OPINION: By Sherif Elsayed-Ali, a multiple founder in tech and AI and co-founder of Nexus Climate.
AI is in every conversation about technology and innovation. It’s dominated tech for a year and a half. It’s been touted as a solution to everything, from productivity, to education, healthcare and the climate crisis. But I’ll let you in on a secret: this AI, the type that everyone is talking about, has done nothing for the climate crisis.
Actually, scratch that. It is responsible for a very large increase in emissions from data centres. This type of AI, known as generative AI or GenAI for short, is what we encounter in large language models (LLMs) in applications like ChatGPT and image generation tools like MidJourney.
There is a lot of investment going into GenAI – whether all the hype is justified or not is for another discussion (my short take: no, it’s an overhyped bubble that will burst). What I am concerned about is that the GenAI hype train is sucking all the air out of the proverbial room when it comes to the role of AI in combating the climate crisis.
GenAI is one subset of the much bigger field of AI and machine learning. It can have some use for people and organizations tackling climate problems, mainly around analysing data and information. When it comes to directly decarbonizing anything though, LLMs and their image and video brethren are pretty much useless.*
But there are numerous AI and machine learning technologies that are directly contributing to decarbonization today. They’re not chatbots and they don’t produce videos at the press of a button. And, they tend to be much less power-hungry than GenAI. A great example is time-series forecasting, which uses past values in a time sequence dataset to predict future values. It’s not sexy, it doesn’t get headlines, but it can be extremely effective.
AI and machine learning are today helping decarbonize numerous sectors and industries with many startups tackling big challenges with their AI products. To name a few: Carbon Re, making fuel combustion in cement production more efficient;** Concrete AI, optimizing concrete mixes; Foresight Data Machines, increasing the efficiency of steelmaking and BrainBox AI making cooling buildings more efficient. Numerous other exciting use cases are emerging, for example Orbital Materials is using AI to design new low carbon materials and Princeton scientists are using AI to control fusion reactions.
These AI solutions are hard. They need not just AI expertise, but deep domain expertise. They need AI engineers and researchers working hand-in-hand with experts in industrial processes, chemistry and materials science, among others. They need lots of data that is almost always proprietary.
Building an AI product that delivers real emissions reductions is much harder than sticking an LLM into a software application. Real AI solutions for climate have huge potential to enable and accelerate decarbonization – and therefore very significant economic and financial return potential. AI is a very important tool in tackling climate change, but we need to stop talking about LLMs.
*This doesn’t take anything away from the transformer architecture underlying GenAI and its broader applications
**For transparency, I am a co-founder of Carbon Re.
