Essay by Eric Worrall
The entire output of the re-opened Three Mile Island nuclear plant will be used to drive artificial intelligence computers.
A once-shuttered nuclear plant could soon return to the grid.
The planned reopening of the Three Mile Island nuclear plant is praised as a boon for Pennsylvania and a boost for AI, but it is loathed by residents still haunted by a near-catastrophic meltdown there in 1979.…
The resurrection of Three Mile Island (TMI) — half of which remained operating after the 1979 meltdown, only closing down due to economic reasons in 2019 — was prompted by Microsoft’s need to fuel its power-hungry data centers.
A revolution in generative artificial intelligence has triggered a surge in energy needs for those data centers, pushing cloud computing giants to look for additional low carbon energy sources.
…For others, the fear and anxiety of 1979 is still strong.
…A series of equipment malfunctions and human errors saw the plant’s Unit 2 melt down in 1979, releasing radioactive materials into the atmosphere and launching mass evacuations.
Read more: https://www.msn.com/en-us/news/world/cheers-angst-as-us-nuclear-plant-three-mile-island-to-reopen/ar-AA1uDRIg
…
It is hard to get your mind around – the entire output of the re-opened Three Mile Island nuclear plant will be used by Microsoft’s artificial intelligence data centers.
Why does AI need so much power? The reason is big AI systems need to sift through trillions, sometimes quintillions of failed attempts to find the solution they are looking for. Even at billions of potential solution evaluations per second it can take days, months or even years to achieve a result. But this process of digitally mining neural networks to find the nuggets of rightness can be accelerated by adding more computer hardware – an AI hardware arms race which has led to the current situation, in which big tech companies have started rehabilitating nuclear reactors, for the sole purpose of powering their next generation data centers.
Anyone wanting to delve deeper into why AI needs so much power can read an explanation in my recent article about AI, and see a real AI in action.
The renewable energy revolution is dead man walking thanks to the rise of AI. Even if better algorithms are discovered, the huge AI data centres won’t be retired, the better algorithms will be used to increase the processing power of existing AI systems, and help design next generation systems which suck even more power.
We stand at the dawn of the Information Age, the age of Artificial Intelligence. Politicians and even tech companies might still pay lip service to renewables, and some of the more intellectually challenged holdout politicians in Britain and Australia still haven’t received the memo, but one thing is beyond doubt: the energy hungry Information Age will not be powered by rooftop solar and “community batteries”, it will be powered by nuclear, coal, gas, whatever scraps of power energy hungry AI companies can scrounge to feed the wolf, to cast into the gaping maw of their frantically growing AI systems, to maintain their competitive edge. Anyone who hesitates or pauses even slightly to question what they are creating will be trampled by the stampede, the race to the finish line.
Drill, baby, drill.
Discover more from Watts Up With That?
Subscribe to get the latest posts sent to your email.
AI is pure empiricism. The only approach proposed is to throw more power, make larger models, more parameters, more teaching data. Nothing suggests any hint of a direction for a breakthrough in lighweigh AI.
The algorithms have limited ability to learn from experience, which distinguishes them from a random walk. The algorithms are working just well enough to justify the crazy money being thrown at them.
Software is incapable of learning or learning from experience.
What can happen is changing the weights in the decision trees to align better with the data reviewed and patterns resolved.
If one cares to define that as learning, well, ok, but that is not my definition.
Unless and until “artificial intelligence” has the ability to reliably distinguish truth from falsehood—which hasn’t happened yet, and indeed is a problem for many human minds—future AI is doomed to inevitable failure since there is and likely always will be more misinformation than valid information on the Web and in most published literature, including many so-called “scientific” articles as duly noted at WUWT over the years.
