze-like website, customers would rather have the answer fed to them in “one shot”, like when they use ChatGPT.

Do Rosario’s suggestion is to use LLM models like ChatGPT to run automated helpdesks. These chatbots could rapidly synthesise information from within a company’s site, quickly producing clear answers to complex questions. The results, he believes, would be companies saving workers and customers time and energy. 

So far, however, chatbots have had a shaky start as replacements for human customer service reps.

In the UK, a disgruntled DPD customer—after a generative AI chatbot failed to answer his query—was able to make the courier company’s chatbot use the F-word and compose a poem about how bad DPD was. 

Parcel delivery firm DPD have replaced their customer service chat with an AI robot thing. It’s utterly useless at answering any queries, and when asked, it happily produced a poem about how terrible they are as a company. It also swore at me. 😂 pic.twitter.com/vjWlrIP3wn

— Ashley Beauchamp (@ashbeauchamp) January 18, 2024

In America, owners of a car dealership using an AI chatbot were horrified to discover it selling cars for $1. Chris Bakke, who perpetrated the exploit, received over 20 million views on his post. Afterwards, the car company announced that it would not be honouring the deal made by the chatbot. They cited the reason that the bot wasn’t an official representative of their business. 

I just bought a 2024 Chevy Tahoe for $1. pic.twitter.com/aq4wDitvQW

— Chris Bakke (@ChrisJBakke) December 17, 2023

Will investors turn against generative AI

Right now, evangelists for the rapid mass deployment of AI seem all too ready to hand over processes like customer relations, technical support, and other more impactful jobs like contract negotiation to AI. This is the same AI that people can convince, without much difficulty it seems, to sell items worth tens of thousands of dollars for roughly the cost of a chocolate bar. 

It appears, however, as though investors are starting to shift their stance. More and more Silicon Valley VS are expressing doubt about throwing infinite money into the generative AI pit. Investor Samir Kumar told TechCrunch in April that he believes the tide is turning on generative AI enthusiasm. 

“We’ll soon be evaluating whether generative AI delivers the promised efficiency gains at scale and drives top-line growth through AI-integrated products and services,” Kumar said. “If these anticipated milestones aren’t met and we remain primarily in an experimental phase, revenues from ‘experimental run rates’ might not transition into sustainable annual recurring revenue.”

Nevertheless, generative AI investment is still trending upwards. Funding for generative AI startups reached $25.2 billion in 2023. Generative AI accounted for over a quarter of all AI-related investments in 2023. However you slice it, it seems as though we’re going to talk to an awful lot more chatbots before the tide recedes. 

It’s neither new nor controversial to say that the world runs on data. Big data analytics are fundamental to maintaining agility and visibility. This is not to mention unlocking valuable insights that let orangisations stay competitive. Globally, the big data market is expected to grow to more than $401 billion by the end of 2028—up from $220 billion last year. 

Business leaders can pretty much universally agree that data is undeniably important. However, actually leveraging that data into impactful business outcomes remains a huge challenge for a lot of companies. Increasingly, focusing on the volume and variety of data alone leaves organisations without the one thing they really need: data they can trust. 

Data quality, not just quantity 

No matter how sophisticated the analytical tool, the quality of data that goes in determines the quality of insight that comes out. Good quality data is data that is suitable for its intended use. Poor quality data fails to meet this criterion. In other words, poor quality data cannot effectively support the outcomes it is being used to generate.

Raw data often falls into the category of poor quality data. For instance, data collected from social media platforms like Twitter is unstructured. In this raw form, it isn’t particularly useful for analysis or other valuable applications. Nonetheless, raw data can be transformed into good quality data through data cleaning and processing, which typically requires time.

Some bad data, however, is simply inaccurate, misleading, or fundamentally flawed. It can’t be easily refined into anything useful, and its presence in a data set can spoil any results. Data that lacks structure or has issues such as inaccuracy, incompleteness, inconsistencies, and duplication is considered poor quality data.

Is AI solving the problem or creating it? 

Concerns over data quality are as old at spreadsheets and maybe even the abacus. Managing, structuring, and creating insights from data only gets more complicated the more data you gather, and organisations today gather a frighteningly large amount of data as a matter of course.They might not be able to do anything with it, but everyone knows that data is valuable, so organisations take a more is more approach and hoover up as much as they can.  

New tools like generative artificial intelligence (AI) promise to help companies capture the value present in their data. The technology exploded onto the scene, promising rapid and sophisticated data analysis. Now, questionable inputs are being blamed for the hallucinations and other odd behaviours that very publicly undermined LLMs’ effectiveness. The current debacle with Google’s AI-assisted search being trained on reddit posts is a perfect example. 

However, AI has also been criticised for muddying the waters and further degrading the quality of data available. 

“How can we trust all our data in the generative AI economy?” asks Tuna Yemisci, regional director of Middle East, Africa and East Med at Qlik in a recent article. The trend isn’t going away either, with reports coming out earlier this year that observe data quality getting worse. A survey by dbt Labs found in April that poor data quality was the number one concern of the 456 analytics engineers, data engineers, data analysts, and other data professionals who took the survey.

The feedback loop 

Not only is AI undermining the quality of existing data, but bad existing data is undermining attempts to find applications for generative AI. The whole issue is in danger of creating a feedback loop that undermines the tech industry’s biggest bets for the future of digital economic activity. 

