What's AI's Environmental Impact?

­Bill Gates recently made a throwaway comment about AI’s environmental impact, kickstarting one of the most critical debates vexing this industry (and the future of this planet) – is artificial intelligence a net positive for Mother Earth?

Gates claimed that AI will help countries hit their climate goals, despite the tech’s insatiable power demands and carbon footprint. And while, yes, Gates has a personal stake in it – the Microsoft co-founder invests in AI through the Gates Foundation – weighing AI’s environmental gains and debits is no small feat.

Earlier this year, we ran an article from Vicor, which noted that by 2027 “AI servers could use 85 to 134 terawatt hours annually” (the yearly electrical energy consumption of Argentina).

Navitas pointed out that incorporating snazzy new purpose-designed AI processors into data center servers could help raise typical server power demands from 30-40 kW today to 100 kW per cabinet in the future.

Then there’s the steep toll for training AI models – one estimate posits that training programs like GPT-3 uses about 1,300 MWh of electricity, with another source claiming that training produces about 626,000 pounds of carbon dioxide. An oft-repeated stat says that producing one generative AI image draws enough juice to power your phone.

On the other hand, innumerable applications use AI for efficiency gains, and sometimes, those gains translate to energy savings.

Scientists at the DOE’s Princeton Plasma Physics Laboratory (PPPL) and Princeton University are using AI deep learning to speed the development of safe, clean, and virtually limitless fusion energy.

The Argonne National Laboratory is applying generative AI techniques, machine learning, and simulations to identify suitable metal-organic framework materials to selectively absorb carbon dioxide.

Over in the People’s Republic, Chinese researchers are using machine learning to estimate solar radiation components, eliminating the need for local ground truth data for calibration.

And it’s not even a recent phenomenon – back in 2019, we were reporting on AI’s involvement in spray-on solar cells and reducing hydropower dams’ carbon footprint.

Even the language models are seeing efficiency gains – just recently, Researchers from the University of California, Santa Cruz have enabled a billion-parameter-scale large language model to run on a mere 13 watts.

So will AI save more energy than it consumes? Bill Gates certainly thinks so, concluding that “Datacenters are, in the most extreme case, a 6% addition [in energy demand] but probably only 2 to 2.5%. The question is, will AI accelerate a more than 6% reduction? And the answer is: certainly.”

Given all the recent developments (and companies using green energy as a PR victory), I’m willing to believe AI can be a net positive for the environment.