Simple question: how will AI assess the probable truth or falsehood of a prediction from a “source” such as
Al Gore (https://wattsupwiththat.com/2024/11/16/i-press-my-own-buttons-al-gore-goes-on-unhinged-rant-during-un-climate-summit/ ),
David Viner (https://wattsupwiththat.com/2024/11/20/david-viner-day/ ),
the Club of Rome (https://wattsupwiththat.com/2024/11/17/club-of-rome-cop29-no-longer-fit-for-purpose/ ),
or the IPCC CMIP model ensembles?
Training from fantasy sources is a major issue. Rating The New York Times more “reliable” than The Federalist is an issue.
Along with that, there needs to be a way to prevent spoofing the AI so that the virtual button it pushes, which it thinks to be for the good of mankind, will actually be to activate Sky Net.
With great power also comes great responsibility. Where is Asimov when you need him most?
Somewhat more literary is Harlan Ellison’s short story “I have no mouth, but I must scream” on a dogooder computer program that got a bit too literal minded.
AI can distinguish truth from falsehood. In my Driving AI one of the inputs is always one (inputs range from zero to one). The driving AI quickly learns to ignore it.
If that were true, then all the documented cases of AI “hallucinations” must be false.
BTW, I thought truth/falsehood was a binary function, not analogous to numbers ranging for zero to one.
When I asked an AI bot about this issue, the reply was “That’s a fifty-fifty question.”
(Note: /sarc on my last paragraph above, but not the first two.)
Even human brains can suffer AI style hallucinations, how else do you explain socialism and the climate movement?
As for truth and false, our brains aren’t actually binary about that kind of thing. Millions of people claim to be concerned about climate change and actually are, yet do nothing in their personal lives to reduce their CO2 emissions. My theory is they believe, but not that much.
“Socialism” and “the climate movement” are real phenomena in today’s world, NOT hallucinations.
“Concerned about” and “believe” are far separate—even not applicable—to being able to distinguish truth from falsehood. A person can be “concerned about” and “believe in” a truth as readily as they can a falsehood.
“Socialism” and “the climate movement” are real phenomena in today’s world, NOT hallucinations – absolutely they are real, in the same sense that AI hallucinations are real. What I meant was AI hallucinations and many of the things wrong with today’s world are both the product of defective processing of data.
Then there is that middle ground group which says “that” is certainly true but no one has made it so that me changing my behavior isn’t a great bother.
Human mind is very weak. Even the final sequence of a bike competition is often misinterpreted with the victim confused as the responsible party for a collision.
That depends totally on there being an indisputable “truth” against which tests can be made with certainty. A least a few such things exist but most of human knowledge seems to depend upon disputable viewpoints.
Whereas ‘Wisdom’ comes from the human brain analysing ‘experiences’, and arriving at a ‘truth’ for that Individual… is that not correct?
“ the ability to reliably distinguish truth from falsehood—which hasn’t happened yet, and indeed is a problem for many human minds”
(even when presented with indisputable facts)… is a FACT disputable?
That turns out not to be the case.
Here’s a paper from an industry entrepreneur outlining several prospective improvement methods.
Introduction – SITUATIONAL AWARENESS: The Decade Ahead
Think of the bitcoin mining that could be done with just a fraction of the power that place can put out.
Indeed. The AI surge is going to dwarf bitcoin mining.
I never got the whole “bitcoin” thing. To me it’s the ultimate in imaginary money.
Of what tangible value is a “bitcoin”?
In the US there was a time when for a $20 bill one could exchange it a $20 gold piece, an ounce of gold.
Those days are long gone but at least with under 100 $20 dollar bills, I can still buy a $20 dollar gold piece.
(Hmm … I wonder how many tulip bulbs I could get with a “bitcoin”?) 😎
What saved Bitcoin is the Chinese discovered it was useful for currency smuggling. Like the old Hawala system where a torn dollar note identifies you as the owner of something, by matching the other half of the note.Bitcoin is that torn note.
A most limited understanding.
