I kept waiting for an Intentional Stance reference that never came. Totally reasonable to avoid explicitly invoking the jargon, and Dennett would be proud either way!
If people are so unwilling to accept basic use of metaphors, I’m going to have to insist that they stop muromorphising their computer pointing peripherals.
Great post, but also, I’m guessing a lot of the people you’re responding to would be skeptical that humans can be reduced to a wave function. And while I think they’re directionally wrong, there is something to be said for the fact that the universe and humans remain mysterious in some fundamental ways. Eg, quantum mechanics doesn’t mesh with general relativity, and neither seem to explain how subjective experiences arise. Of course, none of this justifies thought-terminating metaphors about AI.
Most people are unaware of the circular reasoning inherent in argument by definition and hence believe that they have advanced a real argument that needs a counter with statements like "AI is only a tool".
My pet peeve is the significance that people put on whether AI is conscious or not when this should clearly be seen as irrelevant to any discussion except maybe regarding AI rights.
Nice! AI is not just a tool . But metaphors are not just simplifications as recent philosophers of science such Margaret Morrison, Mary Morgan and Amie Thomasson have suggested.
Agree with the point that quibbling over language is not the most productive part going forward.
But AI people have been anthropomorphizing from the very beginning; indeed, the entire AI project can be described as "build a computational mechanism and then narrate its working in intentional terms."
The computer scientist Drew McDermott once wrote a paper called “Artificial Intelligence Meets Natural Stupidity“ in the 1970s where he argues that many of the feuds over early AI really could have been avoided if the AI researchers had used more technical names for their systems rather than what he calls “wishful mnemonics.” Here's a great quote:
Many instructive examples of wishful mnemonics by AI researchers come to mind once you see the point. Remember GPS? (Ernest and Newell 1969). By now, “GPS” is a colorless term denoting a particularly stupid program to solve puzzles. But it originally meant “General Problem Solver,” which caused everybody a lot of needless excitement and distraction. It should have been called LFGNS--”Local-Feature-Guided Network Searcher.”
I do think some of this careless anthropomorphizing is how you get from talk about safety and complex systems to p-doom.
This reminds me of the tension between Husserl’s phenomenology vs Heidegger describing “things” as characterized by their contexts and how they’re used.
Somehow also reminiscent of social constructivism.
In tech/comms we also have the concept of affordances — that tech/things have the properties based on how they’re used and those become real properties. But as you say that’s only one possible use/property of them, and they’re far from mutually exclusive! (I talk a little about affordances here: https://substack.com/@saraofmagdeline/note/p-174416532?utm_source=notes-share-action&r=9neco)
Good post, and i agree. I think for any metaphorical stance we take towards AI, we should ask ourselves what that stance is doing for us in terms of explaining and predicting its behavior, and what it might be obscuring. I’m not really sure it makes sense to claim there is a fact of the matter wherein one stance is better than another, independent of what we’re trying to accomplish. Different theories are themselves just different tools for giving us purchase on “reality”. (Though I am generally inclined towards pragmatism so it’s an easy sell for me.)
The only thing I’ll say is that I do think it’s also useful / important to try to have some kind of theory linking the more abstract explanatory constructs in question to something a bit more mechanistic. You don’t need to flesh out every explanation in those mechanistic terms, but I sometimes worry about being led to the wrong inference on the basis of abstract constructs or “filler terms” that are doing explanatory work they’re not equipped for. At the same time, I acknowledge this is itself a kind of vague concern.
But yeah, I do fundamentally agree with your take here (and I’m writing up something similar myself).
Yes. And it needs to be perpetually reminded that something is not what we call anthropomorphism, in a sense that it's ascribing human-like properties to a non-human entity which don't actually exist in that entity (be it AI or non-human animals), if those properties are clearly proven to be present. With AI, this dismissal of descriptions usually focuses on their actual documented behavior, and the assertion that a level of intent or consciousness if required for the behaviors to fundamentally exist, or even further, that they must be literal humans for those behaviors to happen. With non-human animals, it usually focuses on their inner experience/sentience, and a denial of biological science when it conflicts with anthropocentric attitudes.
I wish people would stop showing that horrible version of the SM Lagrangian. It's bad on purpose. There's no sensible reason to expand out all those terms.
There are different levels of reality, as we all know. “Fundamental” reality is as much a linguistic or mental categorization as “tool,” or “human.”
Speaking of tools not doing what we expect: thumb, meet hammer.
But I get your point and agree. IIUC, folks get a bit sloppy with labeling and they sometimes let a bunch of cruft around the words (perhaps that’s meaning?) distort or cloud their ability to grok nuance.
I kept waiting for an Intentional Stance reference that never came. Totally reasonable to avoid explicitly invoking the jargon, and Dennett would be proud either way!
I was weighing it but was worried this was jargony enough
Guy who doesn't like ascribing folk psychological concepts to LLMs because he's an eliminative materialist
If people are so unwilling to accept basic use of metaphors, I’m going to have to insist that they stop muromorphising their computer pointing peripherals.
