The other long-term consideration is that whatever we do to animals, AIs may well do to us.
We don’t want future AIs raising us in cramped cages, purely for their own amusement, on the grounds that their utility is much more important than ours. But we also don’t want them to exterminate us on “compassionate” grounds. (Those poor humans, why let them suffer so? Let’s replace them by a few more happy, wire-heading AIs like us!)
Don’t many/most people here want there to be posthumans, which may well cross the species-barrier? I don’t think there is an “essence of humanity” that carries over from humans to posthumans by virtue of descendance, so that case seems somewhat analogous to the wireheading AIs case already. And whether the AI would do wireheading or keep intact a preference architecture depends on what we/it values. If we do value complex preferences, and if we want to have many beings in the world that have them mostly fulfilled, I’d assume there would be more awesome or more effective ways of design than current humans However, if this view implies that killing is bad because it violates preferences, then replacement would, to some extent, be a bad thing and the AI might not do it.
That argument would seem to apply to plants or even to non-intelligent machines as well as to animals, unless you include a missing premise stating that AI/human interaction is similar to human/animal interaction in a way that 1) human/plant or human/washing machine interaction is not, and 2) is relevant. Any such missing premise would basically be an entire argument for vegetarianism already—the “in comparison to AIs” part of the argument is an insubstantial gloss on it.
Furthermore, why would you expect what we do to constrain what AIs do anyway? I’d sooner expect that AIs would do things to us based on their own reasons regardless of what we do to other targets.
Perhaps this is true if the AI is supremely intelligent, but if the AI is only an order of magnitude for intelligent than us, or better by some other metric, the way we treat animals could be significant.
More relevantly, if an AI is learning anything at all about morality from us or from the people programming it I think it is extremely wise that the relevant individuals involved be vegan for these reasons (better safe than sorry). Essentially I argue that there is a very significant chance the way we treat other animals could be relevant to how an AI treats us (better treatment corresponding to better later outcomes for us).
“Other animals” is a gerrymandered reference class. Why would the AI specifically care about how we treat “other animals”, as opposed to “other biological entities”, “other multicellular beings”, or “other beings who can do mathematics”?
That’s the kind of thing I was objecting to. “‘Other animals’ are capable of feeling pain” is an independent argument for vegetarianism. Adding the AI to the argument doesn’t really get you anything, since the AI shouldn’t care about it unless it was useful as an argument for vegetarianism without the AI.
It’s also still a gerrymandered reference class. “The AI cares about how we treat other beings that feel pain” is just as arbitrary as “the AI cares about how we treat ‘other animals’”—by explaining the latter in terms of the former, you’re just explaining one arbitrary category by pointing out that it fits into another arbitrary category. Why doesn’t the AI care about how we treat all beings who can do mathematics (or are capable of being taught mathematics), or how we treat all beings at least as smart as ourselves, or how we treat all beings that are at least 1⁄3 the intelligence of ourselves, or even how we treat all mammals or all machines or all lesser AIs?
Perhaps it should. Being vegan covers all these bases except machines/AIs, which arguably (including by me) also ought to hold some non-negligible moral weight.
The question is really “why does the AI have that exact limit”. Phrased in terms of classes, it’s “why does the AI have that specific class”; having another class that includes it doesn’t count, since it doesn’t have the same limit.
After significant reflection what I’m trying to say is that I think it is obvious that non-human animals experience suffering and that this suffering carries moral weight (we would call most modern conditions torture and other related words if the methods were applied to humans).
Furthermore, there are a lot of edge cases of humanity where people can’t learn mathematics or otherwise are substantially less smart than non-human animals (the young, if future potential doesn’t matter that much; or the very old, mentally disabled, people in comas, etc.). I would prefer to live in a world where an AI thinks beings that do suffer but aren’t necessarily sufficient smart matter in general. I would also rather the people designing said AIs agree with this.
I would prefer to live in a world where an AI thinks beings that do suffer but aren’t necessarily sufficient smart matter in general. I would also rather the people designing said AIs agree with this.
But the original argument is that we shouldn’t eat animals because AIs would treat us like we treat animals. That argument implies an AI whose ethical system can’t be specified or controlled in detail, so we have to worry how the AI would treat us.
If you have enough control over the ethics used by the AI that you can design the AI to care about suffering, then this argument doesn’t show a real problem—if you could program the AI to care about suffering, surely you could just program it to directly care about humans. Then we could eat as many animals as we want and the AI still wouldn’t use that as a basis to mistreat us.
Yes, I guess I was operating under the assumption that we would not be able to constrain the ethics of a sufficiently advanced AI at all by simple programming methods.
Though I’ve spend an extraordinarily large amount of time lurking on this and similar sites, upon reflection I’m probably not the best poised person to carry out a debate about the hypothetical values of an AI as depending on ours. And indeed this would not be my primary justification for avoiding nonhuman suffering. I still think its avoidance is an incredibly important and effect meme to propagate culturally.
Jainism has a remarkably wide concept of creatures not to be harmed (e.g. specifically including insects). I don’t see why are you so focused on the diet.
The other long-term consideration is that whatever we do to animals, AIs may well do to us.
We don’t want future AIs raising us in cramped cages, purely for their own amusement, on the grounds that their utility is much more important than ours. But we also don’t want them to exterminate us on “compassionate” grounds. (Those poor humans, why let them suffer so? Let’s replace them by a few more happy, wire-heading AIs like us!)
