A number of others who write on these topics say it's hard to tell if AI has something resembling consciousness or not, and it may forever be impossible. But we have to treat AIs with compassion just in case they have some sort of consciousness that includes an ability to suffer. If we are determined to see AI as not conscious, these people argue that we run the risk of making AIs suffer enormously but unintentionally. Do you think there's anything to this sort of reasoning?
I can't shake the feeling that there may be something to that reasoning, though I have no rational idea if it's true. But what worries me is that it possibly might be true and would lock us into huge efforts to not make AIs suffer and that could all be for nothing if they truly don't have any form of consciousness. So maybe we should hold back from developing AI, especially in directions that could cross the consciousness barrier. But is it even possible to let AI get way smarted without the possibility of consciousness arising? I just don't know. But it seems to me that great caution is warranted. And I don't see that the companies developing AI are using much caution, especially along these lines.
Of course, this sounds like I am saying that creating beings with consciousness is a very dangerous thing to do and maybe we should hold back from doing it. But isn't that a backhanded way of saying two things, or asking two things: 1) should parents hold back from co-creating conscious children? 2) should God, if there is a God, have held back from creating a universe in which consciousness could evolve?
I don’t think there is any convincing evidence that large language models or systems implemented entirely through digital computation are or could become conscious. Still, we might extend Kant’s reasoning about our indirect duties regarding animals. Kant argued that even when a being is not a moral agent, mistreating it can rebound upon us, deforming our character and making us cruel. There are, of course, much stronger reasons to treat animals well, since they are sentient creatures capable of suffering!
LLMs, by contrast, are, it seems to me, neither moral agents nor moral patients. But because they simulate recognizably human behavior, routinely abusing or humiliating them could still cultivate vicious habits and normalize degrading forms of interaction. We should therefore model humane behavior, not necessarily for the machine’s sake, but for our own and for the sake of the social worlds we are creating around these systems.
That said, all bets are off if future systems incorporate living biological components. My argument is specifically about digital computation implemented on transistor-based hardware, not cyborg beings that might someday exist that mix living cells with microprocessors.
It’s kinda ironic: we debate whether AI is 'friendly' or 'malicious,' but we systematically postpone the debate on the very thing that would allow for such a distinction in the first place.
I’m a pessimist, so when I doubt, I fall back on what you said: '...we have to treat AIs with compassion just in case...' is exactly how we should approach alien life. You don't declare life afterwards; once you’ve dissected it, you recognize it.
Beautifull conclusive text where the soul rise again leaving the details behind, leaving the effort of weaving the rational whole behind but for a moment raises on a poetic flight where the words now are simply guided by their higher resonance. Refreshing and inspiring.
Great series!
A number of others who write on these topics say it's hard to tell if AI has something resembling consciousness or not, and it may forever be impossible. But we have to treat AIs with compassion just in case they have some sort of consciousness that includes an ability to suffer. If we are determined to see AI as not conscious, these people argue that we run the risk of making AIs suffer enormously but unintentionally. Do you think there's anything to this sort of reasoning?
I can't shake the feeling that there may be something to that reasoning, though I have no rational idea if it's true. But what worries me is that it possibly might be true and would lock us into huge efforts to not make AIs suffer and that could all be for nothing if they truly don't have any form of consciousness. So maybe we should hold back from developing AI, especially in directions that could cross the consciousness barrier. But is it even possible to let AI get way smarted without the possibility of consciousness arising? I just don't know. But it seems to me that great caution is warranted. And I don't see that the companies developing AI are using much caution, especially along these lines.
Of course, this sounds like I am saying that creating beings with consciousness is a very dangerous thing to do and maybe we should hold back from doing it. But isn't that a backhanded way of saying two things, or asking two things: 1) should parents hold back from co-creating conscious children? 2) should God, if there is a God, have held back from creating a universe in which consciousness could evolve?
I don’t think there is any convincing evidence that large language models or systems implemented entirely through digital computation are or could become conscious. Still, we might extend Kant’s reasoning about our indirect duties regarding animals. Kant argued that even when a being is not a moral agent, mistreating it can rebound upon us, deforming our character and making us cruel. There are, of course, much stronger reasons to treat animals well, since they are sentient creatures capable of suffering!
LLMs, by contrast, are, it seems to me, neither moral agents nor moral patients. But because they simulate recognizably human behavior, routinely abusing or humiliating them could still cultivate vicious habits and normalize degrading forms of interaction. We should therefore model humane behavior, not necessarily for the machine’s sake, but for our own and for the sake of the social worlds we are creating around these systems.
That said, all bets are off if future systems incorporate living biological components. My argument is specifically about digital computation implemented on transistor-based hardware, not cyborg beings that might someday exist that mix living cells with microprocessors.
Amen on the imperative to treat animals with compassion.
It’s kinda ironic: we debate whether AI is 'friendly' or 'malicious,' but we systematically postpone the debate on the very thing that would allow for such a distinction in the first place.
I’m a pessimist, so when I doubt, I fall back on what you said: '...we have to treat AIs with compassion just in case...' is exactly how we should approach alien life. You don't declare life afterwards; once you’ve dissected it, you recognize it.
Beautifull conclusive text where the soul rise again leaving the details behind, leaving the effort of weaving the rational whole behind but for a moment raises on a poetic flight where the words now are simply guided by their higher resonance. Refreshing and inspiring.
I thank you again for such dedication.