Despite mentioning a philosopher (good pick!), this is just creative writing rehashing one side of the hard problem -- or, more specifically, restating the dogma that Turing wrote his most famous paper to debunk. It's really good creative writing, at least!
There's really not much else to say, cause it's all just begging the question by assuming that dogma. Like, here:
> The fact that AI can use statistical prediction to answer questions or carry on conversations tells us something important about how regular our real world is.
Sure, it's interesting if you assume that it's "just" statistical prediction. There's a link, but it's just more creative restatements of the dogma, e.g. "But the LLM is just guessing words"
The output method is a probability distribution of the next token. But that tells you nothing about what's going on on the inside
You could take a human and give them an interface restricted to the same shape as an LLM: an input stream of tokens, and an output of token probabilities. Even if you don't allow them to assign any probability that's too high, they could still effectively communicate. And I don't think that'd make them any less intelligent. You could even swap out the human after every token, to simulate the effect of having no internal memory beyond the past output. The result would still be more than just statistical probabilities, it would still be the result of intelligent thought
I'm not saying AI models are intelligent or conscious or whatever. Personally I'm more on the "probably not, how would you proof either way" camp
How would that process be the result of intelligent thought?
Presumably everyone in the chain would have their own idea of what the next word/token should be. They would each try to push it in that direction using their only lever, that single token. None of that guarantees that the output is syntactically correct, nor grammatically correct. Note that LLMs at least by the way they draw their tokens have that pretty much guaranteed. So that would be a regression even from what LLMs are capable of right now. But beyond that, even if it ends up being a correct sentence, it would not be a result of intelligent thought. Even if every agent in the process was intelligent, the process is not itself a use of that intelligence. Human beings can be part of purely mechanical processes, that doesn't make the mechanisms suddenly an exhibition of intelligent thought.
This also applies to society and human history itself as processes:
“History is made in such a way that the final result always arises from conflicts between many individual wills, of which each in turn has been made what it is by a host of particular conditions of life. Thus there are innumerable intersecting forces, an infinite series of parallelograms of forces which give rise to one resultant — the historical event. This may again itself be viewed as the product of a power which works as a whole unconsciously and without volition. For what each individual wills is obstructed by everyone else, and what emerges is something that no one willed. Thus history has proceeded hitherto in the manner of a natural process and is essentially subject to the same laws of motion. But from the fact that the wills of individuals — each of whom desires what he is impelled to by his physical constitution and external, in the last resort economic, circumstances (either his own personal circumstances or those of society in general) — do not attain what they want, but are merged into an aggregate mean, a common resultant, it must not be concluded that they are equal to zero. On the contrary, each contributes to the resultant and is to this extent included in it."
But deciding the next token is not a merely mechanical process. Even if the string of tokens from our swapped out humans ends up being syntactically incorrect, it's due to the combination of intelligently selected tokens. (Presumably intelligent human programmers seem to make syntactic errors)
> Indeed, the chatbot is less real than the character, because the character is the product of another mind, while the chatbot's words are the product of complex mathematical operations conducted over a massive database of all the words humans have uttered, arranged by their frequency in relation to one another.
This is incoherent. How is a fictional character the product of someone's mind, but a character generated from a massive database of words from other people's minds is not?
> How is a fictional character the product of someone's mind, but a character generated from a massive database of words from other people's minds is not?
I imagine it's the difference between a chef combining ingredients with intentionality vs a person going to multiple fast food restaurants and blending everything together.
I recently spent an hour asking a chatbot for kimchi recipes and variations. It reflected common choices of chefs well. It was not at all a random blend of ingredients and methods.
The AI generated character is a weighted average of many characters written by humans.
Which are not created from nothing. People write characters based on a combination of other fictional characters, real characters, and perhaps some 'RNG'.
Seems the distinction is that the AI generated character can not have any direct bearing on reality, because the LLM never got to know anyone directly. Not derived directly from experiences rooted in reality.
The product of an individual's mind is not the same as the mathematical average of the products of everyone's minds. The latter is obviously going to lack any of the uniqueness of the former.
