> "We don’t know what is going to happen, we have no idea, and people who tell you what is going to happen are just being silly," he adds. "We are at a point in history where something amazing is happening, and it may be amazingly good, and it may be amazingly bad. We can make guesses, but things aren’t going to stay like they are."
I agree with Hinton: We only know that things will change, not how they will change. We can only make guesses.
Anyone claiming to know with certainty is full of baloney.
Your comment would be fine without that first bit.
By narrow measures of outcome, AI synthesises answers which meet needs in questioners. I think intelligence includes aspects of behaviour (such a word) of a system which go beyond simply providing answers. I don't think AI can do this, yet if ever.
That very phrasing belies the problem with the word: There is no consensus on what intelligence is, what a clear test for it would be or whether such a test could even exist. There are only people on the internet with personal theories and opinions.
So when people say AI is not intelligent, my next questions are whether rocks, trees, flies, dogs, dolphins, humans and “all humans” are intelligent. The person will answer yes/no immediately in a tone that makes it sound like what they’re saying must be obvious, and yet their answers frequently do not agree with each other. We do not have a consensus definition of intelligence that can be used to include some things and exclude others.
The fact that there are degrees of intelligence (dogs > flies) isn't that big of an issue, imo. It's the logically night is day argument - just because we can't point to a clear cut off point between these concepts, doesn't mean they aren't distinct concepts. So it follows with intelligence. It doesn't require consensus, just the same way that "is it night now?" doesn't require consensus
If there's one thing I've found never came true for me, it's almost any sentence of substantive opinion about "philosophy" which starts with "I think we'll agree"
And I do think this AI/AGI question is a philosophy question.
I don't know if you'll agree with that.
Whilst your analogy has strong elements of "consensus not required" I am less sure that applies right now, to what we think about AI/AGI. I think consensus is pretty .. important, and also, absent.
I do think issues of AI/AGI are often Philosophical in nature and I was not claiming that consensus is not required. What I'm arguing against is the requirement for rigorous definitions for words that exist in natural language. Language is messy and often ambiguous - especially around AI.
If we want to have rigorous terms to discuss the state of the art - which I believe we do - then we should define terms for that purpose. i.e: AI needs it's own jargon.
At what point does a human become intelligent? Is a 12 cell embryon intelligent? Is a newborn intelligent? Is a 1 year old intelligent?
> It's the logically night is day argument - just because we can't point to a clear cut off point
Um...what? There may be more than one of them, but precise definitions exist for the transitions between day and night. I think that is a very poor analogy to intelligence.
There are not just degrees of intelligence but different kinds. It is easier for us to understand and evaluate intelligence that is more similar ours and it becomes increasingly harder the more alien it becomes.
Given that, I don't see how you could reject that assertion that LLMs have some kind of intelligence.
Asking
Yes, we don't have clear definitions of intelligence, just like we don't for life, and many other fundamental concepts. And yet it's possible to discuss these topics within specific contexts based on a generally and colloquially shared definition. As long as we're willing to talk about this in good faith with the intention to arrive at some interesting conclusions, and not try to "win" an argument.
So, given this, it is safe to assert that we haven't invented artificial intelligence. We have invented something that mimics it very well, which will be useful to us in many domains, but calling this intelligence is a marketing tactic promoted by people who have something to gain from that narrative.
They're useful. They're not intelligent. He invited the reproach.
The conversation (about whether AI is “intelligent”) was already absurd, I’m just pointing it out ;)
The more important conversation is about whether AI is useful, dangerous, and/or worth it. If AI is competent enough at a task to replace a human for 1/10 the cost, it doesn’t really matter if it “has a mortal soul” or “responds to sensory stimuli” or “can modify its weights in real time”, we need to be talking about what that job loss means for society.
That’s my main frustration: that the “is it intelligent” debate devolves into pointless unsettleable philosophical questions and sucks up all the oxygen, and the actual things of consequence go undiscussed.
This is interesting in its own right, and has propelled the computing industry since it was proposed, but it's not a measurement of intelligence. The reality is that we don't have a good measurement of intelligence, and struggle to define it to begin with.