“There’s a common assumption that the data (companies) have accumulated over the years is AI-ready, but that’s not the case,” Joseph Ours, a Partner at Centric Consulting wrote in a recent blog post. “The reality is that no one has truly AI-ready data, at least not yet… Rushing into AI projects with incomplete data can be a recipe for disappointment. The power of AI lies in its ability to find patterns and insights humans might overlook. But if the necessary data is unavailable, even the most sophisticated AI cannot generate the insights organisations want most.”

The European Parliament has officially approved the Artificial Intelligence act, a regulation aiming to ensure safety and compliance in the use of AI, while also boosting innovation. Expected to come into force in June 2024, the act introduced a set of standards designed to guide organisations in the creation and implementation of AI technology. 

While AI has already been providing businesses with a wide array of new solutions and opportunities, it also poses several risks, particularly with the lack of regulations around it. For organisations to adopt this advanced technology in a safe and responsible way, it is essential for them to have a clear understanding of the regulatory measures being put in place.

The EU AI Act has split the applications of AI into four risk categories: unacceptable risk, high risk, limited risk, and minimal or no risk. Most of its provisions, however, won’t become applicable until after two years – giving companies until 2026 to comply. The exceptions to this are provisions related to prohibited AI systems, which will apply after six months, and those related to general purpose AI, which will apply after 12 months.

 Regulatory advances in AI safety: A look at the EU AI Act

The EU AI Act mandates that all AI systems seeking entry into the EU internal market must comply with its requirements. The act requires member states to establish governance bodies. These bodies will ensure AI systems follow the Act’s guidelines. This mirrors the establishment of AI Safety Institutes in the UK and the US, a significant outcome of the AI Safety Summit hosted by the UK government in November 2023. 

Admittedly, it’s difficult to fully evaluate the strengths and weaknesses of the act at this point. It has only recently been established, but the regulation provided will no doubt serve as stepping stones towards improving the current environment. Currently, AI systems exist with minimal regulations.

These practices will play a crucial role in researching, developing, and promoting the safe use of AI, and will help to address and mitigate the associated risks. That said the EU may have particularly stringent regulations, but the goal in this case is to avoid hindering the progress of AI development as compliance typically applies to the end-product and not the foundational models or creation of the technology itself (with some exceptions).

Article 53 of the EU AI Act is particularly attention-grabbing, introducing AI regulatory sandbox supervised spaces. These spaces have been designed to facilitate the development, testing, and validation of new AI systems before they are released into the market. Their main goal is to promote innovation, simplify market entry, resolve legal issues, improve understanding of AI’s advantages and disadvantages, ensure consistent compliance with regulations, and encourage the adoption of unified standards.

Navigating the implications of the EU’s AI Act: Balancing regulation and innovation

The implications of the EU’s AI acts are widespread, with the potential to affect various stakeholders, including businesses, researchers, and the public. This underlines the importance of striking a balance between regulation and innovation, to prevent these new rules from hindering technological development or compromising ethical standards.

Businesses, especially startups and mid-sized enterprises, may encounter additional challenges, as these regulations can increase their compliance costs and make it difficult to deploy AI quickly. However, it is important to recognise the increased confidence the act will bring to AI technology and its ability to boost ethical innovation that aligns with collective and shared values.

The EU AI Act is particularly significant for any business wanting to enter the EU AI market and involves some important implications in relation to perceived risks. It is comforting to know that act plans to ban AI-powered systems that pose ‘unacceptable risks’, such as those that manipulate human behaviour, exploit vulnerabilities, or implement social scoring. The EU has mandated that companies register AI systems in eight critical falling under the ‘high-risk’ category that impedes safety or fundamental rights. 

What about AI chatbots?

Generative AI systems such as ChatGPT and other models are of limited risk, but they should obey transparency requirements. There is a grey line which means that users can choose whether to use these technologies or not after their interactions with it.

The user’s full knowledge of the situation makes this regulation more open for businesses, as they can provide optimum service to their customers without being hindered by the complicated parts of the law. There are no additional legal obligations that apply to low-risk AI systems in the EU, except for the ones already in place. This gives freedom to businesses and customers to innovate faster in collaboration by developing a compliance strategy. 

Article 53 of the EU AI Act gives businesses, non-profits, and other organisations free access to sandboxes for a limited participation period of up to two years, which is extendable, subject to eligibility criteria. With the agreement on a specific plan and their collaboration with the authorities to outlines the roles, details, issues, methods, risks, and exit milestones of the AI systems, this helps make entry into the EU market straightforward. It provides equal opportunities for startups and mid-sized businesses to compete with well established businesses in AI systems, without worrying too much about costs and the complexities of compliance. 

Where do we go from here?

Regulating AI across different nations is a highly complex task, but we have a duty to develop a unified approach that promotes ethical AI practices worldwide. There is, however, a large divide between policy and technology. As technology becomes further ingrained within society, we need to bridge this divide by bringing policymakers and technologists together to address ethical and compliance issues. We need to create an ecosystem where technologists engage with public policy, to try and foster public-interest. 

AI regulations are still evolving and will require a balance between innovation and ethics, as well as global and local perspectives. The aim is to ensure that AI systems are trustworthy, safe, and beneficial for society, while also respecting human rights and values. To ensure they are working to the best effect for all parties, there are many challenges to overcome first, including the lack of common standards and definitions, and the need for coordination and cooperation among different stakeholders.

There is no one-size-fits-all solution for regulating AI, it necessitates a dynamic and adaptive process supported by continuous dialogue, learning, and improvement.