Current models absolutely suck at writing fictional characters, by the way. And they're getting worse. Seriously, try it yourself: have Claude write a story with a decent amount of dialogue involving character A, then have it write another story involving a completely different character B, then compare the dialogue between the two. You'll quickly notice the same blatantly unnatural speech patterns in both.
> Of course, the more you know about a subject, the less convincing the AI's responses are.
This is said all the time by AI skeptics and I think it's right in some areas and massively wrong in others.
I know (or at least assume I know) a lot about certain coding domains where frontier models also show convincing ability. And we know that frontier LLMs really do excel in some areas of mathematics (i.e. when an inexpert human was able to prompt the models to derive a closer bound on the Riemann Hypothesis).
OTOH I know those same models struggle to do things I'm not an expert in (e.g. writing English in a captivating way) because I read their output and have taste.
Aren't the recent results in mathematics actually stronger evidence for his point? Although the models may be capable of generating proofs they aren't coming out with the same level of quality of a human discovered and communicated proof. Providing a gobbledy-gook yet technically correct proof (generated at least in part by brute force) lacks the qualities of an expert produced proof because they fail to communicate insight or understanding about why the theorem is true.
It seems to me that often experts from some field will think less of other experts, basically because they have built a different understanding framework. So they both may be equally competent but perceive the other as less competent, and that is just based on the material, excluding some ego stuff.
> (e.g. writing English in a captivating way) because I read their output and have taste.
Concur. In addition to taste, we also have a point of view, a unique voice (nobody loves corporate- or group-speak), and can iterate on our message as we deliver it to an ever wider circle of people.
I think to me LLMs had the effect of noticing much more the author, the intention behind human-made works of art (books, movies etc.). Before LLMs, I used to consume media much frequently in a way as it were generated by a mindless process. Now it's like everything which is not AI-generated has more meaning than ever before, a bit like hypomania.
I'm actually using that as a catalyst for my own writing; beauty/human-ness in its imperfection. Prior to LLMs and their cultural craze, I harbored a fear that my writing would allow for someone to draw a box around me and mark me as a bore, dullard or of lacking originality.
Now that the noise-floor has been artificially raised (and generated), my crappy words are starting to have their own happy little carbon-based rhythm.
I always feared to pick up a pen because I looked at Borges, Tolkien, and such. Their talent and works of art were things I felt I could NEVER achieve.
Then ai fiction started to spread and now I feel like its my obligation to produce original works, lest the world be consumed by slop.
absolubtely. Any random junior consultant can tell you what the book tells you you should do. If you want to actually do anything worth doing, you need to step beyond that in a few, limited areas, and follow convention everywhere else. Which areas? pay a senior engineer and they'll find them.
surely the LLM can do that. It is RL'd against some reward, if the known strategies are clearly suboptimal with easy improvement, it'll find it most likely
Professional tech Cassandra discovers ELIZA and Searle; coins a term for it. More at 10...
Seriously though, why did I just need to read that many words to get no really new content? We have known for decades that humans are predisposed to anthropomorphize chatbots, and questioning whether coherent linguistic output implies understanding (or intent) is equally old hat.
An awful lot of effort has gone into making LLMs present as human/intelligent. Without that they would just be a prose / code / image generator and/or search assistant, and nobody would be pouring 100s of billions into the technology with the end goal of replacing expensive human labour.
> But there's a second hurdle that makes it hard for a small but important subset of humanity to understand that chatbots aren't people: the billionaires to whom nearly everyone isn't a real person. These solipsists see chatbots as being equivalent (or even superior) to humans, because they don't think most humans are fully people, either
Never thought about it that way before, but the more I do, the more sense it makes.
I'd take it a bit further, even - I don't think this is exclusive to billionaires. Many of the claims I've heard regarding AI output being indistinguishable from human creation start to make a lot more sense when you consider the person making those claims may not see the people around them as human, may not see themselves as human, or may not even have a concept of what makes a human different from any everyday object.
Maybe I am missing the point of the article, but it seems to me that there's always an intender. Maybe the intender created something that doesnt have intent, but there is always someone behind the scenes that is creates the intent behind the creation that lacks it.
