Microsoft believes AI can hasten development of nuclear fusion as a practical energy source, which could in turn accelerate answers to the question of how to power AI.
Nuclear fusion – in the context of generating electricity – is a technology that, like quantum computing, exists in theory and lab experiments but hasn’t been implemented at practical scale. Test reactors only fleetingly produce more energy than they consume.
Enthusiasts believe it will deliver all-you-can-eat clean energy, a welcome prospect for Microsoft as it develops AI products with a voracious appetite for energy.
Microsoft Research and various fusion luminaries therefore see an opportunity to advance the state of the art using machine learning, now branded “AI.”
“The pursuit of nuclear fusion as a limitless, clean energy source has long been one of humanity’s most ambitious scientific goals,” three Microsoft Research boffins – Kenji Takeda, Shruti Rajurkar, and Ade Famoti – wrote in a post published Wednesday.
“While scalable fusion energy is still years away, researchers are now exploring how AI can help accelerate fusion research and bring this energy to the grid sooner.”
Some of that exploration happened in March at Microsoft Research’s inaugural Fusion Summit, a gathering of scientists who hope to accelerate fusion research using AI – a technology upon which Microsoft, coincidentally, is betting billions.
Ashley Llorens, corporate VP and managing director of the Microsoft Research Accelerator organization, opened the event by musing how grand it would be to advance sustainability through the application of compute and AI.
Such speculation is needed because AI is not currently sustainable. As the UN Environment Program observed last year, “The proliferating datacenters that house AI servers produce electronic waste. They are large consumers of water, which is becoming scarce in many places. They rely on critical minerals and rare elements, which are often mined unsustainably. And they use massive amounts of electricity, spurring the emission of planet-warming greenhouse gases.”
Microsoft’s current attempts at addressing AI’s environmental impact involves paying for carbon offsets, acquiring clean energy, and ongoing hardware and software optimization work to make its AI workloads and datacenters more efficient. Hastening the realization of nuclear fusion might make amends for some of the environmental impact of its business, assuming another decade or so of emissions doesn’t compound the problem beyond repair.
Sir Steven Cowley, Lab Director of Princeton Plasma Physics Laboratory in the US and the former chief executive of the UK Atomic Energy Authority, gave the keynote address and observed that more research is needed to even understand if AI is “the key ingredient in finding an optimum configuration for fusion power that really delivers electricity at a cost the consumer wants to pay for electricity.”
Get ready to wait
AI enthusiasts and fusion fans both need patience, because it will be – of course! – a decade or more before experimental fusion energy plants fire up.
The US National Academies of Sciences, Engineering, and Medicine believes public and private sector investment can deliver a pilot power plant sometime between the years 2035 and 2040. That timeframe overlaps with the target date for operations to commence at ITER, an international fusion project being built in France – and which has already suffered long delays.
On the upside, machine learning is already employed for drug discovery so has been proven capable of assisting complex research tasks.
Microsoft’s assembled researchers hope its application in material discovery and partial differential equations, among other research challenges, will help reveal the path to commercially-viable fusion.
Cowley, whose lab has entered into a Memorandum of Understanding to collaborate with Microsoft, argues AI has the potential to shorten the time required to develop functional fusion and represents an alternative to 70 years of trial-and-error.
“[Fusion] is a technology that we’ve never, never done before, and using calculations and AI, right, to actually find a sure route to something that will work – given that every time we make an experiment, it’s going to be many billion dollars – is surely the way to move forward,” he said. “It’s sort of foolish to imagine that we’ll do fusion by trial and error.”
Enter Clippy: “I see you’re building a fusion reactor. Would you like some help with that?” ®