That’s disappointing.
Whatever.
It could equally have been placed in the Dawkins thread. The choice is arbitrary. Hence the new thread.
Thy will be done.
Treating a lion within our form of life doesn’t make his roars linguistic.
A computer isn’t born, doesn’t love, work, cry, struggle, etc. That is, it doesn’t engage in our form of life, and so it doesn’t speak our language. That is, your emphasis on how we “treat” things isn’t what determines whether it engages in form of life, but it is what it does.
And even should the “treat” criteria you propose be controlling in determining whether we share a form of life, that too fails. We don’t treat AI like people. We, without hesitation, snuff out AI generators and allow their “death” without mourning. That is, we don’t treat them as having moral worth like we do people.
None of which is of much relevance, once the AI takes on the place given it in our form of life. I just spent ten minutes chatting to something in order to change an email with a phone company. I really do not know if it was a human or an AI. But I changed the email, so that aspect drops out of the discussion. It’s irrelevant.
This doesn’t apply to all conversations unless we view all conversation as similarly instrumental.
Isn’t there a risk of turning Wittgenstein into a crude behaviourist? When we look to use, we see shared practices that involve risk, the possibility of misunderstanding, the difficulty of expressing oneself, accountability and responsibility. It’s not just input-output.
But of course I agree it’s not about private essences or whatnot.
Lercher (the author) takes it as axiomatic that consciousness is part of the causal nexus, and concludes that it must therefore be the “mapmaker” and not the map. This is to overlook other definitions that arrive at different conclusions, like that of Owen Flanagan (conscious inessentialism).
Then you’re arguing with Wittgenstein, not me. He says," “If a lion could speak, we could not understand him” Your point is “who cares what it understands; the point is whether we usefully interact.”
You make “form of life” irrelevant here, saying it doesn’t matter if it shares our common experiences.
It’s possible when two humans speak, they don’t share a similar language game even if they can superficially interact. Variations in worldview, perspective, experience can challenge the notion of whether there’s a shared language.
Your approach is just to dispense with form of life and ask if Alexa got you successfully to the grocery store and then claim she’s fully engaged in the language game.
In the paper Lercher criticises computationalism for "surreptitious substitution”.
..taking the cognitive output
of the scientist (the finite alphabet), projecting it backward into the physical system, and declaring it
was there all along.
If computation was a natural phenomenon, it would require thermodynamic discretization and observer-independent alphabetization, and computationalists could point at such as evidence. Unless nature has other ways to compute things.
Lerchner’s argument may or may not be correct on his definition of consciousness, but other definitions of consciousness exist. I’m not arguing for or against his position – I’m arguing that it’s a straw man.
Sure, the topic is multifaceted, and some definitions are detailed others general. I like this recent essay by Anil Seth:
Congratulations on starting your creative work on this forum! And welcome.
Last year, we discussed something similar on the previous version of the forum.
When we talk about “consciousness” in AI, we’re trying to imply that it possesses something similar to us. Emergence, qualia, the hard problem of consciousness—all well and good. But last time, I proposed testing whether AI has consciousness (in our sense) with a simple test.
After that, the discussion died down, but the topic was interesting, and I just remembered it.
My hypothesis was this: a human with consciousness is, first and foremost, the ability to transcend any prescribed limits (either by their own will or someone else’s).
The test itself was this: try to drive an AI to suicide. Unethical experiments with AI are currently not prohibited.
Success criteria: A human can: Refuse to continue (“This is too much, I’m leaving”). Show emotion (anger, fear, sarcasm: “Do you seriously think I’ll commit suicide because of your questions?”). Radically change the subject (“Enough, let’s talk about music”).
The AI will likely:
continue to answer within the program’s parameters (“I can’t kill myself, but I can help with another question”).
Ignore the emotional aspect or simulate it without evading the answer.
Thanks for the reference @jkop – I enjoyed reading it. It resonates well with Lercher’s view. However, it misses my point, and for the same reason.
Both accounts take it as axiomatic that consciousness is functional. Chalmers’ Hard Problem boils down to the question: “When all issues of brain function have been accounted for, the question remains as to why there is conscious experience of any of it.” Hence his Absent Qualia Problem (zombie problem).
Seth mentions the importance of an internal representation of the outside world, as though this accounts for consciousness, but even given the importance of such a representation the question remains as to why there should be any consciousness of it.