Like the hype cycles of just about every technology preceding it, there is a significant chance of a major drawback in the AI market. AI is not a new technology. Previously AI winters all have been foreshadowed by unprecedented AI hype cycles, followed by unmet expectations, followed by pull-backs on using AI.

We are in the very same situation today with GenAI, amplified by an unprecedented multiplier effect.

The GenAI hype cycle is collapsing

Swirled up by the boundless hype around GenAI, organisations are exploring AI usage, often without understanding algorithms’ core limitations, or by trying to apply plasters to not-ready-for-prime-time applications of AI. Today, less than 10% of organisations can operationalise AI to enable meaningful execution.

Adding further pressure, tech companies’ decision to release LLMs to the public was premature. Multiple high profile AI fails followed the launch of public-facing LLMs. The resulting backlash is fueling prescriptive AI regulation. These AI regulations specify strong responsibility and transparency requirements for AI applications, which GenAI is unable to meet. AI regulation will exert further pressure on companies to pull back.

It’s already started. Today about 60% of banking companies are prohibiting or significantly limiting GenAI usage. This is expected to get more restrictive until AI governance reaches an aceptable point from consumers and regulators’ perspectives.

If, or when, a market drawback or collapse does occur, it would affect all enterprises, but some more than others. In financial services, where AI use has matured over decades, analytic and AI technologies exist today that can withstand AI regulatory scrutiny. Forward-looking companies are ensuring that they have interpretable AI and traditional analytics on hand while they explore newer AI technologies with appropriate caution. Many financial services organisations have already pulled back from using GenAI in both internally and customer facing applications; the fact that ChatGPT, for example, doesn’t give the same answer twice is a big roadblock for banks, which operate on the principle of consistency.

The enterprises that will pull back the most on AI are the ones that have gone all-in on GenAI –especially those that have already rebranded themselves as GenAI companies, much like there were Big Data companies a few years ago.

What repurcussions should we expect?

Since less than 10% of organisations can operationalise all the AI that they have been exploring, we are likely to see a return to normal; companies that had a mature Responsible AI practice will come back to investing in continuing that Responsible AI journey. They will establish corporate standards for building safe, trustworthy Responsible AI models that focus on the tenets of robust AI, interpretable AI, ethical AI and auditable AI. Concurrently, these practices will demonstrate that AI companies are adhering to regulations – and that their customers can trust the technology.

Organisations new to AI, or those that didn’t have a mature Responsible AI practice, will come out of their euphoric state, and will need to quickly adopt traditional statistical analytic approaches and / or begin the journey of defining a Responsible AI journey. Again, AI regulation will be the catalyst. This will be a challenge for many companies, as they may have explored AI through software vs. data science. They will need to change the composition of their teams.

Further eroded customer confidence

Many consumers do not trust AI, given the continual AI flops in market as well as any negative experiences they may have had with the technology. These people don’t trust AI because they don’t see companies taking their safety seriously, a violation of customer trust. Customers will see a pull-back in AI as assuaging their inherent mistrust in companies’ use of artificial intelligence in customer facing applications.

Unfortunately, though, other companies will find that a pull-back negatively impacts their AI-for-good initiatives. Those on the path of practising Responsible AI or developing these Responsible AI programmes may find it harder to establish legitimate AI use cases that improve human welfare. 

With most organisations lacking a corporate-wide AI model development / deployment governance standard, or even defining the tenants of Responsible AI, they will run out of time to apply AI in ways that improve customer outcomes. Customers will lose faith in “AI for good” prematurely, before they have a chance to see improvements such as a reduction in bias, better outcomes for under-served populations, better healthcare and other benefits.

Drawback prevention begins with transparency

To prevent major pull-back in AI today, we must go beyond aspirational and boastful claims, to having honest discussions of the risks of this technology, and defining what mature and immature AI look like. 

Companies need to empower their data science leadership to define what constitutes high-risk AI. Companies must focus on developing a Responsible AI programme, or boost Responsible AI practices that have atrophied during the GenAI hype cycle.  

They should start with a review of how AI regulation is developing, and whether they have the tools to appropriately address and pressure-test their AI applications. If they’re unprepared, they need to understand the business impacts if regulatory restrictions remove AI from their toolkit.  

Continuing, companies should determine and classify what is traditional AI vs. Generative AI and pinpoint where they are using each. They will recognise that traditional AI can be constructed and constrained to meet regulation, use the right AI algorithms and tools to meet business objectives. 

Finally, companies will want to adopt a humble AI approach to back up their AI deployments, to tier down to safer tech when the model indicates its decisioning is not 100% trustworthy.

The vital role of the data scientist

Too many organisations are driving AI strategy through business owners or software engineers who often have limited to no knowledge of the specifics of AI algorithms’ mathematics and risks. Stringing together AI is easy. 

Building AI that is responsible and safe is a much harder exercise. Data scientists can help businesses find the right paths to adopt the right types of AI for different business applications, regulatory compliances, and optimal consumer outcomes.

If asked why organisations are hyped about Generative AI (GenAI), it’s sometimes easy to answer, “who wouldn’t be?” The attraction of a technology that can potentially answer any query, completely naturally, is clear to organisations that want to boost user experience. And this in turn is leading to an average $6.7 million investment in GenAI in 2023-24.

Yet while GenAI attracts the headlines, Machine Learning (ML) is quietly doing a huge amount of less glamorous, but equally important, work. Whether acting as the bedrock for GenAI or generating predictive insights that support informed, strategic decisions, ML is a vital part of the enterprise toolkit. With this in mind, it’s no wonder that organisations are still investing heavily in AI in general, to the tune of $21.1 million.