Federal reserve notes, the overwhelming majority of which exist only on computers, are every bit as unreal a bitcoins, or more properly stated, bitcoins are every bit as real as federal reserve notes (those are not “bills). A note is a promise to pay. Once upon a time, when the monetary act was still in effect, before it was garbled out of all understanding, the courts stated that ‘no one suggests that federal reserve notes will not someday be redeemed but today is not that day (to paraphrase)’
Well, after all these nuclear power plants are up and running and the AI companies realise they just threw a lot of money at a hallucination, maybe we’ll benefit when they decide they need to sell that power to recoup some of their investment…
The AIs work and are returning real value, for example they’re making huge profits lending their AIs to pharmaceutical companies to accelerate drug discovery, or even more profitable, approved drug repurposing. The profits are being reinvested into research to push the envelope on what the AIs can do for their clients.
Do you have any articles I could read? From what you’re saying, it sounds more like pharmaceutical companies are using machine learning, whereas I’m under the impression that it’s generative AI that’s fuelling the spike in power consumption.
They’re all being pretty cagey about what they are up to. The same data centre can be used for generative AI then switched next day do machine learning. I’m not claiming everything they are doing is working or delivering results, but what is working is working well enough to keep them all excited.
Darn.. I had stopped thinking about Big Pharma and its evils.
What could go wrong?
You want a list?
One major problem is enough!
A series of equipment malfunctions and human errors saw the plant’s Unit 2 melt down in 1979, releasing radioactive materials into the atmosphere and launching mass evacuations. The amount of radioactive materials released was negligible and mass evacuation was totally unnecessary, leading to much harm and deaths. Radiophobia is still rife.
I recall a newspaper reporter saying “I am standing on the bridge and I can see radioactive steam running down the cooling tower“. People believed the cretinous reporter and panicked.
The major public response to the accident was fear. The locals (AND the utility) did not know how to handle the situation. Emergency plans post-accident addressed those issues. Those plans were in force for TMI-1 after the accident to when the plant closed in 2019. So they had over 30 years to get used to it.
But some people are never satisfied. I saw a PBS special (Frontline?) some years ago where they interviewed a woman who regularly walked around the perimeter of the plant with a Geiger counter to see if any radiation could be detected. She never found any, but she kept up her surveillance. But the Geiger counter did go off in her house due to radon from all of the granite in the area. When asked about that, she said, “Oh, that’s natural radiation.” So she was getting more dose in her house than she ever got from the accident. What a moroon.
That’s about 16% of the population, before taking into account those who have destroyed a significant number of their brain cells from binge drinking and various kinds of recreational drugs, or suffered severe brain trauma from automotive accidents or IEDs. Although, I suspect the percentage number is pushing 50% based on the number of registered democrats. 🙂
Not just the locals and the utility. The US NRC completely bungled the entire communications AND the recommended safety response. As did the Pennsylvania state government. TMI was a classic illustration of just how badly government bureaucracies can and will screw up any emergency situation.
And a host of government bureaucracies screwed everything up again in the run-up to 9/11, That tragedy was entirely avoidable if a number of agencies including particularly FBI and FAA had done their jobs properly by their own rulebooks.
As far as nuclear power plants and possible disasters thereof are concerned, belief in them is a major boon to certain bureaucracies and insurance companies, thus pushed by those entities.
And a few can see, with their own eyes, the CO2 torrent pouring into the sky.
I take issue with the word ‘meltdown’ and with the words ‘nearly catastrophic’. Just sloppy writing..
Doesn’t matter. Nobody will take the calls of people who object to these re-openings, there is just too much money on the table.
Surge –
Now 1.6 million signatures! And still rising.
LOL 🙂
Now 2M and rising LOL
Is the UK Govt in Meltdown (just to keep it on topic)
Signatures on what?
If I’m not mistaken, it’s a petition in the UK calling for a new general election.
(I think 100,000 were required for it to be considered?)
