Can AI Help Manage Nuclear Reactors? (msn.com) 60
America's Department of Energy launched a federally funded R&D center in 1946 called the Argonne National Laboratory, and its research became the basis for all of the world's commercial nuclear reactors.
But it's now developed an AI-based tool that can "help operators run nuclear plants," reports the Wall Street Journal, citing comments from a senior nuclear engineer in the lab's nuclear science and engineering division: Argonne's plan is to offer the Parameter-Free Reasoning Operator for Automated Identification and Diagnosis, or PRO-AID, to new, tech-forward nuclear builds, but it's also eyeing the so-called dinosaurs, some of which are being resurrected by companies like Amazon and Microsoft to help power their AI data centers. The global push for AI is poised to fuel a sharp rise in electricity demand, with consumption from data centers expected to more than double by the end of the decade, the International Energy Agency said Thursday. The owners of roughly a third of U.S. nuclear plants are in talks with tech companies to provide electricity for those data centers, the Wall Street Journal has reported.
PRO-AID performs real-time monitoring and diagnostics using generative AI combined with large language models that notify and explain to staff when something seems amiss at a plant. It also uses a form of automated reasoning — which uses mathematical logic to encode knowledge in AI systems — to mimic the way a human operator asks questions and comes to understand how the plant is operating [according to Richard Vilim, a senior nuclear engineer within the lab's nuclear science and engineering division].
The tool can also help improve the efficiency of the personnel needed to operate a nuclear plant, Vilim said. That's especially important as older employees leave the workforce. "If we can hand off some of these lower-level capabilities to a machine, when someone retires, you don't need to replace him or her," he said... Part of the efficiency in updating technology will come from consolidating the monitoring staff at a utility's nuclear plants at a single, centralized location — much as gas-powered plants already do.
It hasn't found its way into a commercial nuclear plant yet, the article acknowledges. But the senior nuclear engineer points out that America's newer gas-powered plants ended up being more automated with digital monitoring tools. Meanwhile the average age of America's 94 operating nuclear reactors is 42 years old, and "nearly all" of them have had their licenses extended, according to the article. (Those nuclear plants still provide almost 20% of America's electricity.)
But it's now developed an AI-based tool that can "help operators run nuclear plants," reports the Wall Street Journal, citing comments from a senior nuclear engineer in the lab's nuclear science and engineering division: Argonne's plan is to offer the Parameter-Free Reasoning Operator for Automated Identification and Diagnosis, or PRO-AID, to new, tech-forward nuclear builds, but it's also eyeing the so-called dinosaurs, some of which are being resurrected by companies like Amazon and Microsoft to help power their AI data centers. The global push for AI is poised to fuel a sharp rise in electricity demand, with consumption from data centers expected to more than double by the end of the decade, the International Energy Agency said Thursday. The owners of roughly a third of U.S. nuclear plants are in talks with tech companies to provide electricity for those data centers, the Wall Street Journal has reported.
PRO-AID performs real-time monitoring and diagnostics using generative AI combined with large language models that notify and explain to staff when something seems amiss at a plant. It also uses a form of automated reasoning — which uses mathematical logic to encode knowledge in AI systems — to mimic the way a human operator asks questions and comes to understand how the plant is operating [according to Richard Vilim, a senior nuclear engineer within the lab's nuclear science and engineering division].
The tool can also help improve the efficiency of the personnel needed to operate a nuclear plant, Vilim said. That's especially important as older employees leave the workforce. "If we can hand off some of these lower-level capabilities to a machine, when someone retires, you don't need to replace him or her," he said... Part of the efficiency in updating technology will come from consolidating the monitoring staff at a utility's nuclear plants at a single, centralized location — much as gas-powered plants already do.
It hasn't found its way into a commercial nuclear plant yet, the article acknowledges. But the senior nuclear engineer points out that America's newer gas-powered plants ended up being more automated with digital monitoring tools. Meanwhile the average age of America's 94 operating nuclear reactors is 42 years old, and "nearly all" of them have had their licenses extended, according to the article. (Those nuclear plants still provide almost 20% of America's electricity.)
Betteridge's law fail (Score:5, Interesting)
The answer is objectively yes, for the same reason as the question "Does the sky look blue" is yes. These AI tools (as in tools trained with data and operating models) are obvious, they exist, and they have been used by the process industry for a decade already. A nuclear power station is just a process plant that generates electricity.