Original proposal:
"I propose to consider the question, "Can machines think?" This should begin with definitions of the meaning of the terms "machine" and "think." The definitions might be framed so as to reflect so far as possible the normal use of the words, but this attitude is dangerous [...] Instead of attempting such a definition I shall replace the question by another, which is closely related to it and is expressed in relatively unambiguous words."
Clearly Turing is saying "we cannot precisely define what thinking means, so let's instead check if we can tell apart a human and a machine when they communicate freely through a terminal". It's not about fooling humans (what would be the point of it?) but about replacing the ambiguous question "can they think" with an operative definition that can be tested unambiguously. What Turing is saying is that a machine that passes the test is "as good as if it were thinking".
> Machines have arguably been able to do this for decades.
Absolutely not and it's surprisingly uninformed to claim so.
Whether AI has a more powerful effect than it's predecessors remains to be seen. It could.
there is a ton more but you asked for one :)
An alternative view could be, this is just the same as every other technological innovation.
I can't but compare his takes with Stuart Russell takes, which are so well grounded, coherent and easily presented. I often revisit Stuart Russell discussion with Steven Pinker on AI for the clarity he brings to the topic.
Hinton published the seminal paper on backpropagation. He also invented Boltzmann machines, unsupervised learning and mixture of ecperts models. He championed machine learning for 20 years even though there was zero funding for it through the 80s and 90s. He was Yann LeCun's PhD adviser. That means Yann LeCun didn't know ass from tea kettle until Hinton introduced him to machine learning.
Know perchance a fellow by the name of Ilya Sutskever? ChatGPT ring any bells? Also a student of Hinton's. The list is very long.
I know the backprop paper. I've read it in the early 2000s. And I remember Hinton as a co-author. Same with Boltzmann machines. Co-author. "Advisor to that great guy", "Teacher of this great guy", "Nobel price together with that guy" <- all of this leads me to the above conclusion. YMMV
Frankly, this all sounds like hero worship and the language is very cringe.
People on Hacker News seems to idolize the lone genius who somehow pulled himself up from his bootstraps. That person does not exist. The truth is that great minds are made, moulded into shape. That the best people behind our AI technology emerged from his lab is no coincidence. Those trash-talking Hinton on this forum are unlikely to achieve 1/100th of what he has accomplished.
“The housecat may mock the tiger,” said the master, “but doing so will not make his purr into a roar.” [1]
[1] http://www.catb.org/esr/writings/unix-koans/end-user.html
Do these historical accolades give him a blank check to be wrong in the present?
"sounds like he's a poster boy who rode on the success of others"
The person who wrote that didn't even bother checking who Hinton was before pulling that sentence out of their ass.
For context: he once argued AI could handle complex tasks but not drawing or music. Then when Stable Diffusion appeared, he flipped to "AI is creative." Now he's saying carpentry will be the last job to be automated, so people should learn that.
The pattern is sweeping, premature claims about what AI can or can't do that don't age well. His economic framing is similarly simplified to the point of being either trivial or misleading.
Let me start by saying a few things that seem obvious. I think if you work as a radiologist, you're like the coyote that’s already over the edge of the cliff but hasn’t yet looked down
It’s just completely obvious that within five years deep learning is going to do better than radiologists.… It might be 10 years, but we’ve got plenty of radiologists already.”
https://www.youtube.com/watch?v=2HMPRXstSvQ
This article has some good perspective:
https://newrepublic.com/article/187203/ai-radiology-geoffrey...
His words were consequential. The late 2010s were filled with articles that professed the end of radiology; I know at least a few people who chose alternative careers because of these predictions.
---
According to US News, radiology is the 7th best paying job in 2025, and the demand is rising:
https://money.usnews.com/careers/best-jobs/rankings/best-pay...
https://radiologybusiness.com/topics/healthcare-management/h...
I asked AI about radiologists in 2025, and it came up with this article:
https://medicushcs.com/resources/the-radiologist-shortage-ad...