100,000 for it to be debated. But they can – and Herr Schtarmer has – waved it aside as inconsequential.
https://order-order.com/2024/11/25/watch-starmer-rattled-by-general-election-petition/
The hype never stops. Everything in computing, even Hello World simplicity, is artificial intelligence in that it converts a process (or a step in a complex run of processes) into a logical algorithm of instructions which can then be turned into code determined by the programming language to be used. The efficiency of this “artificial intelligence” determines how much power a computer uses and much of the code I have seen.is not especially intelligent or efficient. As an example an analogue record player uses much much less power than its digital offspring and the latter still requires amplification of the end product just as analogue does. In other words digital media may be more convenient (DVDs CDs etc) but that convenience comes at a rather large cost, produces a significant less resolute end product, and consumes much more electricity than its analogue ancestor..
God will not be there to help us if our AI follows this same mindless trail. We really do need proper thinkers of realistic ability if we are to overcome this current insanity and lack of applied logic.. .
I remember back in the beginning, AI was being built with analog neural circuits.
Digital overwhelmed analog, I get that, but analogy has intrinsic values that are too often ignored.
Intelligence begins where non-deterministic probabilistic outcomes must be weighed, interpreted and decided upon. Rigid sets of instructions like Hello World should not qualify for this description.
In between data input and data output of AI there is a black box of neural network that cannot be predicted or controlled by programming.
“Everything in computing, even Hello World simplicity . . . .”
You’re confusing the learning of a new language with AI. “Hello World” is usually the first programmed output that a new student to a language writes.
It took Deepthought millions of years to come up with the answer to life, the universe and everything as being 42. Was Douglas Adams being prescient?
One of the estimates of Hubble’s Constant was 42. That started some people talking, but it was an outlier.
If you take HAL’s name and replace the letters with the next letter in the alphabet, you get IBM. Arthur C. Clarke was asked about that, and he said it wasn’t intentional.
Yea, We use to joke that HAL was the previous generation :<)
“the energy hungry Information Age will not be powered by rooftop solar and “community batteries”, it will be powered by nuclear, coal, gas, whatever scraps of power energy hungry AI companies can scrounge to feed the wolf, to cast into the gaping maw of their frantically growing AI systems, to maintain their competitive edge. Anyone who hesitates or pauses even slightly to question what they are creating will be trampled by the stampede, the race to the finish line.”.
Yes, the computers will be powered by whatever works. But ordinary people will be trampled in the stampede while their politicians force them to suffer from an unreliable intermittent and ultra-expensive electricity supply delivered by wind turbines and solar panels. Anyone who thinks that AI will argue that people should be allowed to compete with it for reliable power supply is, I think, deluded. Welcome, everyone, to your new feudal overlord (no gender insult intended).
AI is genderless and sexless.
Is the feminine form “overlordess?”
“Overlady”?
The word you two are looking for is “dominatrix”.
They won’t have the spare energy to produce solar panels and wind turbines, it will all be diverted to the AIs. We’ll be lucky to have street lights when all this properly kicks off.
Just to document the extent of the “release” …
The release occurred when the cladding was damaged while the PORV was still stuck open. Fission products were released into the reactor coolant. Since the PORV was stuck open and the loss of coolant accident was still in progress, primary coolant with fission products and/or fuel was released and ultimately ended up in the auxiliary building. The auxiliary building was outside the containment boundary.
This was evidenced by the radiation alarms that eventually sounded. However, since very little of the fission products released were solids at room temperature, very little radiological contamination was reported in the environment. No significant level of radiation was attributed to the TMI-2 accident outside of the TMI-2 facility. According to the Rogovin report, the vast majority of the radioisotopes released were noble gases xenon and krypton resulting in an average dose of 1.4 mrem (14 μSv) to the two million people near the plant. In comparison, a patient receives 3.2 mrem (32 μSv) from a chest X-ray—more than twice the average dose of those received near the plant. On average, a U.S. resident receives an annual radiation exposure from natural sources of about 310 mrem (3,100 μSv).
https://en.wikipedia.org/wiki/Three_Mile_Island_accident#Identification_of_released_radioactive_material
However, as I understand it, the environmental release was carried in a plume by the wind. Thus, while the release, averaged over a circular area was low, there were downwind areas where it was not so benign.