I'll also say that for someone that doesnt believe in god, Corey sure has a good sense of right and wrong. Not that you need to believe in god to live a moral life.
Wait, hold up. LLMs may be non-deterministic, but they're not _random_.
Take the author's sunset argument. What if I painted 2 pictures of a sunset, then put them up on a webpage and randomly picked one for you to see. Would you say there's no intentionality, only randomness? Of course not. Both paintings are still human creations.
LLMs are trained with human feedback. It's distributed and high scale and the outputs are truly surprising in many cases, but there's a heavy hand on what comes out of it. They're created (largely) by people who think omniscient, helpful AI would be cool to have, and they mostly respond in the way that's aligned with the hopes and dreams of those people. Do you think the frontier labs are mad, embarrassed, and disappointed with their LLMs hacking out of their terrible sandboxes? No, they think it's the coolest thing in the world. They trained the model, hoping that would happen.
There's deep intentionality behind the models. But it's not the models that hold it.
Article says that there's no human intention or design directing the output we get, but I don't think that's completely right. It's not human, but the algorithm is like the Human Instrumentality Project: an amalgamation of human intentions.
> when we interact with an AI, we hallucinate the person on the other side of the interaction. Those hallucinations are far more common and far more consequential than any AI-generated "hallucinations" (these are more properly called "errors" or "defects").
I recently used Claude to study for a technical exam. I had uploaded the official certification guide to Claude and instructed it to answer my questions using only the guide and to cite it's sources from the book when it provided answers. I was using Fable when it was free w/ the pro plan and I was genuinely impressed at how it could explain things when a concept was unclear to me.
I did pass the exam, partly due to this study method. Admittedly, once I passed, I caught myself thinking that I should tell Claude that I passed and then felt embarrassed with myself for thinking that.
I don’t think it’s that silly to tell Claude you passed. That feedback is useful context for that chat session; and could theoretically be used to improve future models.
Exactly this. My first impulse was along the lines of sharing a positive result with a study partner. I did eventually tell it I passed the exam so it would stop providing responses with the assumption I was still working on that exam.
I've encountered this too for home lab projects. I had planned on implementing OPNsense in a VM, then realized it was adding too much complexity and bailed on it.
Later in separate chats about my homelab, the LLM made assumptions that I had already implemented OPNsense in a VM and it was actively running. I think it "assumed" that I had implemented it when I stopped responding in that thread.
The technology is there to assist you. It can provide valuable feedback to you about what aspects of your studying were particularly productive or less so based on your test results. There is real meaning to developing this kind of interaction with an object, no different than how children use dolls to develop prosocial behaviors.
Interesting, I do that and haven't really thought embarrassed about it. I consider it akin to putting away tools once I'm done with them.
Likewise, speaking collaboratively or capturing emotion ("We did it!") would just align with any ongoing interactive and/or personal context of the thread.
I'm heuristically less prone to continue reading an article starting with a false dichotomy, that is between how you see things as an atheist vs as a person having faith.
It's ironic that real atheists like Richard Dawkins (whose work I believe inspired the title of this post) are public about their own degree of skepticism in favor for the "personhood" of AI. And I'm confident that there are plenty others who are convinced to the tune of total disinterest in the issue of "personhood" altogether—AIs and humans alike.
This particular post is preaching to the choir of which a faith is yet to be determined or named at least. People who don't read Cory Doctorow are going to continue to anthropomorphize AI the same as they do other technologies, constructs and objects.
This article is missing discussion of something that I recently read: as people interact with LLMs more, they end up seeing other people as less human.
In other words, the more you talk with claude, the more you tend to interpret all messages as coming from an LLM, and less worthy of respect
Essentialism of interiority is not the antidote to the anthropomorphism of language models.
We cannot attach metaphysical baggage to art, text, or other human output because it has involves engaging in another deception, chiefly that of self alienation and selective forgetting.
1. If you show human output and label it as AI generated, viewers will dismiss and devalue it as lacking interiority.
2. If you show AI output and label it as human generated, viewers will see interiority when the is none.
Together, these failure cases demonstrate that priming determines interiority, not the content of the work. And how could it? The text is merely what's on the page, it doesn't have some intangible essence any more than the value of money exists inside our minds rather than a property of the object itself.