The Radiologist Shortage: Rising Demand, Limited Supply, Strategic Response
(Ironically, this article feels spammy to me -- AI is probably being too credulous about what's written on the web!)
---
I read Cade Metz's book about Hinton and the tech transfer from universities to big tech ... I can respect him for persisting in his line of research for 20-30 years, while others saying he was barking up the wrong tree
But maybe this late life vindication led to a chip on his shoulder
The way he phrased this is remarkably confident and arrogant, and not like the behavior of respected scientist (now with a Nobel Prize) ... It's almost like Twitter-speak that made its way into real life, and he's obviously not from the generation that grew up with Twitter
Of course, because you have different people all predicting a different future, some of them are bound to get it right. That doesn't mean the same person will be right again.
1. The medical world doesn't accept new technologies easily. Humans get a much higher pass on bad performance than technology and especially than new technology. Things need to be extensively tested and certified, so adoption is slow.
2. AI is legally very different than a radiologist. The liability structure is completely different, which matters a lot in an environment that deals with life or death decisions.
3. Image analysis is not language analysis and generation. This specific machine learning part is not the bit of machine learning that has advanced enormously in the past two years. General knowledge of the world doesn't help that much when the task is to look at pixels and determine whether it's cancer or not. Now this can be improved by integrating the image analysis with all the other possibly relevant information (case history etc.) and diagnosing the case via that route.
The overwhelming likely thing is that radiologist jobs will change, just like programming jobs will change.
e.g. see my comment on: Did Google and Stack Overflow "replace" programmers?
https://news.ycombinator.com/item?id=43013363
That is, I do not think programmers will be "replaced". The job will just be different; people will come to rely on LLMs for their jobs, like they rely on search engines.
Likewise, you can probably hire fewer doctors now because Google appeared in ~2000, but nobody talked about them being "replaced". There is NOT less demand for doctors.
---
It also reminds me of the prediction around self-driving cars, which is 13+ years ago at this point:
https://news.ycombinator.com/item?id=45149270
I believe Hacker News mostly fell for the hype in ~2012-2016. And even though the predictions turned out to be comically wrong, many people are still attached to them
https://en.wikipedia.org/wiki/List_of_predictions_for_autono...
i.e. I don't think Hinton will be proven "right" with ANY amount of time. The whole framing is just off.
It's not humans xor AI, it's humans + AI. And the world is not static
Predictions about self driving were off, but far from "comically wrong". Waymo's operations are proof of that.
And to conclude things based on the state of the replacement of programmers after only 2-3 years of ChatGPT being a thing is folly.
The reality is that AI has far fewer limitations and legacy cruft than humans to deal with. Don't get me wrong, I like humans, but our performance is very close to the peak of what it could ever be. That of AI not so much. Remember that AI has been evolving for less than 100 years and it is already where it is today. That took us/biology orders of magnitude more time.
The only real question is how fast it will replace (which) human labor.
Russell is much more measured in his statements and much more policy driven.
In my mind you need to listen to both and try to figure out where they're coming from.
Ugh, Scientism at its best (worst?). Do you also back up Watson statements about race? I'm sure you don't, as that's not part of your training.
Accomplished researchers can say dumb things too, it happens all the time.
Some condensed source I found on the topic:
https://www.ing.com/Newsroom/News/The-more-famous-an-expert-...
With that stated, if you can see beyond his vitriol, Nassim Taleb has some valid comments toward Tetlock's methodologies. I love Taleb, but hate his tendency to try and shock people. However, he does raise valid concerns about fat tails.
Calling it "scientism" to care about these things as a way of dismissing the argument out of hand is anti-intellectualism at its worst.
Those are not arguments, that's scientism.
I upvoted you anyway, as you're somehow trying.
He was basically concerned that a group of rulers were trying to use science to dominate everybody else. I agree with Hayek and his concerns.
The problem is that it has now morphed into a term that can basically mean many things. While a bit of a rat hole, wikipedia does have a great treatment of the term and its journey.
When you use a term like this, which by it's very nature is pejorative, and you use it without regard for its definition, it is not only anti-intellectualism, but also poor communication. It turns into communication that is set up to foster divisions rather than learning.
moralestapia's reply contains a series of fallacies and thinking errors.
