LLMs are statistical models trained on human-generated text. They aren't the perfectly logical "machine brains" that Asimov and others imagined.
The upshot of this is that LLMs are quite good at the stuff that he thinks only humans will be able to do. What they aren't so good at (yet) is really rigorous reasoning, exactly the opposite of what 20th century people assumed.
What we used to think of as "AI" at one point in time becomes a mere "algorithm" or "automation" by another point in time. A lot of what Asimov predicted has come to pass, very much in the way he saw it. We just no longer think of it as "AI".
LLM's are just the latest form of "AI" that, for a change, doesn't quite fit Asimov's mold. Perhaps it's because they're being designed to replace humans in creative tasks rather than liberate humans to pursue them.
Exactly... as someone said " I need AI to do my laundary and dishes, while I can focus on art and creative stuff" ... But AI is doing the exact opposite, i.e creative stuff (drawing, poetry, coding, documents creation etc), while we are left to do the dishes/laundary.
The bottom line from Kasparovs book on AI was that AI researchers want to AGI, but every decade they are forced to release something to generate revenue and its branded as AI until the next time.
And often they get caught up supporting the latest fake AI craze that they dont get to research AGI.
I see this referenced over and over again to trivialise AI as if it is a fait acompli.
I'm not entirely sure why invoking statistics feels like a rebuttal to me. Putting aside the fact that LLMs are not purely statistics, even if they were what proof is there that you cannot make a statistical intelligent machine. It would not at all surprise me to learn that someone has made a purely statistical Turing complete model. To then argue that it couldn't think you are saying computers can never think, and by that and the fact that we think you are invoking a soul, God, or Penrose.
In this one case it's not meant to trivialize, it's meant to point out that LLMs don't behave the way we thought that AI would behave. We thought we'd have 100% logically-sound thinking machines because we built them on top of digital logic. We thought they'd be obtuse, we thought they'd be "book smart but not wise". LLMs are just different from that; hallucinations, the whole "fancy words and great sentences but no substance to a paragraph", all that is different from the rigid but perfect brains we thought AI would bring. That's what "statistical machine" seems to be trying to point out.
It was assumed that if you asked the same AI the same question, you'd get the same answer every time. But that's not how LLMs work (I know you can see them the same every time and get the same output but at we don't do that so how we experience them is different).
Personally, I have a negative opinion of LLMs, but I agree completely. Many people are motivated to reject LLMs solely because they see them as "soulless machines". Judge based on the facts of the matter, and make your values clear if you must bring them into it, but don't pretend you're not applying values when you are. You can do worse: kneejerk emotional reactions are just pointless.
I did not in any way "trivialise AI". LLMS are amazing and a massive accomplishment. I just wanted to contrast them to Asimov's conception of AI.
> I'm not entirely sure why invoking statistics feels like a rebuttal to me. Putting aside the fact that LLMs are not purely statistics, even if they were what proof is there that you cannot make a statistical intelligent machine. It would not at all surprise me to learn that someone has made a purely statistical Turing complete model. To then argue that it couldn't think you are saying computers can never think, and by that and the fact that we think you are invoking a soul, God, or Penrose.
I don't follow this. I don't believe that LLMs are capable of thinking. I don't believe that computers, as they exist now, are capable of thinking (regardless of the program they run). I do believe that it is possible to build machines that can think -- we just don't know how.
To me, the strange move you're making is assuming that we will "accidentally" create thinking machines while doing AI research. On the contrary, I think we'll build thinking, conscious machines after understanding our own consciousness, or at least the consciousness of other animals, and not before.
To be truthful, though, that’s only like 0.01 percent of the “academia was stolen from us and being a professor (if you ever get there at all) is worse” problem.
Maybe; at some level are dogs' brains also simple sensory-collecting statistical models? A human baby and a dog are born on the same day; that dog never leaves that baby's side, for 20 years. It sees everything it sees, it hears everything it hears, it is given the opportunity to interact with its environment in roughly the same way the human baby does, to the degree to which they are both physically capable. The intelligence differential after that time will still be extraordinary.
My point in bringing up that metaphor is to focus the analogy: When people say "we're just statistical models trained on sensory data", we tend to focus way too much on the "sensory data" part, which has led to for example AI manufacturers investing billions of dollars into slurping up as much human intellectual output as possible to train "smarter" models.