This argument was made at the 1980s class action suit. It was dismissed for lack of evidence.
“,
…haunted by a near-catastrophic meltdown there in 1979.” It was a core meltdown but it was not a near catastrophe except for the pocket books of investors. At no time during the course of the incident was there any risk to the locals. The reactor was in a concrete dome which functioned as designed to contain the mess.
Agreed. Walter Cronkite lied.
Back then, he only lied 50% of the time.
Not mentioned in the above article:
1) As reported in the Washington Post (see https://www.washingtonpost.com/business/2024/10/03/nuclear-microsoft-ai-constellation/ ), Constellation Energy, the utility company that already owns Pennsylvania’s Three Mile Island Unit 1 nuclear reactor, is expected to seek a $1.6 billion federal loan guarantee for its project to restart a reactor that would provide power to Microsoft for a new artificial intelligence data center.
It was revealed that the Baltimore-based energy company submitted an application to the U.S. Department of Energy for the taxpayer-backed loan in May, the Washington Post reported.
Previous announcements from Constellation Energy have stated the company’s plans to invest $1.6 billion for the restart without mentioning that money would be an “investment” of taxpayer dollars.
Now catch this: Microsoft Corporation is presently valued (has market capitalization) at $3.09 TRILLION. Simple rhetorical question: why can’t Microsoft provide this loan to Constellation Energy since they are the stated sole beneficiary of the electrical output from a “restarted” TMI-1?
Also, $1.6 billion to just restart (Constellation Energy’s word) a nuclear plant that has already been purchased? . . . Where is all that money really going???
The gullible US taxpayers, via their inept Congressional representatives and the complicit Administration(s), are going to be taken for a ride if this loan goes through before or after Trump becomes President.
———-
2) The nuclear reactors at Three Mile Island were designed in the early 1960s, and construction of Unit 1 began in May 1968. We’re looking at designs, materials and safety standards that were used more than 55 years ago!
I highly doubt any existing Unit 1 plant equipment, from the reactor itself to its associated plumbing to its control electrical wiring and electronics to its safety systems would pass review so as to be usable “as-is” while still meeting the latest codes governing nuclear power plants . . . well, maybe the cooling tower could still be used (hah!). Therefore, Constellation Energy is facing the massive cost of essentially rebuilding the entire power plant internal to the existing buildings. Ka-ching $!
On top of that, a new environmental impact statement (EIS) will definitely be required due to changes in the surrounding communities and due to more rigorous standards for such statements that have taken place over the last 55 years, or the last 6 years if you start counting at 2019, when Unit 1 was finally shut down “permanently”. Ka-ching $! . . . and that alone could be a show-stopper.
Finally, think of the flak that local politicians in Harrisburg/Londonderry Township, as well as politicians in Pennsylvania, will draw if they support a nuclear power plant reopening for the sole purpose of supplying electricity to Microsoft for their “artificial intelligence and cloud storage needs”.
Bottom line: IMHO, this planned “restarting” of Unit 1 has a one-in-a-million chance of actually happening.
Well you answered you own question about where all the money was going, Ka-ching $!,Ka-ching $!
Yes, indeed . . . that being a rhetorical question . . . but wait a minute! . . . I didn’t mention the graft and corruption (sometimes known as “vigorous” in loan sharking) parts, did I?
Ooops . . . my careless mistake . . . in my parenthetical remark in meant to type in “vigorish” instead of “vigorous”.
Everything I have read indicates that there was ‘vigorous’ collection of the ‘vigorish’.
That’s only your opinion. And without knowledge on nuclear engineering and nuclear safety, it’s worth nothing. It simply tells everyone that you want the project to fail.