There are already several vendors on the market that offer such tools for new plants, and vendors that will also allow you to train models based on your existing plants historical data.
Yes we were using AI before you kids thought it was cool.
Re: (Score:2)
The answer is objectively yes, for the same reason as the question "Does the sky look blue" is yes. These AI tools (as in tools trained with data and operating models) are obvious, they exist, and they have been used by the process industry for a decade already. A nuclear power station is just a process plant that generates electricity.
There are already several vendors on the market that offer such tools for new plants, and vendors that will also allow you to train models based on your existing plants historical data.
Yes we were using AI before you kids thought it was cool.
Yeah but ya didn't have the trendy term "AI" back then.
Re:Betteridge's law fail (Score:5, Insightful)
The answer is objectively yes, for the same reason as the question "Does the sky look blue" is yes. These AI tools (as in tools trained with data and operating models) are obvious, they exist, and they have been used by the process industry for a decade already. A nuclear power station is just a process plant that generates electricity.
There are already several vendors on the market that offer such tools for new plants, and vendors that will also allow you to train models based on your existing plants historical data.
Yes we were using AI before you kids thought it was cool.
Yeah but ya didn't have the trendy term "AI" back then.
We did call it "AI", but some people made wild promises and claims, far beyond what the "AI" software could do. Claimed it was "thinking" etc.
So there was massive backlash and the "AI" industry
was totally destroyed. At that point, if you put the word
"AI" or even just "intelligence" on anything, you would
go out of business instantly because of the bad rap.
But the software never went away.
It was very successful and has been for 40 years.
Of course, it was not based on LLMs,
which is a fundamentally unreliable approach.
Re:Betteridge's law fail (Score:4, Insightful)
Automation is not intelligence. Control systems have never been, and are not referred to as intelligent in any way.
Sure, using AI to help with analysis of fatigue and behaviour might be helpful. Providing helpful teaching hints is good too.
Predictive system warnings is risky, you'd want to at least have independent verification of what it's warning of. And that means the operators have to be trained for confidently making judgement calls over the AI.
Throwing a modern AI directly onto the controls of something that requires rigorous design and build would be a catastrophe in the making. It'd be the real Homer Simpson live act.
Re: (Score:2)
Automation is not intelligence. Control systems have never been, and are not referred to as intelligent in any way.
Sure, using AI to help with analysis of fatigue and behaviour might be helpful. Providing helpful teaching hints is good too.
Predictive system warnings is risky, you'd want to at least have independent verification of what it's warning of. And that means the operators have to be trained for confidently making judgement calls over the AI.
Throwing a modern AI directly onto the controls of something that requires rigorous design and build would be a catastrophe in the making. It'd be the real Homer Simpson live act.
It is kind of a strange thing isn't it? Sometimes I think that there is some sort of "anti-skill" race going on. Where somehow, we must make all things perform well with people who otherwise would be stocking shelves at Wal-Mart.
So work will simply require skill and better than average thinking ability. And I agree, it would be a quick disaster to just have Beany pressing buttons based on what some hallucinating LLM tells Beany to do.
One thing is for certain, novel situations can and will come up. An
Re: (Score:3)
Sometimes I think that there is some sort of "anti-skill" race going on. Where somehow, we must make all things perform well with people who otherwise would be stocking shelves at Wal-Mart.
That's exactly what's happening. People with advanced skill sets command high salaries, and worse, there are only a relatively small number of them, so even if you have plenty of money to pay them with, you'll be limited in how many you can hire at any price, which limits your company's ability to scale up.
So, the push is on to McDonalds-ize everything; i.e. reduce every task to a set of well-documented, standardized, simple steps that can be performed by anyone, and then hire anyone who is willing/able to
Re: (Score:2)
Sometimes I think that there is some sort of "anti-skill" race going on. Where somehow, we must make all things perform well with people who otherwise would be stocking shelves at Wal-Mart.
That's exactly what's happening. People with advanced skill sets command high salaries, and worse, there are only a relatively small number of them, so even if you have plenty of money to pay them with, you'll be limited in how many you can hire at any price, which limits your company's ability to scale up.