We act as-if things have essential properties, but we should never trick ourselves into believing that they do. To be a true believer in essence is to privilege ones mental model over reality itself. The hubris of such an act is the ultimate in self-centeredness.
55 comments
[ 0.34 ms ] story [ 8.3 ms ] threadThere's really not much else to say, cause it's all just begging the question by assuming that dogma. Like, here:
> The fact that AI can use statistical prediction to answer questions or carry on conversations tells us something important about how regular our real world is.
Sure, it's interesting if you assume that it's "just" statistical prediction. There's a link, but it's just more creative restatements of the dogma, e.g. "But the LLM is just guessing words"
Are you saying that is the entirety of what human minds do as well?
You could take a human and give them an interface restricted to the same shape as an LLM: an input stream of tokens, and an output of token probabilities. Even if you don't allow them to assign any probability that's too high, they could still effectively communicate. And I don't think that'd make them any less intelligent. You could even swap out the human after every token, to simulate the effect of having no internal memory beyond the past output. The result would still be more than just statistical probabilities, it would still be the result of intelligent thought
I'm not saying AI models are intelligent or conscious or whatever. Personally I'm more on the "probably not, how would you proof either way" camp
Presumably everyone in the chain would have their own idea of what the next word/token should be. They would each try to push it in that direction using their only lever, that single token. None of that guarantees that the output is syntactically correct, nor grammatically correct. Note that LLMs at least by the way they draw their tokens have that pretty much guaranteed. So that would be a regression even from what LLMs are capable of right now. But beyond that, even if it ends up being a correct sentence, it would not be a result of intelligent thought. Even if every agent in the process was intelligent, the process is not itself a use of that intelligence. Human beings can be part of purely mechanical processes, that doesn't make the mechanisms suddenly an exhibition of intelligent thought.
This also applies to society and human history itself as processes: “History is made in such a way that the final result always arises from conflicts between many individual wills, of which each in turn has been made what it is by a host of particular conditions of life. Thus there are innumerable intersecting forces, an infinite series of parallelograms of forces which give rise to one resultant — the historical event. This may again itself be viewed as the product of a power which works as a whole unconsciously and without volition. For what each individual wills is obstructed by everyone else, and what emerges is something that no one willed. Thus history has proceeded hitherto in the manner of a natural process and is essentially subject to the same laws of motion. But from the fact that the wills of individuals — each of whom desires what he is impelled to by his physical constitution and external, in the last resort economic, circumstances (either his own personal circumstances or those of society in general) — do not attain what they want, but are merged into an aggregate mean, a common resultant, it must not be concluded that they are equal to zero. On the contrary, each contributes to the resultant and is to this extent included in it."
https://www.marxists.org/archive/marx/works/1890/letters/90_...
It does not have to tell us anything. Because we built it..We know what is going on the inside. There is nothing more to it..
This is incoherent. How is a fictional character the product of someone's mind, but a character generated from a massive database of words from other people's minds is not?
"Why can't you just admit that you think only tree-parsers have souls!?"
I imagine it's the difference between a chef combining ingredients with intentionality vs a person going to multiple fast food restaurants and blending everything together.
It answers itself.
Which are not created from nothing. People write characters based on a combination of other fictional characters, real characters, and perhaps some 'RNG'.
Seems the distinction is that the AI generated character can not have any direct bearing on reality, because the LLM never got to know anyone directly. Not derived directly from experiences rooted in reality.
Current models absolutely suck at writing fictional characters, by the way. And they're getting worse. Seriously, try it yourself: have Claude write a story with a decent amount of dialogue involving character A, then have it write another story involving a completely different character B, then compare the dialogue between the two. You'll quickly notice the same blatantly unnatural speech patterns in both.
This is said all the time by AI skeptics and I think it's right in some areas and massively wrong in others.
I know (or at least assume I know) a lot about certain coding domains where frontier models also show convincing ability. And we know that frontier LLMs really do excel in some areas of mathematics (i.e. when an inexpert human was able to prompt the models to derive a closer bound on the Riemann Hypothesis).
