Why claims that AI will help solve climate change are just hope coda

The Hope Coda

For decades, whenever a journalist wrote about climate change they almost always had to add a “hope coda” - the section at the end that started with “But its not all doom and gloom…” or words to that affect that would try to lighten the mood before the end of the article. The reasoning was that scaring people would incapacitate them and they would therefore not act on climate change. Or perhaps there was a concern about losing readers. Whatever, despite - or perhaps because of - the regular dose of hope, we didn’t act on climate change in a timely way, and we still aren’t acting with anywhere near the urgency needed. I’m using “we” loosely, I’m not going into who did or didn’t act just now.

I’m noticing a similar trend in AI articles which may talk about the risks of AI to society (e.g. in the second half of this podcast) or the heavy energy and water use of AI approaches that process (e.g. this article). This is the hope coda that “AI will help us solve climate change”. I think we need to challenge these claims about AI solving climate before the repeated claim become an illusory truth (if it hasn’t already).

How do we solve climate change?

Firstly, what are the general issues that we need to address to solve climate change? In a literal sense, there’s just one issue - we need to bring greenhouse gases in the atmosphere and ocean to levels that stabilise the climate within safe limits. With this aim, the UK’s Climate Change Committee divide the 6th Carbon Budget into 11 main sectors that need action to stabilise greenhouse gas emissions:

1. surface transport
2. buildings
3. manufacturing and construction
4. electricity generation
5. fuel supply
6. agriculture and land use
7. aviation
8. shipping
9. waste
10. F-gases
11. greenhouse gas removals

This tell us materially what we need to do but there are still social and political - systemic - issues that are preventing action. Stoddard et al. (2021) looked through 9 lenses on why we haven’t started reducing greenhouse gas emissions in 3 decades:

1. climate governance
2. the vested interests of the fossil fuel industry
3. geopolitics and militarism (manifest in global extractivism)
4. economics and financialization
5. mitigation modelling
6. energy supply systems
7. inequity
8. high-carbon lifestyles
9. social imaginaries

This decomposition of the problem of climate change to both material and systemic aspects is useful for considering where any possible solution can apply.

What can AI solve?

I would break down the applications of AI into 3 groups (but I’d be happy to hear your thoughts):

1. measure and monitor climate damage, adaptation and mitigation
2. optimise existing products and processes
3. accelerate the development of new products and processes

Measure and monitor climate damage, adaptation and mitigation

The most common form of hope coda is “all is not doom and gloom, we can leverage AI to detect wildfires/flooding/methane/penguins/etc.” This is all very well, but it isn’t really tackling any of the core material or systemic areas. Its useful to know how well or badly we are doing. However, given that generally the sensor is more important than the data processing, the processing-hungry AI approaches that are the topic of these articles are rarely justified.

Optimise existing products and processes

Most AI is optimisation of some sort and thus searching across a high-dimensional space for better solutions is an excellent application of these technologies. They can take the tedium out of logistics, optimising supply chains, or improving the efficiency of a vehicle or a turbine blade. Unfortunately, increasing efficiency can result in a consequent increase in use - Jevons Paradox. It is unlikely that improving efficiencies alone is really going to have much impact on our emissions. Instead, inefficient activities (flying for instance, except perhaps one day with dirigibles), need to have their supply massively limited.

Accelerate the development of new products and processes

As a search process, optimation may also accelerate the search for materials such as for batteries or greenhouse gas removal compounds. Again, this is not to be sniffed at but these are specific carefully-designed applications of AI, which are quite different from the applications (usually generating text or images) to which the hope coda article alludes.

Why I distrust the hope coda

Despite what the tech oligopoly would have us believe, AI is not being more creative than humans. Even the quirky outputs from some generative AIs are not truly original, they are simply samples from a distribution defined by huge datasets of works made by loads of people. AIs tend to interpolate well, where there are lots of examples in these datasets, but extrapolation beyond the examples in the dataset is usually dreadful. Thus, AI will not have some wild new insight that will convince corporates to ignore their profit motive for the sake of humanity, governments to prioritise social over economic issues, or individuals to change their carbon-intensive lifestyles.

Yes, AI provides tools to help us understand how well or badly we are doing for the climate, to gain us a little more efficiency in transport, buildings and energy storage and supply, and maybe one day to draw down a little more greenhouse gas. But that leaves most aspects of climate action untouched by AI. It would be good to ultimately put numbers on the benefits of these climate benefitting AIs. However, I also believe that the hope coda is too often equivocation - solution-AI rarely requires the most energy-intensive algorithms or socially risky applications that make up problem-AI.

The hope coda worries me because it reads as a lazy excuse to carry on AI-ing without any real thought about the consequences. At its best it draws attention to the climate crisis. However, at its worst it’s another form of greenwash and climate delayism that gives governments, corporates and anyone else looking for one, an excuse for continuing to put off real urgent climate action.