So, the push is on to McDonalds-ize everything; i.e. reduce every task to a set of well-documented, standardized, simple steps that can be performed by anyone, and then hire anyone who is willing/able to do that menial job for the least amount of money. In the endgame, that least-common-denominator worker will be a robot, and the investors will keep all the salary money for themselves.
The problem with that is if they succeed, there won't be money to extract from now unemployed people. At that point, the uber wealthy will start to eat their own.
Re: (Score:2)
Automation is not intelligence. Control systems have never been, and are not referred to as intelligent in any way.
There have certainly been "AI" systems (using OPS-5 type or Prolog "unification" conferencing rules) connected to automation and making real-time control decisions. We did that since the mid-1980s.
Marketing-wise, there have been all kinds of things that I would not consider AI advertised as "AI". In the domain of feedback control systems, I remember an electric shaving razor ca. 1995 being advertised as having "AI". These days, people put the "AI" label on all kinds of things, just because "AI" is hot.
They
Re: (Score:2)
Spell corrector didn't like "inferencing" apparently.
I wonder if it has AI?
Re: (Score:2)
Expert System was one buzzword I remember well.
Fuzzy Logic was another one.
Then there was this one,
https://www.chemicalonline.com... [chemicalonline.com]
I worked with that package, it inspired my Ph.D. However, it was also prone to what they now call hallucinations.
The computer it ran on was a Vax Station 60 with 40 MB of RAM.
Re: (Score:2)
Expert systems claimed to be AI, but were just boolean if/then. Which would make nearly all computer programs "AI"
Re: (Score:2)
Yeah but ya didn't have the trendy term "AI" back then.
What? Of course we did! The term has been around since the 1956 Dartmouth conference.
Re:Betteridge's law fail (Score:5, Insightful)
The key word is "help". They can offer useful advice, but are no substitute for experienced and knowledgeable staff.
They don't solve the fundamental problem, which is that such staff are expensive, take a long time to train up, and need continual updating and replacement over the 50+ year lifespan of the plant. It doesn't solve issues like them needing to be empowered to take actions that hurt profits, and resist corporate pressure to reduce safety.
In fact if anything these kinds of "AI" are a threat to that, because corporate will demand they are programmed to meet their requirements, and use them as an excuse to reduce staffing costs.
Re: (Score:2)
I'm sure the operators at Chornobyl thought the same.
Re: (Score:2)
They don't solve the fundamental problem, which is that such staff are expensive
Actually that is something they definitely do. Datamining is HARD, you can throw a lot of people at the problem. Automatic inference based on a digital twin using secondary sources is far cheaper and requires less manpower than going out and taking primary measurements.
In one area I have worked at we have objectively reduced the amount of manual work in the field collecting primary data on large compressors as the result of the trained digital twin models. We no long have to go out and take monthly measurem
Re: (Score:2)
They don't solve the fundamental problem, which is that such staff are expensive, take a long time to train up, and need continual updating and replacement over the 50+ year lifespan of the plant.
Well, these guys don't seem to agree. "If we can hand off some of these lower-level capabilities to a machine, when someone retires, you don't need to replace him or her,"
Re: (Score:2)
I think this is the "duh" case
Betteridge Holds...sort of (Score:3)
So while machine learning could definitely help I woul
Re: (Score:3)
The answer is objectively yes, for the same reason as the question "Does the sky look blue" is yes. These AI tools (as in tools trained with data and operating models) are obvious, they exist, and they have been used by the process industry for a decade already. A nuclear power station is just a process plant that generates electricity.
There are already several vendors on the market that offer such tools for new plants, and vendors that will also allow you to train models based on your existing plants historical data.
Yes we were using AI before you kids thought it was cool.
A nuclear power plant is not just a process plant that generates electricity. It is also a giant bomb waiting to go off and spew radioactive steam into the atmosphere, turning everything within a twenty-mile radius into an exclusion zone for tens of thousands of years or more, and turning every water well within an even larger distance downstream into the fake grail from Indiana Jones. I mean, sure, Chernobyl was worse than other plants would have been because of its design (graphite moderator), but even
Re: (Score:2)
A nuclear power plant is not just a process plant that generates electricity. It is also a giant bomb waiting to go off and spew radioactive steam into the atmosphere, turning everything within a twenty-mile radius into an exclusion zone for tens of thousands of years or more, and turning every water well within an even larger distance downstream into the fake grail from Indiana Jones. I mean, sure, Chernobyl was worse than other plants would have been because of its design (graphite moderator), but even with a modern plant, a catastrophic failure is a spectacularly big deal.
