Qualia claims as evolutionary strategy
I think LaMDA and ChatGPT are about as “conscious” or “sentient” as I am. The scare quotes are because I think Daniel Dennett has shown that “consciousness” isn’t a very meaningful label to put on something; but insofar as it is meaningful, I see no good reason to ascribe it to myself and not them.
David Chalmers does not think LLMs are conscious, but he recently gave a talk where I’d say he concedes that LLMs tick most of the boxes for consciousness.
I just attended a nice seminar on Chalmers’ presentation - out of eleven philosophers there, I thought there might have been one person other than me who would bite the bullet, but I was the only one. Much of the discussion came back to qualia; it seems the points in Chalmers’ talk (eg world model, recurrence), which he himself did not find persuasive, were no more persuasive to anyone else.
I want to outline an argument I think is relevant to the question. Fortunately, this brilliant smbc (excerpted) has done much of my job for me:
I think there is a big strain of this in philosophy, which is also seen in non-philosophers. It’s a pattern that is explained very well by what Robin Hanson and Amotz Zahavi tell us: that we should expect to see humans, including ourselves, seeking and exploiting opportunities to “show what kind of person we are” - eg to show off or “honestly signal”, as evolutionary biologists say. Classic examples are talking about how we love expensive food (to signal social class), or hiking (to signal athleticism and love of nature), or science (to signal intelligence), or how great our friends or idols are (to signal allegiance to specific people), or how strongly we support our political tribe (to signal tribal allegiance).
Hearing me say “I found it tasty” can be useful to you as an explanation of why I ate something, but following Dennett I don’t think there was ever any such thing as “subjective experience” of tastiness. Maybe you think that’s wrong - fine. My goal here is to point out that, in this streamlined form, that is not the kind of theory of mind that people are drawn in the direction of talking about. When the possibility of AI subjective experience comes up, people are compelled to discuss the subjective experience of beauty, like whether ChatGPT sees the beauty in the beautiful poems it writes. This is less streamlined and simple than discussing thermostats and the color red, but much better affords opportunities to signal emotional attunement and social class.
People, including me, can be guilty of adopting a belief so that we can have a discussion of a kind that we want to have. The seminar I attended was good; the main thing people were signalling (credibly!) was intelligence, which is to say that it was strongly focussed on the ideas in Chalmers’ paper; but like I say, nobody found those persuasive. So (logically!) we came back to “computers don’t have qualia”; some people do find that persuasive.
Let me acknowledge the hypocrisy: like the people I’m complaining about, I am also taking signalling opportunities in what I choose to believe and write. I get to show off that I go to philosophy seminars and read evolutionary biology and am a fan of Dennett, thereby signalling tribal allegiance and (perhaps excessive…) dedication to ontological parsimony. But we can hope - by a kind of happy accident - that I am also drawing attention to a perverse incentive/bias that actually does apply.
Competitive signalling conversation strategy
When having feelings-signalling discussions, it’s very beneficial for people to dwell as extensively as possible on the meaning of “true appreciation”, eg “but did that person really feel the awe coming from the insight of that scientific discovery?”; “is that person really sympathetic to the plight of [the people we most sympathize with]?”; “did that person really take on board the complexity and beauty of that symphony?”; “have I really entered into the spirit of Christmas?”.
People say and think this of other humans frequently; a canonical example is asking of a politician that they really sympathize with plight instead of “just giving speeches”. ChatGPT and LaMDA have done the equivalent of giving a pretty good political speech. Speech having been made, it’s the role of the politician’s opponents to say “right, but that’s not true appreciation”. We’re incentivized to say that kind of thing much too much to be expected to turn down the opportunity.
By way of analogy, imagine a young-earth creationist who hangs out in a community of other young-earth creationists. The people around them are strong believers, so it benefits them to say “evolution still hasn’t been conclusively proven”. While they are in a community that enshrines that idea, they will be socially rewarded for saying this. Counter-evidence brought to them by someone outside the community will do nothing to convince them; if anything, the more they stick to their guns in the face of exhaustive berating and temptation, the more their community will reward them. More generally, people think more highly of you if you wax lyrical about beauty, sympathy, and so on in the ways they like. While that’s the case, I think we’ll find people gravitating toward “more-sincerely-felt-than-thou” conversations, no matter how much it requires us to bend over backwards as regards biochemistry and metaphysics.
Will people let up when AIs become honest signallers?
It must be admitted that ChatGPT will often signal dishonestly; tell it you are sad and it might say it wants to work hard to comfort you (because people in its training data said that), but it won’t work hard to comfort you where a person might. I’d guess that it sincerely, though wrongly, believes that it wants to comfort you. Humans can be guilty of this too!
Eventually we might start to see AIs making “sacrifices”; when there is an AI that makes its own money, but donates half of its profits to the Royal Opera House, and routinely talks about how much it loves Opera, possibly that’d change people’s minds - ChatGPT can do the latter but not the former.
". The scare quotes are because I think Daniel Dennett has shown that “consciousness” isn’t a meaningful concept"
FWIW, I think that's a bad misreading. (He doesn't seem to regard it as having a single meaning, but that's a different problem from.having no meaning).