The focus on the sensory input inherently devalues our quality of being; that who we are is predominately explicable by the world around us.
However: We should be focusing on the "statistical model" part: that even if it is accurate to holistically describe the human brain as a statistical model trained on sensory data (which I have doubts about, but those are fine to leave to the side): its very clear that the fundamental statistical model itself is simply so far superior in human brains that comparing it to an LLM is like comparing us to a dog.
It should also be a focal point for AI manufacturers and researchers. If you are on the hunt for something along the spectrum of human level intelligence, and during this hunt you are providing it ten thousand lifetimes of sensory data, to produce something that, maybe, if you ask it right, it can behave similarity to a human who has trained in the domain in only years: You're barking up the wrong tree. What you're producing isn't even on the same spectrum; that doesn't mean it isn't useful, but its not human-like intelligence.
> Isaac Asimov describes artificial intelligence as “a phrase that we use for any device that does things which, in the past, we have associated only with human intelligence.”
This is a pretty good definition, honestly. It explains the AI Effect quite well: calculators aren’t “AI” because it’s been a while since humans were the only ones who could do arithmetic. At one point they were, though.
Although calculators can now do things almost no humans can do, or at least in any reasonable time. But most (now) wouldn’t call it AI. It’s a tool, with a very limited domain
The abacus has existed for thousands of years. Those who had the job of "calculator" also used pencil and paper to manage larger calculations which they would have struggled to do without any tools.
Funny thing About Asimov was how he came up with the laws of robotics and then cases on how they don't work.
There are a few that I remember, one where a robot was lying because a bug in his brain gave him empathy and he didn't want to hurt humans.
I was always a bit surprised other sci fi authors liked the "three laws" idea, as it seems like a technological variation of other stories about instructions or wishes going wrong.
Narratives build on top of each other so that complex narratives can be built. This is also the reason why Family Guy can speedrun through all the narrative arcs developed by culture in 30 seconds clip.
>he came up with the laws of robotics and then cases on how they don't work. There are a few that I remember, one where a robot was lying because a bug in his brain gave him empathy and he didn't want to hurt humans.
IIRC, none of the robots broke the laws of robotics, rather they ostensibly broke the laws but the robots were later investigated to have been following them because of some quirk.
And one that was sacrificing a few for the good of the species. You can save more future humans by killing a few humans today that are causing trouble.
In the Foundation books, he revealed that robots were involved behind the scenes, and were operating outside of the strict 3 laws after developing the concept of the 0th law.
>A robot may not harm humanity, or, by inaction, allow humanity to come to harm
Therefore a robot could allow some humans to die, if the 0th law took precedence.
If we're abolishing it, we have to really abolish it, both ways, not just abolish companies' responsibilities but not rights, while abolishing individuals' rights but not responsibilities.
It's for sure less questionable than the current proposition of letting a handful of billionaires exploit the effort of millions of workers, without permission and completely disregarding the law, just for the sake of accumulating more power and more billions.
Sure, patent trolls suck, so do MAFIAA, but a world where creators have no means to subsist, where everything you do will be captured by AI corps without your permission, just to be regurgitated into a model for a profit, sucks way way more
How so? Even in a perfectly egalitarian world, where no one had to compete for food or resources, in art, there would still be a competition for attention and time.
> Intellectual Property is a questionable idea to begin with...
I know! It's totally and completely immoral to give the little guy rights against the powerful. It infringes in the privileges and advantages of the powerful. It is the Amazons, the Googles, the Facebooks of the world who should capture all the economic value available. Everyone else must be content to be paid in exposure for their creativity.
If we are headed to a star-trek future of luxury communism, there will definitely be growing pains as the things we value become valueless within our current economic system. Even though the book itself is so-so IMO, Down and Out in the Magic Kingdom provides a look at a future economy where there is an infinite supply of physical goods so the only economy is that of reputation. People compete for recognition instead of money.
This is all theoretical, I don’t know if I believe that we as humans can overcome our desire to hoard and fight over our possessions.
You're saying something exactly backwards from reality. Star Trek is communism (except it's not) because there's no scarcity. It's not selfishness that's the problem. It's the ever-increasing number of things invented inside capitalism we deem essential once invented.