There's many kinds of nuclear power plants, and you want me to believe they all could blow up and "spew radioactive steam"? Explain how TerraPower's Natrium reactor could do that? It has a molten sodium coolant, there's no water in the system that could flash into steam like Chernobyl. Or consider the liquid fluoride thorium reactors from Flibe Energy. None have been built yet, and there's been a few variations offered, but surely you can explain how a reactor that has molten fuel can melt down and "spew radioactive steam", right?
By burning through the floor down to the water table. The only way to guarantee that you won't have problems like that, short of building it somewhere with a very deep water table (which likely would not be close to where the power will be used, which causes a different set of problems) is to limit the size of the plant, which likely results in lower efficiency. You can do it, but there are tradeoffs, and either way, none of that applies to plants that already exist.
When it comes to turning an area into an exclusion zone for tens of thousands of years, what specific isotopes would create this hazard? You must know. Around Chernobyl the primary hazards are caesium-137 and strontium-90, both with a half life of about 30 years. If we use the "rule of thumb" on 10 half lives to where an isotope is considered "gone" then that's 300 years. That's not nothing, but it's also far from thousands. That's also an exceptional case for all kinds of reasons. That's also an estimate based on theories that has yet to be tested in the real world, Chernobyl is the first such test, and it could take 300 years to run that experiment to the end. As it is now there's thriving wildlife in much of the exclusion zone, not all of it because the radiation was not spread evenly.
There are people who live in parts of t
Re: (Score:1)
By burning through the floor down to the water table.
If you believe that possible then you watched The China Syndrome a few too many times. I thought I might go over how that can't happen but then realized you'd either not believe me, and it would be better if you heard how a Natrium reactor works elsewhere, or you'd not believe anyone that says different because you've drank the kool-aid or something.
When you build nuclear plants, you tend to build them near major metropolitan areas, because that's where the power will be used, and doing so reduces transmission losses. This makes safety rather critical, because permanently evacuating a major city, even for a few hundred years, would be catastrophic.
Um, that's easy. Have rules that prohibit nuclear power plants near population centers even if it means higher transmission losses.
If you want to be a "not in
Re: (Score:2)
It is also a giant bomb waiting to go off and spew radioactive steam into the atmosphere
That shows you don't know much about the process industry. Most things a giant bombs waiting to go off. So far nuclear hasn't done remotely the damage or caused remotely the fatalities or injuries of the wider process industry.
So although the answer is ostensibly yes, I would argue that the correct question is "Should AI be used for this," and the answer to that is solidly "no", which means the answer to "Can it be used" also should be "no".
You don't seem to understand the subject. These tools have made our ability to predict and plan for unforeseen events better. They reduce risk. They improve maintenance targeting. The answer of should AI be used for this is definitely "yes" when you understand what these tools are, ho
Re: (Score:2)
You fail to understand what these system is doing. Sensors make a decision based on a point in time. Experts draw conclusions of changes over time. AI does the latter. We actually already used trained models specifically to address your concern: sensors failing. Shit that's an off the shelf product from Palantir which uses a trained model of your plant to tell you when a sensor is failing based on a deviation from the expected behaviour of the digital twin. Specifically the kind of thing which could identif
not really. (Score:2)
and then absolutely not. like, not at all
Automated reasoning? (Score:2)
Doc Ock (Score:2)
Sure (Score:3)
Just as long as the C-suite is still protected. Put it anywhere but there!
PR fail (Score:2)
it's drawing power directly from the main reactor (Score:2)
How long will it be before all of us "simply get in the way? "
Hmmm - I have questions (Score:5, Insightful)
"The tool can also help improve the efficiency of the personnel needed to operate a nuclear plant, Vilim said. That's especially important as older employees leave the workforce. "If we can hand off some of these lower-level capabilities to a machine, when someone retires, you don't need to replace him or her," he said... Part of the efficiency in updating technology will come from consolidating the monitoring staff at a utility's nuclear plants at a single, centralized location.
So this is a way to have a rando off the streets replace experienced older people who have a lot of skill and knowledge. Why they were going to apply for a job at Burger King, but they'll do great managing a humongous nuclear reactor and power generating plant and don't need to know anything at all. "Just do as the AI tells you, Binky. Push the AZ-5 button."
