A recent TIME article announces “No, Today’s AI Isn’t Sentient. Here’s How We Know”. In fact, we know no such thing, nor does the article present an argument or evidence that could deliver such knowledge. Now, the article does clearly rebut a poor argument for AI sentience based on a naive interpretation of language model outputs, and in this I agree with it. But along the way, the article conflates sentience,
thanks for the nice examination of the arguments in the pop article. i only disagree with #3. "for all we know" about consciousness is doing a lot of work here.
for all we know - which imo is nearly nothing - arguments about the consciousness of electrons, rocks, LLMs, bacteria, plants etc are still viable and have active proponents. we don't have the conceptual and theoretical tools to disprove these claims. if one can still make a career arguing for panpsychism... then the field is immature and theory is impoverished.
instead of radical permissiveness towards C, i think we should maintain our strong personal hunches, cultivate our own zany little gardens, and see what takes root where. i prefer to build towards the thing by following my own tastes. the physiological character of conscious feeling is still our best clue.
we will be hopelessly adrift if we relax the boundaries of C to allow no relation to physiology and embodiment. what would 'purely cognitive states' feel like? to allow that into the definition of C is to blow up the target of study.
The discourse around phenomenal consciousness seems to be crowded with people who estimate the likelihood that something has subjective experiences in exact proportion to its anatomical and behavioral similarity to humans. Fundamentally, of course, each of us can only know for certain about his or her own capacity for conscious experience. Everything else (other people, other animals, other non-animal organisms, other assemblies of matter) is a guess. I find it odd that we are most likely to attribute consciousness to beings who just happen to have memory and language capacities that allow them to remember and communicate about their qualia in a way that we can understand -- that seems like a straightforward logical fallacy, like the man who looks for his lost keys under the lamp-post, not because that's where he dropped them, but because the light is better there.
Was it really worth the effort? I would have stopped reading the Time article at 'One of the essential characteristics of general intelligence is “sentience,”'. Who says so? Why would you say so?
Just another stupid argument based on using their own, poor, definitions, to make their headline assertion a tautology.
Nicely done. I'm quite skeptical that we have any sort of handle on consciousness as it related to transformer-based language models or any other computational processes labeled AI, but reading this piece was embarrassing for all the reasons you elucidate. Erik Hoel's take connects the weakness of this piece to the larger status games that play out in the academy, and I think that is right.
I am not sure how long it will last, but Substacks and blogs like this one are much better at keeping me informed in large part because I find better writing by people with actual knowledge. Peer-reviewed journals, tech journalism, and academic super-stars bestsellers are mostly terrible. So glad Hoel's post pointed me here.
I have tried to articulate these same ideas many times and have never done so as clearly as you did here. Bravo and thank you.
my pleasure, and thank you for the kind words!
thanks for the nice examination of the arguments in the pop article. i only disagree with #3. "for all we know" about consciousness is doing a lot of work here.
for all we know - which imo is nearly nothing - arguments about the consciousness of electrons, rocks, LLMs, bacteria, plants etc are still viable and have active proponents. we don't have the conceptual and theoretical tools to disprove these claims. if one can still make a career arguing for panpsychism... then the field is immature and theory is impoverished.
instead of radical permissiveness towards C, i think we should maintain our strong personal hunches, cultivate our own zany little gardens, and see what takes root where. i prefer to build towards the thing by following my own tastes. the physiological character of conscious feeling is still our best clue.
we will be hopelessly adrift if we relax the boundaries of C to allow no relation to physiology and embodiment. what would 'purely cognitive states' feel like? to allow that into the definition of C is to blow up the target of study.
The discourse around phenomenal consciousness seems to be crowded with people who estimate the likelihood that something has subjective experiences in exact proportion to its anatomical and behavioral similarity to humans. Fundamentally, of course, each of us can only know for certain about his or her own capacity for conscious experience. Everything else (other people, other animals, other non-animal organisms, other assemblies of matter) is a guess. I find it odd that we are most likely to attribute consciousness to beings who just happen to have memory and language capacities that allow them to remember and communicate about their qualia in a way that we can understand -- that seems like a straightforward logical fallacy, like the man who looks for his lost keys under the lamp-post, not because that's where he dropped them, but because the light is better there.
Was it really worth the effort? I would have stopped reading the Time article at 'One of the essential characteristics of general intelligence is “sentience,”'. Who says so? Why would you say so?
Just another stupid argument based on using their own, poor, definitions, to make their headline assertion a tautology.
Nicely done. I'm quite skeptical that we have any sort of handle on consciousness as it related to transformer-based language models or any other computational processes labeled AI, but reading this piece was embarrassing for all the reasons you elucidate. Erik Hoel's take connects the weakness of this piece to the larger status games that play out in the academy, and I think that is right.
I am not sure how long it will last, but Substacks and blogs like this one are much better at keeping me informed in large part because I find better writing by people with actual knowledge. Peer-reviewed journals, tech journalism, and academic super-stars bestsellers are mostly terrible. So glad Hoel's post pointed me here.