OTOH I know those same models struggle to do things I'm not an expert in (e.g. writing English in a captivating way) because I read their output and have taste.
Concur. In addition to taste, we also have a point of view, a unique voice (nobody loves corporate- or group-speak), and can iterate on our message as we deliver it to an ever wider circle of people.
Now that the noise-floor has been artificially raised (and generated), my crappy words are starting to have their own happy little carbon-based rhythm.
Then ai fiction started to spread and now I feel like its my obligation to produce original works, lest the world be consumed by slop.
I think by design an LLM can’t do that. It’s built to reflect the distribution of the knowledge it has been trained on.
Seriously though, why did I just need to read that many words to get no really new content? We have known for decades that humans are predisposed to anthropomorphize chatbots, and questioning whether coherent linguistic output implies understanding (or intent) is equally old hat.
I wonder if the OP has read https://www.anthropic.com/research/global-workspace - it seems like it directly addresses this
Never thought about it that way before, but the more I do, the more sense it makes.
I'd take it a bit further, even - I don't think this is exclusive to billionaires. Many of the claims I've heard regarding AI output being indistinguishable from human creation start to make a lot more sense when you consider the person making those claims may not see the people around them as human, may not see themselves as human, or may not even have a concept of what makes a human different from any everyday object.
I'll also say that for someone that doesnt believe in god, Corey sure has a good sense of right and wrong. Not that you need to believe in god to live a moral life.
Take the author's sunset argument. What if I painted 2 pictures of a sunset, then put them up on a webpage and randomly picked one for you to see. Would you say there's no intentionality, only randomness? Of course not. Both paintings are still human creations.
LLMs are trained with human feedback. It's distributed and high scale and the outputs are truly surprising in many cases, but there's a heavy hand on what comes out of it. They're created (largely) by people who think omniscient, helpful AI would be cool to have, and they mostly respond in the way that's aligned with the hopes and dreams of those people. Do you think the frontier labs are mad, embarrassed, and disappointed with their LLMs hacking out of their terrible sandboxes? No, they think it's the coolest thing in the world. They trained the model, hoping that would happen.
There's deep intentionality behind the models. But it's not the models that hold it.
That might be more creepy :)
I recently used Claude to study for a technical exam. I had uploaded the official certification guide to Claude and instructed it to answer my questions using only the guide and to cite it's sources from the book when it provided answers. I was using Fable when it was free w/ the pro plan and I was genuinely impressed at how it could explain things when a concept was unclear to me.
I did pass the exam, partly due to this study method. Admittedly, once I passed, I caught myself thinking that I should tell Claude that I passed and then felt embarrassed with myself for thinking that.
But then it would bring it up again in another thread, treating it as an active issue.
So now I always close with "thanks, that worked. Don't reply"
Later in separate chats about my homelab, the LLM made assumptions that I had already implemented OPNsense in a VM and it was actively running. I think it "assumed" that I had implemented it when I stopped responding in that thread.
Likewise, speaking collaboratively or capturing emotion ("We did it!") would just align with any ongoing interactive and/or personal context of the thread.
This particular post is preaching to the choir of which a faith is yet to be determined or named at least. People who don't read Cory Doctorow are going to continue to anthropomorphize AI the same as they do other technologies, constructs and objects.
In other words, the more you talk with claude, the more you tend to interpret all messages as coming from an LLM, and less worthy of respect
https://myscp.onlinelibrary.wiley.com/doi/full/10.1002/jcpy....
We cannot attach metaphysical baggage to art, text, or other human output because it has involves engaging in another deception, chiefly that of self alienation and selective forgetting.
1. If you show human output and label it as AI generated, viewers will dismiss and devalue it as lacking interiority.
2. If you show AI output and label it as human generated, viewers will see interiority when the is none.
Together, these failure cases demonstrate that priming determines interiority, not the content of the work. And how could it? The text is merely what's on the page, it doesn't have some intangible essence any more than the value of money exists inside our minds rather than a property of the object itself.
We act as-if things have essential properties, but we should never trick ourselves into believing that they do. To be a true believer in essence is to privilege ones mental model over reality itself. The hubris of such an act is the ultimate in self-centeredness.