In the early days of growing environmental awareness, the ‘paperless office’ was hailed as a release from the burden of deforestation, then the most urgent concern. The machines that replaced filing cabinets came with other, less visible, environmental costs. The latest generation of machines are the dirtiest we have ever produced, and we need to factor their carbon impact into our environmental planning.
When mandatory ESG reporting was introduced in the UK, the technology sector was not among the first sectors required to comply. Part of the reason that the tech sector draws less attention to itself is that we don’t have we don’t have clear headline busting statistics to rely on. For example, according to Google.com, one internet search produces approximately 0.2g of CO2. If your website gets around 10,000 views per year, that’s around 211 kg per year. Add a chatbot functionality to that website and you jump into a whole different league.
The hidden costs of new algorithms
Chatbots are based on Large Language Model algorithms, which have very little in common with the search browsers that we’re more familiar with, even if their interfaces look familiar. Every time you run your query in a service like Bard, LLama or Co-Pilot, the machine has to traverse over every data point in its network. We don’t know for certain how big that network is, but estimates for exemple, that ChatGPT4, runs on around 4 x 1.7 trillion bytes are plausible.
We aren’t yet able to measure how much CO2 that produces with every query. Estimates range from 15 to 100 times more carbon produced on one sophisticated chatbot request compared to a regular search query, depending on how you factor into the equation the trillions and trillions of times that the machine had to run over that data set during the ‘training’ phase, before it was even released. And many of us are ‘entering queries’ with the casual back-and-forth conversational style like we’re chatting to a friend.
Given that these machines are now responding daily to trivial and minor requests across organisational networks, the CO2 production will quickly add up. It is time to look at the environmental bottom line of these technologies.
Solutions on the horizon
Atmospheric carbon may come under some control soon. In the heart of Silicon Valley, the California Resources Corporation saw their plans for carbon capture and storage reach the draft permission stage earlier this month. There are another 200 applications for similar projects waiting in line. Under such schemes, carbon is returned to the earth in ‘TerraVaults’. The idea is to remove it from the atmosphere by injecting it deep into depleted oil reserves left behind after fossil fuel extraction. It’s the kind of solution that is popular, because it takes the onus of lifestyle change away from the public. However, it’s a controversial technology that divides environmental experts.
Only half an answer to a complicated problem
It also only addresses half the problem. These supercomputers burn through carbon at a shocking rate when they power up. They also need electricity to cool down. In fact, it is estimated that 43% data centre electricity could go on cooling alone. Regional water stress is a major part of the climate problem, too. Data centres guzzle water to run their cooling systems at a rate of millions of litres of water per year. This is nothing, however, compared to the volume of water needed to run the steam turbines to generate the electricity. It’s a vicious cycle of depletion.
It is an irony that the supercomputers that threaten the environment are also needed to save it. Without the kind of climate modelling that a supercomputer can provide, it will be harder to respond to climate challenges. Supercomputers are also improving their own efficiency. Manufacturers today use processors that constantly try to operate at maximum efficiency – a faster result means less energy consumption. These top end dilemmas over whether to use these machines are similar to those faced at an organisational level. At what point does it become worthwhile?
What you can do
We need to develop a culture of transparency around the true cost of these sophisticated technologies. Transparency supports accountability and it benefits those who are doing the right thing. There are data centres that use 100% renewable energy today. Some, like Digital Realty, have even achieved carbon net neutrality in their operations in France. As more of us ask uncomfortable questions about where our chatbots are powered, we’ll start to get better answers.
In the meantime, the solution lies mostly in sensible deployment of these technologies. If your organisation is committed to the drive to net neutrality, it is worth considering where and how you apply these advanced technologies to meet with commitments your organisation has made. A customer facing chatbot may not be the optimal solution for your business or environmental needs.
- Data & AI
- Sustainability Technology