The closest thing to a time machine

At its core, machine learning is currently the nearest technology we have to a time machine. By learning from the past to predict the future, it can drive actionable insights that the business can act on with confidence. However, to realise these benefits, organisations need the right approach.

First, they need to be able to measure, monitor and understand any impact on performance, efficiency and competitiveness. To do this, they need to integrate ML into operations and decision-making processes. It also needs to be fed the right data. Data sets must be extensive, so the AI can recognize and learn from patterns, and make accurate predictions. And data needs to be real-time, so that the AI is learning from and acting on the most up-to-date information possible. After all, as most of us know, what we thought was true yesterday, or even five minutes ago, isn’t always true now. It’s this combination of large volumes of real-time data that will give ML the analytical horsepower it needs to forecast demand; predict market trends; give customers unique experiences; or ensure supply chains are as optimised as possible.

For ML to create these contextualised, hyper-personalised insights that inform strategic decisions, the organisation needs the right data strategy in place.

One data strategy to rule them all

A successful strategy is one that combines historical data – with its rich backdrop of information that highlights long-term trends, patterns and outcomes – with real-time data that gives the most up-to-the-minute information. Without this, AI producing inaccurate insights could send enterprises a wild goose chase. At best, they will lose many of the efficiency benefits of AI through having to constantly double-check its conclusions: an issue already affecting 23% of development teams that use GenAI.

What does this strategy look like? It needs to include complete control over where data is stored, who has access and how it is used to minimise the risk of inappropriate use. Also, it needs to enable accessing, sharing and using data with minimal latency so AI can operate in real time. It needs to prevent proprietary data from being shared outside the organisation. And as much as possible it should consolidate database architecture so there is no risk of AI applications accessing – and becoming confused by – multiple versions of data.

This consolidation is key not only to reduce AI hallucinations, but to ensure the underlying architecture is as simple – and so easy to manage and protect – as possible. One way of reducing this complexity and overhead is with a unified data platform that can manage colossal amounts of both structured and unstructured data, and process them at scale.

This isn’t only a matter of eliminating data silos and multiple data stores. The more streamlined the architecture, the more the organisation can concentrate on creating a holistic view of operations, customer behaviours and market opportunities. Much like human employees, the AI can then concentrate its energies on the data itself, becoming more agile and precise.

Forging ahead with machine learning in the GenAI age

A consolidated, unified approach isn’t only a case of improved performance. As the compute and infrastructure demands of AI grow, and commitments to Corporate Social Responsibility and environmental initiatives drive organisations towards greater efficiency, it will be essential to ensuring enterprises can meet their goals.

While GenAI is at the centre of much AI hype, organisations still need to recognise the importance and potential of predictive AI based on machine learning. At its heart, the principles are the same. 

Organisations need both in-depth historical information and real-time data to create a strategic asset that aids insightful decision making. Underpinning all of these is a data strategy and platform that helps enterprises adopt AI efficiently, effectively and safely.

Rahul Pradhan, is Vice President of Product and Strategy for database-as-a-service provider Couchbase.

Apple looks like it’s finally getting into the generative artificial intelligence (AI) space, even though some say that the company is late to the party. Nevertheless, lagging behind Microsoft, Google, OpenAI, and other major players in the generative AI space, experts expect the Cupertino-based to make its first major generative-AI-related announcement later today. 

AI on Apple’s agenda (at last) 

At Apple’s annual World Wide Developers Conference (starting on Monday, June 10th), insiders report that the company’s move into generative AI will dominate the agenda. Tim Cook, Apple’s CEO, will likely unveil Apple’s new operating system, iOS 18 later today. Industry experts predict that the software update will be a major element underpinning the company’s generative AI aspirations. 

In addition to software, Apple typically also unveils its next hardware generation at the conference.

The next generation of Apple products will likely be the first to have AI capabilities baked in. Apple is far from the first company to hit the market with devices designed with AI in mind, however. Google’s Pixel 8 smartphone launched late last year and Samsung’s Android-based S24, which hit the market in January, are both use Google’s Gemini AI.  

Tech giants are launching a growing wave of “AI” devices designed to do more AI computing locally rather than in the cloud (like Chat-GPT, for example), which supposedly reduces strain on digital infrastructure and speed sup performance. Reception to the first generation of AI PCs, smartphones, and other devices like the Rabbit R1 has been mixed, however. 

However, the technology is advancing rapidly, and Apple’s reputation for user-friendly, high quality consumer devices could mean it has the potential to capture a large slice of the AI device market. Apple currently controls just under a third of the global smartphone market, while iOS computers have a market share just above 10%. 

Late to the generative AI party?

Some more optimistic experts suggest that Apple’s reticence to release generative AI products before being confident in the quality of life improvements the technology can deliver is a good thing. “Apple’s early reticence toward AI was entirely on brand,” wrote Dipanjan Chatterjee vice president and principal analyst at Forrester. “The company has always been famously obsessed with what its offerings did for its customers rather than how it did it.”

However, Leo Gebbie, an analyst at CCS Insight, told the Financial Times that Apple’s leap into the AI pool may not be as calculated as some believe. “With AI, it does feel as though Apple has had its hand forced a little bit in terms of the timing,” she said. “For a long time Apple preferred not to even speak about ‘AI’ — it liked to speak instead about ‘machine learning.’”