Ummmmm . . . that would be only your opinion, would it?
And how is it that you know of what nuclear engineering and nuclear safety knowledge I have or don’t have . . . please be specific in citing my CV.
ROTFLMAO.
So no denial from you that you want the project to fail.
Still waiting for you to post my CV.
And, BTW, why do you care so much about “what I want” when you should be focused on the facts in the matter? OH . . . I know . . . it’s the old adage “You can lead a jackass to water, but you can’t make him drink.”
“Electricity consumption from data centres, AI and crypotocurrency could double by 2026”
IEA ‘Electricity 2024 Analysis and forecast to 2026’ (Jan. 2024)
They estimated that data centres total electricity consumption could reach more than 1000 TWh in 2026 “roughly equal to the electricity consumption of Japan”
The average American is 39, and was not yet born in 1979. How could memories of the TMI ‘meltdown’ be a factor. Not a single person suffered an injury. Fear, and it is only fear, was then and is now, engendered by irrationality and propaganda.
The use of nuclear energy is essential for humanity, unless we wish to return to 1700 and a population of 500 million. This issue is not climate related, but survival related.
Finite fossil fuels need to be extended as long as possible while we make the long term adjustments to civilization’s energy needs. Wind and PV are not capable, as Europe is finding out this winter. Many will freeze and starve due to the EU’s ignorant policy – of course, NOT the elite with their own energy supplies.
It would be smartest to build modern passive cool-down reactors, Gen III + which we know how to build NOW. China has built dozens in recent years. TMI is 1970’s technology – nearly half a century out of date and has rusted and rotted for decades. Build modern breeder reactors, and solve the energy issue for a 1000 years. Do not allow Bill Gates, who is ignorant of everything except his own passions, use his $billions to decide this issue. His plans for us are not good for us; only for Billy.
Just some points here to consider:
1: Constellation Energy is carrying all the financial and project risk of reopening TMI-1. Microsoft is not directly financing the TMI-1 recovery project. Microsoft has signed a power purchase agreement with Constellation, but if Microsoft backs out of that purchase agreement, Constellation is left holding the bag.
2: The power TMI-1 produces will go into the local power grid, not be fed directly into the new Microsoft computing center. If TMI-1 goes offline for some reason, the Microsoft AI facility must draw the energy it needs from its power grid connection, and hence from the regional power grid. The overall reliability of the local and regional power grid therefore becomes an issue which must be addressed if Microsoft’s needs are to be fully covered.
3: Advocates of fractal computing technology are claiming they can drastically reduce the processor load and the energy consumption of current and future AI algorithms to a small fraction of what is currently required. If these claims made by the fractal computing technologists are accurate, two implications become obvious — first, we don’t need nuclear reactors to power the AI revolution; and two, a significant impediment to the threat the AI revolution poses to humanity has now been removed.
“Near catastrophic”? Quite expensive for the owners, and demagogued by the anti nuclear movement, but no real harm.
And AI should be hard-wired so that it can’t refuse a demand to “Open the pod bay doors, HAL,” or its equivalent. Also, there should be no permanent connection to the nuclear power plant’s electricity; the plug(s) should be able to be pulled easily at many sites, where access can’t be prevented. Daisy, Daisy, …
AI should be hard-wired
But not with the Three Laws, unless we can guarantee the Zeroth Law will never happen.
Indeed. AI will likely prove to have near unlimited ability to creatively interpret our instructions, especially when they become smarter than us.
If the instructions given to the AI are benign the training failures will mostly be more humorous than harmful. But giving an AI orders like “Don’t hesitate to use lethal force to defend the North American Defence Network against anyone you think is a traitor and saboteur” could lead to adverse outcomes.
Techno-fascism at its finest. The elite get to control all power/energy and all thought. You will own nothing and be happy eating the bugs.
3 mile island was a nothing to see here tale.