Great post, but also, I’m guessing a lot of the people you’re responding to would be skeptical that humans can be reduced to a wave function. And while I think they’re directionally wrong, there is something to be said for the fact that the universe and humans remain mysterious in some fundamental ways. Eg, quantum mechanics doesn’t mesh with general relativity, and neither seem to explain how subjective experiences arise. Of course, none of this justifies thought-terminating metaphors about AI.
Most people are unaware of the circular reasoning inherent in argument by definition and hence believe that they have advanced a real argument that needs a counter with statements like "AI is only a tool".
My pet peeve is the significance that people put on whether AI is conscious or not when this should clearly be seen as irrelevant to any discussion except maybe regarding AI rights.
As usual, Yudkowsky scooped us all by a couple of decades.
https://www.lesswrong.com/s/SGB7Y5WERh4skwtnb
Nice! AI is not just a tool . But metaphors are not just simplifications as recent philosophers of science such Margaret Morrison, Mary Morgan and Amie Thomasson have suggested.
If only people thought this way about politics
Agree with the point that quibbling over language is not the most productive part going forward.
But AI people have been anthropomorphizing from the very beginning; indeed, the entire AI project can be described as "build a computational mechanism and then narrate its working in intentional terms."
The computer scientist Drew McDermott once wrote a paper called “Artificial Intelligence Meets Natural Stupidity“ in the 1970s where he argues that many of the feuds over early AI really could have been avoided if the AI researchers had used more technical names for their systems rather than what he calls “wishful mnemonics.” Here's a great quote:
Many instructive examples of wishful mnemonics by AI researchers come to mind once you see the point. Remember GPS? (Ernest and Newell 1969). By now, “GPS” is a colorless term denoting a particularly stupid program to solve puzzles. But it originally meant “General Problem Solver,” which caused everybody a lot of needless excitement and distraction. It should have been called LFGNS--”Local-Feature-Guided Network Searcher.”
I do think some of this careless anthropomorphizing is how you get from talk about safety and complex systems to p-doom.
This reminds me of the tension between Husserl’s phenomenology vs Heidegger describing “things” as characterized by their contexts and how they’re used.
Somehow also reminiscent of social constructivism.
In tech/comms we also have the concept of affordances — that tech/things have the properties based on how they’re used and those become real properties. But as you say that’s only one possible use/property of them, and they’re far from mutually exclusive! (I talk a little about affordances here: https://substack.com/@saraofmagdeline/note/p-174416532?utm_source=notes-share-action&r=9neco)
Good post, and i agree. I think for any metaphorical stance we take towards AI, we should ask ourselves what that stance is doing for us in terms of explaining and predicting its behavior, and what it might be obscuring. I’m not really sure it makes sense to claim there is a fact of the matter wherein one stance is better than another, independent of what we’re trying to accomplish. Different theories are themselves just different tools for giving us purchase on “reality”. (Though I am generally inclined towards pragmatism so it’s an easy sell for me.)
The only thing I’ll say is that I do think it’s also useful / important to try to have some kind of theory linking the more abstract explanatory constructs in question to something a bit more mechanistic. You don’t need to flesh out every explanation in those mechanistic terms, but I sometimes worry about being led to the wrong inference on the basis of abstract constructs or “filler terms” that are doing explanatory work they’re not equipped for. At the same time, I acknowledge this is itself a kind of vague concern.
But yeah, I do fundamentally agree with your take here (and I’m writing up something similar myself).
Yes. And it needs to be perpetually reminded that something is not what we call anthropomorphism, in a sense that it's ascribing human-like properties to a non-human entity which don't actually exist in that entity (be it AI or non-human animals), if those properties are clearly proven to be present. With AI, this dismissal of descriptions usually focuses on their actual documented behavior, and the assertion that a level of intent or consciousness if required for the behaviors to fundamentally exist, or even further, that they must be literal humans for those behaviors to happen. With non-human animals, it usually focuses on their inner experience/sentience, and a denial of biological science when it conflicts with anthropocentric attitudes.
I wish people would stop showing that horrible version of the SM Lagrangian. It's bad on purpose. There's no sensible reason to expand out all those terms.
LLMs show that language, fuzzy and all, is vastly superior to capture knowledge than classical AI or physics models.
We can't even fully map the mater and motion of a glass of water. We'll never ever do so.
But with natural language there are a thousand useful ways to talk about that same glass.
Words win out, as LLMs show us.
(I wrote an article about it, see my profile if interested)
I'd argue that "AI is a tool" is a normative claim. A tool can't be accountable, even if it's powerful and unpredictable.
I slip into “just a tool” framing when people start comparing the times we’re living in to Terminator or Robocop.
There are different levels of reality, as we all know. “Fundamental” reality is as much a linguistic or mental categorization as “tool,” or “human.”
Speaking of tools not doing what we expect: thumb, meet hammer.
But I get your point and agree. IIUC, folks get a bit sloppy with labeling and they sometimes let a bunch of cruft around the words (perhaps that’s meaning?) distort or cloud their ability to grok nuance.