As much fun as it is to hate on those older experienced people, They know their shit, and can come up with novel solutions that might save our asses in an unexpected emergency. This so called solution seems like something dreamt up by accountants. I've seen cases where smart and skilled people were replaced by plug-ins to save money, until it didn't.
Which reminds me, when the AI hallucinates and ends up scramming the reactor or overspins the turbines - who is responsible for any damage? Remember even if it directs the Burger King kid to do it - seems like a question worth asking.
Re: (Score:2)
The business is responsible for any damage regardless of whether it was caused by an employee, software, or even hardware.
Given how expensive down time is for a nuclear reactor, smart operators lean heavily towards safe dependable operation.
Not all are smart though, as seen with Fukushima not moving their generators or installing hydrogen reformulators as recommended.
Re: (Score:2)
Those sort of businesses are setup to be folded again the moment any negativity sets in.
There's a limit to shell companies (Score:2)
That works for dodgy home contractors, much less so for nuclear power plants.
Re: (Score:2)
As far as I know no commercial nuclear power station in the US has ever caused a utility to go bankrupt. There is always a bailout and the people in charge never are held per
Re: (Score:2)
Your are 100% wrong.
I'm as wrong as this sentence is grammatically correct.
Except for things like slip and fall or industrial workplace accidents nuclear facilities are uninsurable.
Wow, what a way to shoot yourself in the foot with an unresearched declaration. You need to read up on the Price-Anderson Act. [wikipedia.org]
There's actually three tiers of insurance for nuclear power:
Every single nuclear power plant carries $500M in private insurance. [amnucins.com]
Next, there's a sort of co-insurance where all the nuclear plants would get together to cover stuff, that's around $15B.
Only after that would the government maybe have to pony up anything.
AI nukes (Score:2)
Re: AI nukes (Score:1)
File this under things that sound like bad ideas (Score:4, Insightful)
Re: (Score:2)
In times of high energy demand, control rods are raised allowing more reactions to take place. In times of low demand, control rods are lowered to absorb neutrons and hence decrease the number of reactions; less reactions = less heat energy produced.
Re: (Score:2)
Re: (Score:2)
Does this sound like a bad idea though? Maybe if they had an AI model a specific incident from the 80s wouldn't have happened. It was clear that the actual people involved weren't making sound decisions. Replacing them with a computer may not be so bad.
It reminds me of an Isaac Asimov short story. I can't remember the name of it, but the plot was something about an AI gone rogue on a space solar collector that beams microwave energy down to earth has locked humans out of the facilities systems and there was
Re: (Score:1)
This is https://en.wikipedia.org/wiki/... [wikipedia.org]
Re: (Score:2)
It was clear that the actual people involved weren't making sound decisions. Replacing them with a computer may not be so bad.
If you're going to build a nuclear reactor, you need to be confident that the people involved are going to make sound decisions. If you're not sure about them, then you shouldn't build a nuclear reactor. Replacing the people with a computer doesn't help, because you then need to be confident that the computer will make sound decisions... and the computer is also designed by people, so you haven't actually taken people out of the process at all, you've just made their job harder since now your people have
Relaaax (Score:2)
Nothing can go wrong, go wrong, go wrong, go wrong, go wrong, go wrong...
Re: (Score:2)
I'm still waiting for fuzzy logic (Score:2)
To show its results
all accidents will be human error (Score:2)
Oh this can’t be happening! You’re ope (Score:2)
Oh this can’t be happening! You’re operating without a T-437, Springfield!
Yes (Score:2)
Humans in control? (Score:2)
But AI will become increasingly competent.
And it is certain that one day it will transition to a supervisory role, with humans no longer able to keep up with the spectacular advancements we must expect.
It is illusory to think that humans will always be in control.
Re: (Score:2)
But AI will become increasingly competent.
No, it's not.
And it is certain that one day it will transition to a supervisory role, with humans no longer able to keep up with the spectacular advancements we must expect.
Now you're just writing bad science fiction.
Still Relevant... (Score:2)
https://xkcd.com/463/ [xkcd.com]
I am insulted (Score:2)
As a former Navy trained Reactor Operator on a submarine I am really insulted by this idea that a computer AI do what I did. Remember, humans make mistakes sometimes but to really foul things it takes a computer.