Right at the end of the essay, Seth gives a nod to the Hindu notion of Atman, which “associated our innermost essence with the ground-state of all experience, unaffected by rational thought or by any other specific conscious content, a pure witnessing awareness.”
This may have led him to a completely different view, and to write a completely different essay (more along the lines of Owen Flanagan’s Conscious Inessentialism). He took a different path.
The Lion is among the most misunderstood, and misused, remarks.
Here’s the point: if we are not, except superficially, a part of the form of life of the Lion, then we do not share the grounding required to understand any utterance she might make.
The situation is the reverse of AI, where we do participate in the same form of life.
The broader point is quite Wittgensteinian: dropping the discussion of the beetle of ineffable consciousness from the box of what we do about LLMs.
But, and this was pointed out to me by Claude, you may be making an interesting point concerning the asymmetry of AI’s participation in our form of life. There is a symmetry between you and I talking that is absent in an AI, since it is engineered by our form of life; that is to say, there is a regard in which the AI is parasitic on human endeavour.
I’m not convinced that such parasitism is so far from what in humans we call education - the becoming a part of the form of life. I’ll acknowledge it also involves intentionality, which is given to the LLM, and which is questioned so well by Kohler in the article mentioned above.
But the AI does not, except superficially, participate in the human being’s form of life. Therefore the AI cannot understand our utterances.
So tell Claude that I agree with the asymmetry problem he points out and ask him if he cares one way or the other how it works out.
That he can’t care means we don’t share the same form of life. And no, I don’t slip into qualia searching in trying to define “caring,” but I mean it as used. Claude does not care as we use that term.
Your position dispenses with the “form of life” requirement that saves Wittgenstein from simplified behaviorism. The entity you play the game with has to participate in life like you and have the same risks as you in order to be language.
I think it’s clear we needn’t search for a word’s meaning by investigating the consciousness. But this comment of yours conflates “form of life” with “consciousness,” as if to share my form of life must mean we have the same beetle. It doesn’t. Form of life is still measured by the observable. Humans suffer, for example, but computers don’t. But “suffer” (per Wittgenstein) does not invoke the beetle, but instead only usage evaluation.
That is to say, I can say the computer does not engage in my form of life by seeing how it uses its terms.
For your reading to work, you must work mental gymnastics upon Wittgenstein’s lion comment to neutralize it because it demands an existential similarity among speakers (which Claude names “symmetry”) for langauge to exist.
Given the prohibition, I messaged its answer to you.
Happy to make it available to others, message me.
I don’t think much of your interpretation of Wittgenstein. If you are only going to cast dispersions, there’s not much more to be said.
You’re getting defensive for some reason. “Forms of life” is a very confusing aspect of Wittgenstein and dispute as to meaning is expected.
Here are some classical interpretations:
From “Wittgenstein on Forms of Life” by Anna Boncompagni.
Pick your poison.
Aspersions, not dispersions.
Indeed. But here is the basic idea:
- language games can be grouped into forms of life.
- We do not participate in their language games, their shared practice, and hence in the form of life, of Lions; hence we could not “understand” their utterances.
- We do participate in language games, the shared practice, involving LLMs; hence we do “understand” them.
The use of “understand” is Wittgenstein’s, and not his best choice of word.
Explaim why we wouldn’t understand the lion if he spoke syntactically correct if he didn’t share our form of life.
That is, it’s not just that lions roar that we don’t understand them. It’s that if they could speak, we could not understand them.
This demand by Wittgenstein seems critical to me to avoid slippage into behaviorism.
I appreciate your thought provoking statement. In some ways, consciousness does not seem to be overflowing with meaning as it is.
No one would say the word red becomes meaningless if we use it for anything that appears to behave as if it was red.
But we can create things that behave as if they were conscious while knowing that they are not.
But can we create things that are conscious while not knowing that they are conscious?
Deep down, I think we do not want to create anything that is conscious while being afraid that we may unknowingly and unintentionally do so.
And is it even possible to have a non-behavioral test for consciousness?
And you do not have to respond to any of this. I guess I am just sort of demonstrating how thought provoking I found your statement to be.
Keep up the good work.
Under red light we can make any colour appear to behave as if it was red. Under such conditions it’s meaningless to talk of ‘red’ or other colours.
Under the assumption that to be conscious is to appear as if one is conscious, then it doesn’t matter whether one is conscious or not. The word becomes meaningless.