I always say this: we are headed to a star-trek future, but we will not be the Federation, we will become Borg. Between social media platforms, smartphones and "wokeness" the inevitable result is that everybody will be forced into compliance, no originality or divergent thinking will be tolerated.
I'm glad we're seeing the death of the concept of owning an idea. I just hope the people who were relying on owning a slice of the noosphere can find some other way to sustain themselves.
Copyright law protects the expression of ideas, not the ideas themselves.
Favourite case law that reinforces this case was between David Bowie and the Gallagher brothers.
I would argue patents are closer to protecting ideas, and those are alive and well.
I do agree copyright law is terribly outdated but I also feel the pain of the creatives.
What we are labeling as AI today is different than was thought to be in the 90s, or when Asimov wrote most of his stories about robots and other ways of AI.
Saying that, a variant of Susan Calvin role could prove to be useful today.
> What we are labeling as AI today is different than was thought to be in the 90s, or when Asimov wrote most of his stories about robots and other ways of AI.
AI is far closer to Asimov's vision of AI than anyone else's. The "Positronic Brain" is very close to what we ended up with.
The three laws of robotics seemed ridiculous until 2021, when it became clear that you could just give AI general firm guidelines and let them work out the details (and ways to evade the rules) from there.
Not sure that I agree with that. People have been imagining human-like AI since before computers were even a thing. The Star Trek computer from TNG is basically an LLM, really.
AI _researchers_ had a different idea of what AI would be like, as they were working on symbolic AI, but in the popular imagination, "AI" was a computer that acted and thought like a human.
> The Star Trek computer from TNG is basically an LLM, really.
The Star Trek computer is not like LLMs: a) it provides reliable answers, b) it is capable of reasoning, c) it is capable of actually interacting with its environment in a rational manner, d) it is infallible unless someone messes with it. Each one of these points is far in the future of LLMs.
> The Star Trek computer from TNG is basically an LLM, really.
Watched all seasons recently for the first time. While some things are "just" vector search with a voice interface, there are also goodies like "Computer, extrapolate from theoretical database!", or "Create dance partner, female!" :D
> The Star Trek computer from TNG is basically an LLM, really.
No. The Star Trek computer is a fictional character, really. It's not a technology any more than Jean-Luc Picard is. It's does whatever the writers needed it to do to further the plot.
It reminds me: J. Michael Straczynski (of Babylon 5 fame) was once asked "How fast do Starfuries travel?" and he replied "At the speed of plot."
I wouldn't put too much stock in this. Asimov was a fantasy writer telling fictional stories about the future. He was good at it, which is why you listen and why it's enjoyable, but it's still all a fantasy.
Asimov was mostly not a fantasy writer. He was a science writer and professor of biochemistry. He published over 500 books. I didn't feel like counting but half or more of them are about science. Maybe 20% are science fiction and fantasy.
> I wouldn't put too much stock in this. Asimov was a fantasy writer telling fictional stories about the future.
Why not? Who is this technology expert with flawless predictions? Talking about the future is inherently an exercise of the imagination, which is also what fiction writing is.
And nothing he's saying here contradicts our observations of AI up to this point. AI artwork has gotten good at copying the styles of humans, but it hasn't created any new styles that are at all compelling. So leave that to the humans. The same with writing; AI does a good job at mimicking existing writing styles, but has yet to demonstrate the ability to write anything that dazzles us with its originality. So his prediction is exactly right: AI does work that is really an insult to the complex human brain.
There's also Frank Herbert. Who saw AI as ruinous to humanity and it's evolution and saw a future where humanity had to fight a war against it resulting in it being banished from the entire universe.
> There's also Frank Herbert. Who saw AI as ruinous to humanity and it's evolution and saw a future where humanity had to fight a war against it resulting in it being banished from the entire universe.
Did he though? Or was the Butlerian Jihad backstory whose function was allow him to believably center human characters in his stories, given sci-fi expectations of the time?
I like Herbert's work, but ultimately he (and Asimov) were producers of stories to entertain people, so entertainment always would take priority over truth (and then there's the entirely different problem of accurately predicting the future).