She added: “That dynamic shifted maybe six months ago when Tim Cook started talking about ‘AI’ and reassuring investors. It was quite fascinating to see Apple, for once, dragged into a conversation that was not on its own terms.”

Whether or not Apple’s entrance to the generative AI race is entirely willing or not, there’s no doubt that the inclusion of the technology in Apple devices could mark another major inflection point for AI adoption among consumers. 

Industry experts believe that this week’s announcements will constitute a major milestone for the tech sector. Given the widespread use of Apple devices, the success or failure of generative AI embedded into the iPhone, iPad, Apple Watch, Mac computers and other devices will undeniably have some serious consequences for the technology.

It’s been a year and a half since Chat-GPT and other such AI tools were released to the public. Since then, generative artificial intelligence (AI) has attracted massive media attention, investment, and controversy. Now, new data from McKinsey suggests that generative AI tools are already seeing relatively widespread adoption in enterprise environments. 

Generative AI investment doubled last year

The value of private equity and venture capital-backed investments in generative AI companies more than doubled last year. Even bucking an otherwise sluggish investment landscape. According to S&P Global Market Intelligence data, generative AI investments by private equity firms reached $2.18 billion in 2023. This is compared to $1 billion the year before.

However, there’s a difference between investment and real-world applications that support a profitable business model. Just ask Uber, Netflix, WeWork, or any other “disruptive” tech company. 

In 2023, generative AI captivated the attention of everyone from the media to investors. Since then, the debate has raged over what exactly the technology will actually do. 

Is AI coming for our jobs? 

According to many prominent tech industry figures, from Elon Musk to the “godfather of AI” Geoffrey Hinton, AI is definitely coming for our jobs. Any day now. If Musk is to be believed, we can all expect to be out of work imminently. He claimed recently that “AI and the robots will provide any goods and services that you want”. Jobs would be, he concluded reduced to hobbies. 

However, studies like the one recently performed at MIT suggest that AI may not be ready to take our jobs just yet… or any time soon, for that matter. The last few weeks’ tech news has been dominated by Google’s AI search melting down, hallucinating, and giving factually inaccurate answers. A crop of AI apps designed to help identify mushrooms have been performing poorly, with potentially deadly results—part of what Tatum Hunter for the Washington Post describes as “emblematic of a larger trend toward adding AI into products that might not benefit from it.” 

According to Peter Cappelli, a management professor at the University of Pennsylvania Wharton School, generative AI is regularly being over-applied to situations where simple automation will suffice. According to Capelli, generative AI may be creating more work for people than it alleviates. LLMs are difficult to deploy. “It turns out there are many things generative AI could do that we don’t really need doing,” he added.

Generative AI is delivering return on investment

Nevertheless, generative AI adoption is accelerating at a meaningful pace among enterprises, according to McKinsey’s new data. Not only that, but “Organisations are already seeing material benefits from gen AI use, reporting both cost decreases and revenue jumps in the business units deploying the technology,” note authors Alex Singla, Alexander Sukharevsky, Lareina Yee, and Michael Chui, with Bryce Hall on behalf of Quantum Black, MicKinsey’s AI division. 

Most organisations using gen AI are deploying it in both marketing and sales and in product and service development. The biggest increase from 2023 took place in marketing and sales, where MicKinsey found that adoption had more than doubled. The function where the most respondents reported seeing cost decreases was human resources. However, respondents most commonly reported “meaningful” revenue increases in their supply chain and inventory management functions. 

So, are we headed for a radical employment apocalypse? 

“The technology’s potential is no longer in question,” said Singla. “And while most organisations are still in the early stages of their journeys with gen AI, we are beginning to get a picture of what works and what doesn’t in implementing—and generating actual value with—the technology.” 

According to Brian Merchant at Blood in the Machine, “regardless of how this is framed in the media or McKinsey reports or internal memos, ‘AI’ or ‘a robot’ is never, ever going to take your job. It can’t. It’s not sentient, or capable of making decisions. Generative AI is not going to kill your job — but your manager might.” 

He adds that, while “there will almost certainly be no AI jobs apocalypse,” this doesn’t necessarily mean that people won’t suffer as the technology continues to be more widely adopted. “Your boss isn’t going to use AI to replace jobs, or, more likely, going to use the spectre of AI to keep pay down and demand higher productivity,” Merchant adds.

The smartphone, like the gun or high speed rail, is approaching being a “solved technology.” Each year’s crop of flagship devices might run a little faster, bristle with even more powerful optics, and even fold in half like the world’s most expensive piece of origami. At the core of it, however, smartphones have been doing the things that are actually central to their design for over five years at this point. 

Smartphones are ubiquitous, connected, and affordable. Their form factor has defined the past decade. The question, however, is will it define the next decade? What about the next century? Or, as some suggest, is the age of the smartphone already drawing to a close? 

A post-smartphone world

Ever since the smartphone rose to prominence, people have been looking for the technology that will supplant it. From the ill-fated Google Glass to Apple’s new Vision Pro VR headset, the world’s smartest people have invested billions of dollars and hundreds of thousands of hours looking for something better than a rectangle of black glass. 

“In the long run, smartphones are unlikely to be the apotheosis of personal technology,” wrote technology strategist Don Philmlee last year for Reuters. When something does come along that breaks the smartphone’s hold on us, Philmlee expects it to be a “more personal and more intimate technology. Maybe something that folds, is worn, is embedded under our skin, or is ambiently available in our environment.” 

Right now, a new generation of AI-powered gadgets are giving us a glimpse into what that could look like. 

