Google and Westinghouse lean on AI to speed US nuclear plant builds

Google and atomic power biz Westinghouse Electric claim that AI will speed construction and cut the cost of building the new US power plants it is planning in response to rising demands for energy to fuel AI. The pair announced back in July that they are working together to transform how nuclear reactors are constructed and …

[ajadedcynicaloldfart]

Dear God

Please make this A.I. thing go away. Because A.I. plus nuclear reactors does not fill me with confidence.

[Dan 55]

Re: Dear God

All that process, design, and testing was slow and inefficient, now we're designing nuclear reactors by bingo number machine. I don't see what could possibly go wrong.

[JohnSheeran]

Re: Dear God

Up next.......Quantum A.I. !!!!!

[vtcodger]

Re: Dear God

I don't see what the problem is here. AI designed nuclear plants should be perfectly OK. As long as they are restricted to orbits around Neptune. (I'm pretty sure that's far enough from my back yard that their probable spectacular demise isn't much of a problem).

[Neil Barnes]

So...

Nuclear power is required to power AI servers.

AI servers are required to design the nuclear power generators.

Seem a bit circular. Couldn't we do away with both, and save a metric shedload of money as a by product? Or at least use the power generators for people's needs, y'know, heating, lighting...

[codejunky]

cool

"However, while this shows that AI does have useful applications, we're not convinced that optimizing schedules is something that requires AI to accomplish."

Thankfully we have the market to allow them to test their theory with their money and if it succeeds we get to benefit from the knowledge and technology to improve construction times.

[vtcodger]

Re: cool

That's fine. But how about we spend a decade of three proofing the technology by optimizing the design and construction of supermarket parking lots and hiking trails in wilderness areas before applying it to stuff that might blow up?

[codejunky]

Re: cool

@vtcodger

"That's fine. But how about we spend a decade of three proofing the technology by optimizing the design and construction of supermarket parking lots and hiking trails in wilderness areas before applying it to stuff that might blow up?"

As already pointed out by another commenter the AI is about scheduling the development. I wouldnt be shocked if AI is already being turned to those things along with everything else that is being tried just to find uses for this technology. Which again comes back to thankfully the market lets people try stuff and risk their own money so we can get the benefits of what works.

[Anonymous Coward]

Re: cool

Which again comes back to thankfully the market ...

Careful codejunky, your religious fervour for The Holy Market might lead some to consider you a religious zealot. Blind to everything except Mammon.

(Next you'll be using AI to generate canticles to the commodities traders.)

[codejunky]

Re: cool

@AC

"Careful codejunky, your religious fervour for The Holy Market might lead some to consider you a religious zealot. Blind to everything except Mammon."

If it sounds like worship to you then you probably need to go back to school. Pointing out the observation of what works and why this isnt a problem shouldnt seem like such a religious thing to you unless your upbringing has yet to educate you on a thing called life.

Good luck

[Anonymous Coward]

Re: cool

As you cast shade on others' education, remind us all which degree you studied for? Engineering, science? Or was it an arts/humanities "McDegree" as you term them? (Theology perhap?)

[vtcodger]

Re: cool

Sorry, but I don't share your rather naive faith in markets. Recommended reading -- "Only Yesterday" and "Since Yesterday" by Frederick Lewis Allen. Both very readable. The first deals with the 1920s -- a period almost as whacked out as today. The second with the 1930s which deals with the catastrophic consequences of the failure of financial marketplaces in the 1920s to limit excess.

Markets can be very useful. But history tells us that they don't always self regulate properly. So I have to be pretty skeptical that they are the appropriate single tool for dealing with anything as touchy as nuclear power plants. Particularly at a time where it appears to many of us that the financial markets have gone totally off the rails and there is likely one hell of a storm brewing.

[codejunky]

Re: cool

@vtcodger

"Recommended reading -- "Only Yesterday" and "Since Yesterday" by Frederick Lewis Allen."

Cheers I will take a look at those.

"Markets can be very useful. But history tells us that they don't always self regulate properly."