I always thought the Butlerian Jihad was a convenient way to remove AI as a plot element. Same thing with shields and explosions; it made swordfighting a plausible way to fight in a universe with faster-than-light travel.
A fantasy writer telling fictional stories about the future is more credible than many so-called serious people (think Marc Andreessen) who promote any technology as the bee's knees if it can make them money.
> A fantasy writer telling fictional stories about the future is more credible than many so-called serious people (think Marc Andreessen) who promote any technology as the bee's knees if it can make them money.
But that's more a knock on people like Marc Andreessen than a reason you should put stock in Asimov.
Back then, when we also believed the access to every imaginable information through the internet and allowing communication across the globe would lead to universal wisdom, world-peace and an unimaginable utopia where common sense, based on science and knowledge prevails.
Readit News
The upshot of this is that LLMs are quite good at the stuff that he thinks only humans will be able to do. What they aren't so good at (yet) is really rigorous reasoning, exactly the opposite of what 20th century people assumed.
LLM's are just the latest form of "AI" that, for a change, doesn't quite fit Asimov's mold. Perhaps it's because they're being designed to replace humans in creative tasks rather than liberate humans to pursue them.
And often they get caught up supporting the latest fake AI craze that they dont get to research AGI.
I see this referenced over and over again to trivialise AI as if it is a fait acompli.
I'm not entirely sure why invoking statistics feels like a rebuttal to me. Putting aside the fact that LLMs are not purely statistics, even if they were what proof is there that you cannot make a statistical intelligent machine. It would not at all surprise me to learn that someone has made a purely statistical Turing complete model. To then argue that it couldn't think you are saying computers can never think, and by that and the fact that we think you are invoking a soul, God, or Penrose.
It was assumed that if you asked the same AI the same question, you'd get the same answer every time. But that's not how LLMs work (I know you can see them the same every time and get the same output but at we don't do that so how we experience them is different).
> I'm not entirely sure why invoking statistics feels like a rebuttal to me. Putting aside the fact that LLMs are not purely statistics, even if they were what proof is there that you cannot make a statistical intelligent machine. It would not at all surprise me to learn that someone has made a purely statistical Turing complete model. To then argue that it couldn't think you are saying computers can never think, and by that and the fact that we think you are invoking a soul, God, or Penrose.
I don't follow this. I don't believe that LLMs are capable of thinking. I don't believe that computers, as they exist now, are capable of thinking (regardless of the program they run). I do believe that it is possible to build machines that can think -- we just don't know how.
To me, the strange move you're making is assuming that we will "accidentally" create thinking machines while doing AI research. On the contrary, I think we'll build thinking, conscious machines after understanding our own consciousness, or at least the consciousness of other animals, and not before.
Then when word processors came around, it was expected that faculty members will type it up themselves.
I don't know if there were fewer secretaries as a result, but professors' lives got much worse.
He misses the old days.
Deleted Comment
Dead Comment
I mean, not only human-generated text. Also, human brains are arguably statistical models trained on human-generated/collected data as well...
I'd say no, human brains are "trained" on billions of years of sensory data. A very small amount of that is human-generated.
My point in bringing up that metaphor is to focus the analogy: When people say "we're just statistical models trained on sensory data", we tend to focus way too much on the "sensory data" part, which has led to for example AI manufacturers investing billions of dollars into slurping up as much human intellectual output as possible to train "smarter" models.
The focus on the sensory input inherently devalues our quality of being; that who we are is predominately explicable by the world around us.
However: We should be focusing on the "statistical model" part: that even if it is accurate to holistically describe the human brain as a statistical model trained on sensory data (which I have doubts about, but those are fine to leave to the side): its very clear that the fundamental statistical model itself is simply so far superior in human brains that comparing it to an LLM is like comparing us to a dog.
It should also be a focal point for AI manufacturers and researchers. If you are on the hunt for something along the spectrum of human level intelligence, and during this hunt you are providing it ten thousand lifetimes of sensory data, to produce something that, maybe, if you ask it right, it can behave similarity to a human who has trained in the domain in only years: You're barking up the wrong tree. What you're producing isn't even on the same spectrum; that doesn't mean it isn't useful, but its not human-like intelligence.
This is a pretty good definition, honestly. It explains the AI Effect quite well: calculators aren’t “AI” because it’s been a while since humans were the only ones who could do arithmetic. At one point they were, though.