The AI gadget era? 

Tech giants and startups alike are racing to capitalise on the potential of generative AI to power a new wave of devices and gadgets. 

Right now, the first wave of devices, including Humane’s AI Pin, Rabbit’s R1, and Brilliant Labs’ AI-powered smart glasses are among the first wave of these devices to hit the market. 

Most of these devices substitute the traditional smartphone form factor for something smaller and voice controlled. They have a microphone and a camera for inputting commands. The devices then either dispense information via speaker or limited visual displays. Humane’s AI-Pin even contains a projector that can shine text or simple images onto a nearby surface or the user’s hand. 

The specifics differ, but all these gadgets put artificial intelligence at the forefront of the user experience. A series of large language models then pars the queries. The results are generated by image analysers, large language models, and other cutting edge AI. “AI is not an app or a feature; it’s the whole thing,” writes the Verge’s tech editor, David Pierce. 

However, creating novel hardware is difficult. Creating novel hardware that outperforms the smartphone? Things don’t necessarily look good for the first crop of AI tech. 

A shaky start for the first crop of AI gadgets

Despite Pierce’s bold proclamation that “we’ll look back on April 2024 as the beginning of a new technological era,” even he is forced to admit that, when it comes to Humane’s AI Pin, “After many days of testing, the one and only thing I can truly rely on the AI Pin to do is tell me the time”. 

Other reviewers have been similarly critical of this first generation of AI gadgets. When reviewing the AI Pin, Marques Brownlee wrote, “this thing is bad at almost everything it does, basically all the time.”

However, devices like the Rabbit R1 have shown promise and generated excitement. By combining a Large Language Model with a “Large Action Model”, the device can not only understand requests, but execute on them. For example, in addition to providing suggestions for a healthy dinner, Rabbit can reportedly place an order with a local restaurant, or purchase ingredients for delivery. 

“The Large Action Model works almost similarly to an LLM, but rather than learning from a database of words, it is learning from actions humans can take on websites and apps — such as ordering food, booking an Uber or even super complex processes,” wrote one reviewer. He explains that the Rabbit R1 isn’t trying to replace the smartphone. However, he notes that he “wouldn’t be surprised if it becomes a handset substitute. This is a breakthrough product that I never knew I needed until I held one in my hands.” 

Generative artificial intelligence (AI) companies like OpenAI, Google, and Microsoft are on the hunt for new training data. In 2022 a research paper warned that we could run out of high quality data on which to train stable diffusion algorithms and large language models (LLMs) as soon as 2026. Since then, AI firms have reportedly found a potential source of new information: social media. 

Social media offers “vast” amounts of usable training data

In February, it was revealed that the social media site reddit had struck a deal with a large AI company. The $60 million per year agreement will see the company train its generative AI using content created by reddit’s users. The buyer was later revealed to be Google, which is locked in a bitter AI race with OpenAI and Microsoft.

This will allegedly provide Google with an “efficient and structured way to access the vast corpus of existing content on Reddit.” 

The move caused significant controversy in the ramp up to an expected public offering by the company. A week later, social media platform tumblr and blog hosting platform WordPress also announced that they would be selling their users’ data to Midjourney and OpenAI. 

The race for AI training data  

These developments mark an evolution of an existing trend. Increasingly the AI industry is shifting from unpaid data scraping towards a model where the owners of data are paid for it. Recently, OpenAI was revealed to be paying between $1 million and $5 million a year to licence copyrighted news articles from outlets like the New York Times and the Washington Post to train its AI models.  

In December 2023, OpenAI also signed an agreement with Axel Springer. The German publisher is being paid an undisclosed sum for access to articles published Politico and Business Insider. OpenAI has also struck deals with other organisations, including the Associated Press, and is reportedly in licensing talks with CNN, Fox, and Time. 

However, a content creation (or journalistic) organisation licensing out the content it creates and distributes is one thing. The sale of public and private user data generated on social media is an entirely different matter. Of course, such data is already sold and mined heavily for advertising purposes. Income from the sale of personal data makes up the majority of social media sites like Facebook’s revenue.

If social media content is mined to train the next generation of AI, it’s essential that user data is anonymised. This may be less of an issue on sites like Reddit and Tumblr, where user identities are already concealed. However, the race for AI training data continues to gather pace. Soon, AI companies may look towards less anonymised sites like Instagram and X (formerly Twitter).

There’s no denying the speed at which the generative artificial intelligence (AI) sector has grown over the past year. 

In the UK, AI experimentation has been widespread. Research by Ofcom found that 31% of adults and 79% of 13–17-year-olds in the UK had used a generative AI tool, such as ChatGPT, Snapchat My AI, or Bing Chat (now called Copilot). This included for personal, educational, or professional reasons. Recent ONS data shows that around 15% of UK businesses are currently using at least one form of AI. Larger companies were also the most likely to adopt an AI tool.  

Since the launch of Chat-GPT at the tail end of 2022, the potential economic, political, and societal implications of AI have cast a long shadow. 

AI has attracted enthusiastic investment from businesses looking to be the first to adopt. The technology has also attracted criticism for a mixture of reasons. These range from the unethical use of intellectual property to train large AI models like Chat-GPT, to the potential devastation of the job market. 

Now, the UK’s Competition and Markets Authority (CMA) has highlighted the fact it has serious reservations over the “whirlwind pace” at which AI is being developed. 