1. Straw Man Fallacy and Misrepresentation
moralestapia distorted my statement that Hinton is a well-regarded researcher into the idea that I am suggesting any statement he makes about the AI ramp is correct. This is actually the opposite of what I wrote. Despite Hinton's groundbreaking work in AI, I said he is not good at understanding commercial ramps, as talented people often cannot conceive that others cannot follow their vision.
moralestapia created a classic straw man argument: attacking a misrepresented version of the other’s position.
2. Red Herring and Totally Irrelevant Reference
moralestapia brings in Watson's racist viewpoints that got him ostracized.
Watson’s controversial statements about race have no connection to the topic at hand or to the argument being made and serve only to distract.
3a. Anchoring (Behavioral Economics) and Innuendo Effect
This next point comes from my fascination with Kahneman and Tversky—understanding their framework is incredibly important in how we relate. But more than that, I think they basically pull back the curtain and allow us to see what we do to hide the truth.
The first bit of information we’re exposed to disproportionately influences how we interpret subsequent information or people—that’s anchoring. By introducing Watson’s racist statements in response to my comments, he “anchored” the discussion with something emotionally and morally loaded.
The phrase “Do you also back up Watson's statements about race? I’m sure you don’t, as that’s not part of your training.” superficially acknowledges that I am most likely not defending racist statements. However, just raising the hypothetical (“do you also defend X?”) puts the idea into the discussion.
Research in psychology and behavioral economics confirms that once an accusation or association is mentioned, it becomes part of the mental framework—making “I know you are not” less effective at removing the implication than never introducing it in the first place.
In other words, moralestapia does a great job of saying, "look, there seems to be somebody who could possibly support racist thought," by specifically saying, "I know you don't have this viewpoint."
3b. Guilt by Association and Ad Hominem (Implied)
This is slightly more subtle: if we unpack it, moralestapia is suggesting that if I did support Hinton's commercial ramp (which I don't), then somehow I must also accept all of another person’s viewpoints. As referenced above, moralestapia introduces the notion that I may (although he says he doubts it) be a racist.
4. Hasty Generalization
moralestapia's blanket statement “Accomplished researchers can say dumb things too, it happens all the time” is true in a trivial sense, but it’s being used to generalize and dismiss Hinton’s credibility out of hand, regardless of the context or specifics. This is a generalization that bypasses engagement with the actual substance of Hinton’s expertise or my original point.
I’m a bit less picky here because you could argue that this was exactly my point: although Hinton is clearly brilliant, he can't always appreciate that others may not be able to follow or implement his vision.
But specifically, moralestapia doesn't seem to understand or appreciate Hinton. I’ll try to explain how to listen to a brilliant mind and not dismiss him with statements like "can say dumb things too." I think moralestapia is cutting himself off from an enormous amount of learning by making such dismissals or by accusing others of “Scientism.” In this case, I would also submit that moralestapia does not understand the historic context of how this was derived by Hayek. As somebody that leans heavily toward Austrian School of Economics and Hayek, it's funny of being accused of a term that I would say that I agree with as per Hayek's definition.
So, let’s discuss Hinton.
It’s already been cited in this thread that Hinton has lost credibility because he predicted in 2016 that radiologists would be replaced.
But I’d hope that most people can recognize that Hinton is very capable of making a technical judgment, just not necessarily a marketing or commercial one. That’s the root of my comment: listen to him, but understand his background.
From a technical perspective, using a tensor-based approach and all the tools we have today should, in theory, allow us to replace radiologists with AI. There’s nothing technically preventing this if a business case could be made for the investment. From this standpoint, Hinton was—and is—100% correct.
So, why was he so off?
While it’s not a rigorous argument, I’ll just mention that 1 out of every 5 dollars of our GDP is spent on medicine—basically double any other country—and that number is growing faster than inflation or GDP growth.
I’ll propose, without full argument, that a quick review shows the field is flooded with rent-seeking, as defined by Krueger & Tullock.