Don’t many/most people here want there to be posthumans, which may well cross the species-barrier? I don’t think there is an “essence of humanity” that carries over from humans to posthumans by virtue of descendance, so that case seems somewhat analogous to the wireheading AIs case already. And whether the AI would do wireheading or keep intact a preference architecture depends on what we/it values. If we do value complex preferences, and if we want to have many beings in the world that have them mostly fulfilled, I’d assume there would be more awesome or more effective ways of design than current humans However, if this view implies that killing is bad because it violates preferences, then replacement would, to some extent, be a bad thing and the AI might not do it.
That argument would seem to apply to plants or even to non-intelligent machines as well as to animals, unless you include a missing premise stating that AI/human interaction is similar to human/animal interaction in a way that 1) human/plant or human/washing machine interaction is not, and 2) is relevant. Any such missing premise would basically be an entire argument for vegetarianism already—the “in comparison to AIs” part of the argument is an insubstantial gloss on it.
Furthermore, why would you expect what we do to constrain what AIs do anyway? I’d sooner expect that AIs would do things to us based on their own reasons regardless of what we do to other targets.
Perhaps this is true if the AI is supremely intelligent, but if the AI is only an order of magnitude for intelligent than us, or better by some other metric, the way we treat animals could be significant.
More relevantly, if an AI is learning anything at all about morality from us or from the people programming it I think it is extremely wise that the relevant individuals involved be vegan for these reasons (better safe than sorry). Essentially I argue that there is a very significant chance the way we treat other animals could be relevant to how an AI treats us (better treatment corresponding to better later outcomes for us).
“Other animals” is a gerrymandered reference class. Why would the AI specifically care about how we treat “other animals”, as opposed to “other biological entities”, “other multicellular beings”, or “other beings who can do mathematics”?
Because other animals are also sentient beings capable of feeling pain. Other multicellular beings aren’t in general.
That’s the kind of thing I was objecting to. “‘Other animals’ are capable of feeling pain” is an independent argument for vegetarianism. Adding the AI to the argument doesn’t really get you anything, since the AI shouldn’t care about it unless it was useful as an argument for vegetarianism without the AI.
It’s also still a gerrymandered reference class. “The AI cares about how we treat other beings that feel pain” is just as arbitrary as “the AI cares about how we treat ‘other animals’”—by explaining the latter in terms of the former, you’re just explaining one arbitrary category by pointing out that it fits into another arbitrary category. Why doesn’t the AI care about how we treat all beings who can do mathematics (or are capable of being taught mathematics), or how we treat all beings at least as smart as ourselves, or how we treat all beings that are at least 1⁄3 the intelligence of ourselves, or even how we treat all mammals or all machines or all lesser AIs?
Heh.
Have you been nice to your smartphone today? Treat your laptop with sufficient respect?
DID YOU EVER LET YOUR TAMAGOTCHI DIE?
Perhaps it should. Being vegan covers all these bases except machines/AIs, which arguably (including by me) also ought to hold some non-negligible moral weight.
The question is really “why does the AI have that exact limit”. Phrased in terms of classes, it’s “why does the AI have that specific class”; having another class that includes it doesn’t count, since it doesn’t have the same limit.
After significant reflection what I’m trying to say is that I think it is obvious that non-human animals experience suffering and that this suffering carries moral weight (we would call most modern conditions torture and other related words if the methods were applied to humans).
Furthermore, there are a lot of edge cases of humanity where people can’t learn mathematics or otherwise are substantially less smart than non-human animals (the young, if future potential doesn’t matter that much; or the very old, mentally disabled, people in comas, etc.). I would prefer to live in a world where an AI thinks beings that do suffer but aren’t necessarily sufficient smart matter in general. I would also rather the people designing said AIs agree with this.
But the original argument is that we shouldn’t eat animals because AIs would treat us like we treat animals. That argument implies an AI whose ethical system can’t be specified or controlled in detail, so we have to worry how the AI would treat us.
If you have enough control over the ethics used by the AI that you can design the AI to care about suffering, then this argument doesn’t show a real problem—if you could program the AI to care about suffering, surely you could just program it to directly care about humans. Then we could eat as many animals as we want and the AI still wouldn’t use that as a basis to mistreat us.
Yes, I guess I was operating under the assumption that we would not be able to constrain the ethics of a sufficiently advanced AI at all by simple programming methods.
Though I’ve spend an extraordinarily large amount of time lurking on this and similar sites, upon reflection I’m probably not the best poised person to carry out a debate about the hypothetical values of an AI as depending on ours. And indeed this would not be my primary justification for avoiding nonhuman suffering. I still think its avoidance is an incredibly important and effect meme to propagate culturally.
Go start recruiting Jains as AI researchers… X-/
I don’t see why. Jainism is far from the only philosophy associated with veganism.
Jainism has a remarkably wide concept of creatures not to be harmed (e.g. specifically including insects). I don’t see why are you so focused on the diet.
Vegans as a general category don’t unnecessarily harm and certainly don’t eat insects either. I’m not just focused on the diet actually.
Come to think of it, what are we even arguing about at this point? I didn’t understand your emoticon there and got thrown off by it.
I’m yet to meet a first-world vegan who would look benevolently at a mosquito sucking blood out of her.
I don’t think we’re arguing at all. That, of course, doesn’t mean that we agree.
The emoticon hinted that I wasn’t entirely serious.