“When we started this work, we were curious. Now, we have real concerns,” said Sarah Cardell, CEO of the CMA, speaking to the 72nd Antitrust Law Spring Meeting in Washington DC.

AI foundation models pose risk to “fair, effective, and open competition”

Cardell’s speech—along with an update to the CMA’s earlier report on AI foundational models released last year— highlighted the growing presence of a few incumbent tech companies further cementing their control over the sector, and the foundational AI market specifically.

“Without fair, open, and effective competition and strong consumer protection, underpinned by these principles, we see a real risk that the full potential of organisations or individuals to use AI to innovate and disrupt will not be realised, nor its benefits shared widely across society,” warned Cardell. She added that the foundational model sector of the AI market was developing at a “whirlwind pace.” 

“As exciting as this is, our update report will also reflect a marked increase in our concerns,” she explained. Specifically, Cardell and the CMA believe the growing presence across the foundation models value chain of a small number of incumbent technology firms, which already hold positions of market power in many of today’s most important digital markets. These firms,she argued, “could profoundly shape these new markets to the detriment of fair, open and effective competition, ultimately harming businesses and consumers, for example by reducing choice and quality and increasing price.” 

It’s difficult to say exactly how much data exists on the internet at any one time. Billions of gigabits are created and destroyed every day. However, if we were to try and capture the scope of the data that exists online, estimates suggest that the figure was about 175 zettabytes in 2022. 

A zettabyte is equal to 1,000 exabytes, or 1 trillion gigabytes, by the way. That’s (roughly) 3.5 trillion blu ray copies of Blade Runner: The Director’s Cut. If you converted all the data on the internet into blu-ray copies of Blade Runner: The Director’s Cut, and smashed every disk after watching it, you could spend about 510 times longer than the universe has existed watching Blade Runner before you ran out of copies. 

Was that a weird, tortured metaphor? Yes. Was it any more weird and unnecessary than Jared Leto’s presence in Blade Runner: 2049? Absolutely not. But I digress. The sheer amount of data that’s out there in the world is mind-boggling. It’s hard to fit into metaphors and defies real-world examples. 

Also, it seems we’re going to run out of it, and it might happen as early as 2030. 

We’re running out of (good) data?

The value of data has skyrocketed over the past few years. A global preoccupation with extracting, measuring, analysing, and—above all—monetising data defined the past decade. Big data has profoundly impacted our politics, entertainment, social spheres, and economies. 

Awareness of the things that can be accomplished with data—from optimising e-commerce revenues to cybercrime and putting people like Donald Trump in positions of political power—has led to a frenzied scramble for the stuff. Data is the world’s most valuable resourse. Like many other valuable resources, the rate at which we’re consuming it is turning out to be unsustainable. Organisations have tried frantically to gather as much data as possible. Any and all information about environmental conditions, personal spending habits, racial demographics, political bias, financial markets, and more has been gathered up into huge pools of Big Data.  

AI training models are to blame

However, there’s a problem related to the hot new use for huge data sets: training AI models.

“The gigantic volume of data that people stored but couldn’t use has found applications,” writed Atanu Biswas, a Professor at the Indian Statistical Institute in Kolkata. “The development and effectiveness of AI systems — their ability to learn, adapt and make informed decisions — are fuelled by data.” 

Training a large language model like the one that fuels OpenAI’s ChatGPT takes a lot of data. It took approximately 570 gigabytes of text data–about 300 billion words—to train ChatGPT. AI image generators are even hungrier, with stable diffusion engines like those powering DALL-E and Midjourney requiring over 5.8 billion image-text pairs to generate weird, unpleasant pictures where the hands are all wrong that Haiyo Miyazaki described as “an insult to life itself.”

This is because these generative AI models “learn” by intaking an almost unfathomable amount of data then using statistical probability to create results based on the observable patterns in that data. 

Basically, what you put in defines what you get out.

Bad data poisons AI models

Increasingly, the huge reserves of data used to train these generative AI models are starting to look thin on the ground. Sure, there’s a brain-breakingly large amount of data out there, but putting low quality—even dangerous—data into a model can produce low quality—even dangerous—results. 

Information sourced from social media platforms may exhibit bias, prejudice, or potentially disseminate disinformation or illicit material, all of which may be unwittingly adopted by the model. 

For example, Microsoft trained an AI bot using Twitter data in 2016. Almost immediately, the endeavour resulted in outputs tainted with racism and misogyny. Another problem is that, as the amount of AI-generated content on the internet increases, new models could end up being trained by cannibalising the content created by old models. Since AI can’t create anything “new”, only rephrase existing content, development would stagnate. 

As a result, developers are locked in an increasingly desperate hunt for “better” content sources. These include books, online articles, scientific papers, Wikipedia, and specific curated web material. For instance, Google’s AI Assistant was trained using around 11,000 romance novels. The nature of the data supposedly made it a better conversationalist (and, one presumes, a hornier one?). The problem is that this kind of data—books, research papers, and so on—is a limited resource. 

The paper Will we run out of data? suggests that the point of data exhaustion could be alarmingly close. Comparing the projected “growth of training datasets for vision and language models” to the growth of available data, they concluded that “we will likely run out of language data between 2030 and 2050.” Additionally, they estimate that “we will likely run out of vision data between 2030 to 2070.” 

Where will we get our AI training data in the future? 

There are several ways this problem could resolve itself. Popular solutions include smaller language models and even synthetic data created specifically to train AIs. There has even been a proposed freeze on all new AI research and development, signed by Elon Musk and Steve Wozniak, amojng others. 