As an investor in the medical field, you quickly realize that breaking into medicine is a minefield of regulation, bureaucracy, and entrenched interests.
Navigating the labyrinth of government approvals and encountering medical practitioners—who often seem intent on protecting their own positions—can open your eyes to what rent-seeking really looks like. Why do we pay U.S. doctors 200–300% more than Scandinavian doctors? This is not a rigorous argument, but it should highlight that something looks wrong from 50,000 ft.
Penetration into this field is not a technical problem; it’s a political one. Hinton is a talented researcher who finished his career at a massive company called Google, shielded from market realities. That doesn’t mean we should dismiss his ability to perceive the capabilities of the technology. However, to get a valid time frame, we need to process his output.
Finally, do we say that "well I'll just listen to Stuart Russell." While I think you should listen to him, you need to recognize his impact on the technical community is much less. More than this, the specifics of what has redefined AI, which I would like to call a tensor based approach, is NOT were he made his mark in AI. On the flip side of this, Hinton is all about applying Tensors, and is the reason why he is so widely cited and recognized.
I wonder if he's a HN commenter as well, in that case.
I do appreciate your mention of Stuart Russell however. I've recently been watching a few of his talks and have found them very insightful.
The issue is the increasing imbalance of capital being overvalued compared to labor, and how that has a negative impact on most individuals.
A statement like this from someone influential is important to break that narrative, despite the HN crowd finding it obvious.
Inequality has increased but it’s no longer clear that it’s as severe an increase as Piketty and Saez once argued. So, yes, things could certainly be much better. The US could, for example, benefit from a more progressive taxation and a stronger social safety net. But at the same time, we aren’t all headed to hell.
If I make 100 tokens and that buys me 100 food, thats better than making 1000 tokens that buys me 1 food.
https://en.wikipedia.org/wiki/Robot_tax
Make it 50% of the sales price like with cigarettes, since "AI" makes people dumber.
Rich, greedy people ruin everything.
Moreover, most of the rest of the world’s poverty exists so that a few greedy pigs here can be even more wealthy. We have the CIA and the one-party system that controls it to thank for that.
B2C (sell to people)
B2B (sell to B2C companies)
If the “C” is broke, it seems like there won’t by any rich people. In other words, if the masses are poor and jobless, who is sending money to the rich?
What would they need it for? Remember, money is just the accounting of debt. Under the old world model, workers offer a loan to the rich — meaning that they do work for the rich and at some point in the future the rich have to pay that work back with something of equal value [e.g. food, gadgets, etc.].
But if in the new world the rich have AI to do the work they want done, the jobless masses can simply be cut out of the picture. No need to be indebted to them in the first place.
As one individual, you don't really owe anything to anyone. The only time you owe something is in social terms, when you borrow it in your name, or promise reward for work. And even then, people try to get out of paying things back, but in most cases, the courts, the police or the payees themselves get them to do it anyway.
If you own some land, and suddenly, you can get work on it done without giving almost anything in return (except electrical power), you don't owe anything to anyone. And if you can defend that land effectively, you don't physically need anyone else.
This concept of the social contract, where some abstract group of rich owes something to an abstract group of workers, is actually just a series of consequences that happened to a bunch of individuals when debts weren't paid. But if you're rich, the consequences are no longer an issue, and you're not motivated by some other thing (morals or empathy, for example), the social contract breaks down in your favor.
It's a good thing to remind oneself that social contracts don't maintain themselves, we need to maintain them.
The debt to the workers almost never goes unpaid. The workers quickly call the debt to get food and shelter in return.
More often the workers fail to repay their debts to the rich. This is how you get entities like Berkshire Hathaway or Apple sitting on mountains of money. That money is the symbol of the loans that were extended to the workers, with the workers not being able to offer equivalent value in return.
Even among the rich, holding money is unusual, though. They usually like to call the debt for something of real value (e.g. land) as well.
Your original comment said:
> at some point in the future the rich have to pay that work back with something of equal value [e.g. food, gadgets, etc.].
So, I agree with you there. Then, when you said:
> But if in the new world the rich have AI to do the work they want done, the jobless masses can simply be cut out of the picture.