The natural and real world is to 'correct' for mistakes/mis-allocations of resources. The market is to allow some to take risk (and create rewards) and explore options which is the natural way to figure things out. There is no self regulation beyond the natural correction that occurs because something doesnt work. Governments are not self correcting, authoritarian management isnt and markets are just many people making their own risks. The regulation is the correction regardless. In the end it is all people and we have our flaws and limits to our ability.

"So I have to be pretty skeptical that they are the appropriate single tool for dealing with anything as touchy as nuclear power plants."

Yet it worked otherwise how do we have competing designs to make a better power plant? So it already worked. As for this AI story it seems to be about scheduling the build not designing, but even if it was it would still be people on the hook for it going wrong. Markets need regulation but people being free to try, risk and fail is necessary for progress and made the difference between the US and USSR or even East v West Germany. The fast pace rise of China shows that it works vs the heavy regulation before.

"Particularly at a time where it appears to many of us that the financial markets have gone totally off the rails and there is likely one hell of a storm brewing."

I agree. And of course governments are incentivising and inflating the bubble hard. Here in the UK we are seeing the government reach its limits in over inflated spending. And corrections are just the natural regulation that brings us all back to reality.

[Helcat]

Thankfully AI isn't designing, just organising the build schedule.

However, all that means is if it's only looking at dependencies, it'll miss that building D before B means you can't get the equipment in to build B as D is now in the way.

Think of it like this: You're building the reactor, but there's a delay, so you bring forwards the staff canteen: No dependency here. Only the canteen is being built on the spot where you need the crane to be positioned to lower the reactor shielding into place. There'd be no problem at all if the shielding was already placed, but that's delayed and there's nothing to tell the AI that the canteen will become a hard blocker if built before a specific point in the project - a thing a human might spot.

[VicMortimer]

That's terrifying.

While I think more nuclear power is probably a good idea, and if the choice is between nuclear or fossil I'd rather live near the nuke every time, I'm not sure that would continue to be the case if a LLM designed any part of the nuclear plant.

Why are people this stupid?

[Tron]

Re: That's terrifying.

quote: I'd rather live near the nuke

You're in luck. There is loads of land to be purchased cheaply around Fukushima Daiichi.

[Brewster's Angle Grinder]

"we're not convinced that optimizing schedules is something that requires AI to accomplish"

Of course, conventional software could handle it. But a bespoke, deterministic solution would take months if not years to develop. (And optimisation problems are notorious hard. Hard as in NP)

So, instead, we hand it to a general purpose LLM and pray the screw ups cost less than the cost of developing software that is provably correct.

[Tron]

Nuclear power, built cheaply and quickly, with AI.

I wonder what that will do to property values downwind.

[thames]

AI?

I suspect there's very little if any AI involved in this software. More likely is that they called optimization algorithms "AI" for marketing purposes.

[elDog]

Main page shows 9 comments. When I come to be my pithy self there are no comments

Is this such a touchy subject that every comment so far is ending up in the hoosegow?

[Dutch Eagle]

Re: Main page shows 9 comments. When I come to be my pithy self there are no comments

I've seen something similar myself. I found that going back to the article and then clicking the Comments link then shows the comments.

Unsure if a simple page refresh would also cure it.

[Anonymous Coward]

Isn't the usual biggest roadblock Greenpeace and their irrational hysterical ilk?

[IGotOut]

Grok, tell us how to safely build a nuclear reactor

Sure, what a great question, the sort only a white man could come up with.

Well Chernobyl was caused by the glorious Russians being attack by a slany eyed wave from Japan. This in turn caused a radioactive cloud of zeta rays to leak into the North Sea, thus contaminating 3 mile island in the beautiful USA.

[Dan 55]

Re: Grok, tell us how to safely build a nuclear reactor

Grok can't answer you at the moment, all GPU power is dedicated to simping for Elon:

Elon Musk’s Grok AI tells users he is fitter than LeBron James and smarter than da Vinci