“This is an existential risk,” commented Geoffrey Hinton, one of AI’s most prominent figures, shortly after quitting Alphabet last year. “It’s close enough that we ought to be … putting a lot of resources into figuring out what we can do about it.”

One hellish vision for the future appeared during the 2023 actors’ strike. During the strike, the MIT Technology Review reported that tech firms extended an opportunity to unemployed actors. They could earn $150 per hour by portraying a range of emotions on camera. The captured footage was them used to aid in the ‘training’ of AI systems.

At least we won’t all lose our jobs. Some of us will be paid to write new erotic fiction to power the next generation of Siri. 

Generative artificial intelligence (AI) has arrived. However, if 2022 was the year that generative AI exploded into the public consciousness, 2023 was the year the money started rolling in. Now, 2024 is the year when investors start to scrutinise their returns. PitchBook estimates that generative AI startups raised about $27 billion from investors last year. OpenAI alone was projected to rake in as much as $1 billion in revenue in 2024, according to Reuters.

This year, then, is the year that AI takes all-important steps towards maturity. If generative AI is to deliver on its promises, it needs to develop new capabilities and find real-world applications.

Currently, it looks like multimodal AI is going to be the next true step-change in what the technology can deliver. If investor are right, multimodal AI will deliver the kind of universal input to universal output functionality that would make Generative AI commercially viable.

What is multimodal AI? 

A multimodal AI model is a form of machine learning that can process information from different “modalities”. This includes images, videos, and text. They can then, theoretically, spit out results in a variety of formats as well. 

For example, an AI with a multimodal machine meaning model at its core could be fed a picture of a cake and generate a written recipe as a response and vice versa.

Why is multimodal AI a big deal? 

Multimodal models represent the next big step forward in how developers enhance AI for future applications. 

For instance, according to Google, its Gemini AI can understand and generate high-quality code in popular languages like Python, Java, C++, and Go, freeing up developers to create more feature-rich apps. This code could be generated in response to anything from simple images to a voice note. 

According to Google, this brings us closer to AI that acts less like software and more like an expert assistant.

“Multimodality has the power to create more human-like experiences that can better take advantage of the range of senses we use as humans, such as sight, speech and hearing,” says Jennifer Marsman, principal engineer for Microsoft’s Office of the Chief Technology Officer, Kevin Scott.

The emergence of sophisticated generative artificial intelligence (AI) applications—including image generators like Midjourney and conversational chatbots like OpenAI’s Chat-GPT—has sent shockwaves through the economy and popular culture in equal measure. The technology,  made accessible to a massive audience in a short span of time, has attracted immense interest, investment, and controversy. However, the data used to train large language models

Aside from criticisms rooted in the role played by generative AI in creating sexually explicit deepfakes of Taylor Swift, spreading misinformation, and enforcing prejudicial biases, the most prominent controversy surrounding the technology stems from the legal and ethical issues relating to the data used to train large language models (LLMs).

Generative AI large language models on unstable ethical ground

According to Chat-GPT 3.5 itself, LLMs are “trained on a vast dataset of text from various sources, including books, articles, websites, and other publicly available written material. This data helps us learn patterns and structures of language to generate responses and assist users.” 

Essentially, an LLM scrapes billions of lines of text from across the internet in order to train its learning model. Because generative AI consumes so much information, it can convincingly mimic, response, and “create” responses based on the data it has examined. However, authors, journalists, and several news organisations have raised concerns. The issue they highlight is that an LLM scraping content written by human authors is, in effect, uncredited and unpaid use of those writers’ work. 

Chat-GPT generates the response that “while large language models learn from existing text, they do so within legal and ethical boundaries, aiming to respect intellectual property rights and promote responsible usage.” 

A statement by to the European Writers’ Council contradicts the claim. “Already, numerous criminal and damaging “AI business models” have developed in the book sector – with fake authors, fake books and also fake readers,” the council says in a letter. “The fundamental process of developing large language models such as GPT, Meta, StableLM, and BERT rest on using uncredited copyrighted work. These works, asserts the Council, are sourced from “shadow libraries such as Library Genesis (LibGen), Z-Library (Bok), Sci-Hub and Bibliotik – piracy websites.”  

More ethical generative AI? Start by thinking smaller

AI developers train the most publicly visible forms of generative AI, like Chat-GPT and Midjourney, using billions of parameters. Therefore, these large language models need to crawl the web for every possible scrap of information in order to build up the quality of their responses. However, several recent developments in generative AI are “challenging the notion that scale is needed for performance.” 

For example, the most recent version of OpenAI’s engine, Chat-GPT-4, operates using 1.5 billion parameters. That might sound like a lot, but the previous version, GPT-3.5, uses 175 billion. 

Large language models are, one generation at a time, shrinking in size while their performance improves. Microsoft has created two small language models (SLMs) called Phi and Orca which, under certain circumstances, outperform large language models. 

Unlike earlier generations—trained on vast diets of disorganised, unvetted data—SLMs use “curated, high-quality training data” according to Vanessa Ho from Microsoft.

They are more specific in scope, use less computing power (and therefore less energy—another relevant criticism of generative AI models), and could produce more reliable results when trained with the right data—potentially making them more useful from a business point of view. In 2022, Deepmind demonstrated that training smaller models on more data yields better performance than training larger models on fewer data. 

AI needs to find a way of escaping its ethically dubious beginnings if the technology is to live up to its potential. The transition from large language models to smaller, higher quality data training sets would be a valuable step in the right direction.