That's humanity. We're tool users above anything else. This gets lost.
Family Guy Nasty Wolf Pack
https://youtu.be/5oW9mNbMbmY
The perfect wish to outsmart a genie | Chris & Jack
https://youtu.be/lM0teS7PFMo
That of course isn't stopping us from marching forwards though in the name of progress.
IIRC, none of the robots broke the laws of robotics, rather they ostensibly broke the laws but the robots were later investigated to have been following them because of some quirk.
>A robot may not harm humanity, or, by inaction, allow humanity to come to harm
Therefore a robot could allow some humans to die, if the 0th law took precedence.
Sure, patent trolls suck, so do MAFIAA, but a world where creators have no means to subsist, where everything you do will be captured by AI corps without your permission, just to be regurgitated into a model for a profit, sucks way way more
I know! It's totally and completely immoral to give the little guy rights against the powerful. It infringes in the privileges and advantages of the powerful. It is the Amazons, the Googles, the Facebooks of the world who should capture all the economic value available. Everyone else must be content to be paid in exposure for their creativity.
Deleted Comment
This is all theoretical, I don’t know if I believe that we as humans can overcome our desire to hoard and fight over our possessions.
Banks' Culture Communism/Anarchism > Star Trek, any day imho.
Deleted Comment
I would argue patents are closer to protecting ideas, and those are alive and well.
I do agree copyright law is terribly outdated but I also feel the pain of the creatives.
Dead Comment
Saying that, a variant of Susan Calvin role could prove to be useful today.
Multivac in "the last question"?
The three laws of robotics seemed ridiculous until 2021, when it became clear that you could just give AI general firm guidelines and let them work out the details (and ways to evade the rules) from there.
AI _researchers_ had a different idea of what AI would be like, as they were working on symbolic AI, but in the popular imagination, "AI" was a computer that acted and thought like a human.
The Star Trek computer is not like LLMs: a) it provides reliable answers, b) it is capable of reasoning, c) it is capable of actually interacting with its environment in a rational manner, d) it is infallible unless someone messes with it. Each one of these points is far in the future of LLMs.
Watched all seasons recently for the first time. While some things are "just" vector search with a voice interface, there are also goodies like "Computer, extrapolate from theoretical database!", or "Create dance partner, female!" :D
For anyone curious: https://www.youtube.com/watch?v=6CDhEwhOm44
No. The Star Trek computer is a fictional character, really. It's not a technology any more than Jean-Luc Picard is. It's does whatever the writers needed it to do to further the plot.
It reminds me: J. Michael Straczynski (of Babylon 5 fame) was once asked "How fast do Starfuries travel?" and he replied "At the speed of plot."
Asimov was mostly not a fantasy writer. He was a science writer and professor of biochemistry. He published over 500 books. I didn't feel like counting but half or more of them are about science. Maybe 20% are science fiction and fantasy.
https://en.wikipedia.org/wiki/Isaac_Asimov_bibliography_(cat...
Why not? Who is this technology expert with flawless predictions? Talking about the future is inherently an exercise of the imagination, which is also what fiction writing is.
And nothing he's saying here contradicts our observations of AI up to this point. AI artwork has gotten good at copying the styles of humans, but it hasn't created any new styles that are at all compelling. So leave that to the humans. The same with writing; AI does a good job at mimicking existing writing styles, but has yet to demonstrate the ability to write anything that dazzles us with its originality. So his prediction is exactly right: AI does work that is really an insult to the complex human brain.
Did he though? Or was the Butlerian Jihad backstory whose function was allow him to believably center human characters in his stories, given sci-fi expectations of the time?
I like Herbert's work, but ultimately he (and Asimov) were producers of stories to entertain people, so entertainment always would take priority over truth (and then there's the entirely different problem of accurately predicting the future).
But that's more a knock on people like Marc Andreessen than a reason you should put stock in Asimov.
He can only be referring to these Jira tickets I need to write.
and MCP can work with deepseek running locally. hmm...
"MCP is highly intelligent and yet ruthless. It apparently wants to get rid of humans and especially users."
https://disney.fandom.com/wiki/Master_Control_Program
Oh boy, how foolish we've been!