So, if any rich individual gets that equation back in their favor, no one will uphold the social contract (i.e., balance the equation) unless the rest of us do it.
For this, you don't need AGI or robots that can do anything a human can, or a large pool of non-workers. You just need reliable autonomous weapons to coerce most work, or enough automation to get work done when coercion fails; humans can do the rest.
> The workers quickly call the debt to get food and shelter in return.
I'm not as worried about that other debt, where some money becomes worthless, or the state stops enforcing its currency, or people stop accepting money for goods. I'm more worried about the former debt, where someone can renege or just pay much less, because rich and state interests align. So, even if the state takes over with UBI, and you continue working, you'll just have less and less overall, and there'll be no real leverage against it.
Only if it was the employer that originally extended the debt and it is the worker repaying the debt with their labor. Otherwise the employer still has a debt outstanding. The IOUs are distributed, so, practically speaking, that debt may be to another entity (e.g. a bank), but it really makes no difference how many other people get involved. No matter how you slice it, the math is all the same.
does not mean you cannot get rich by some other means
I was really surprised to hear a scientist like him, who knows how the tech works, to go full scare of a Skynet AI.
Most revolutions (bolsheviks, cuba, iran, arab spring etc) have made people significantly poorer, while most innovations have made people significantly richer (railroads, electricity, first and second agricultural revolutions, manufacturing)
I assume it will be technologically possible to run "medium LLM" (for the lac of a better name) on your phone. A medium LLM is something that knows a limited vocabulary (say slightly larger than simple English), and perhaps doesn't remember the capital of France; but it can reason well with the limited vocabulary. So it can answer what's the capital of France by reading Wikipedia, and likewise, it can work with complicated words using their definitions in terms of simpler words.
Now, everybody would run an AI like that on their phone. It would help people solve real world problems of navigating the world and talking to others. Most importantly, it would help people unite by surpassing the Dunbar number. If you (with help of your phone AI) can keep track of 15 million contacts rather than 150, it is life changing and increases trust-building by orders of magnitude. And soon, machines will be able to do that for us.
Socialists have always emphasized education and communication, for everyone, because these are the true constituents of sovereignty and emancipation. We have, I believe, technological means (widely available universal computation and telecommunication) to provide a kind of mind extension for the mind that will surpass our neocortex and will allow everyone to engage with much larger number of people. Lot more human cooperation will result.
I think activists should look into building and embracing such app - a decentralized communication frontend agent, which would let you build trust with large number of people, without much effort, and help you by coordinating with them (really, learning from them their skills and their struggles). We don't need social media giants to do this for us in a centralized way.
So I posit this techno-anarchism as an antidote against techno-feudalism.
> “What’s actually going to happen is rich people are going to use AI to replace workers,” he says. “It’s going to create massive unemployment and a huge rise in profits. It will make a few people much richer and most people poorer. That’s not AI’s fault, that is the capitalist system.”
It's kind curious how that would happen.
In the old days, if you want to maintain a monopoly, you can try drain out the talent pool so no one else can hire the best people to do the work that you're doing, and you can also try patent wall to delay your competitors from launching their product.
But if a worker can be replaced by an AI, it could also mean that the competitiveness of the work is significantly reduced, to the point that theoretically everybody can do it. The only way (i guess) to remain the monopoly is then tightening control on the AI while optimize the process to kill off all potential competitors etc. It's all Red Ocean policies (https://www.wallstreetprep.com/knowledge/red-ocean-strategy/).
"Massive unemployment" maybe, but I don't think "huge rise in profits" is guaranteed.
Back then, you have people, which is hard to duplicate and thus can act as a barrier of entry. But AI is just a program, which can be copied with ease, and runs on maybe expensive but standardized hardware.
Now he is claiming he left to watch Netflix?
bigyabai•5mo ago
pajamasam•5mo ago
[1]: https://www.techtimes.com/articles/240511/20190329/godfather...
bigyabai•5mo ago
I'd prefer a second opinion from someone with credentials that aren't cosmetically related to the source material.