Why do you think deep ecology values would not get locked in to an AI system? Presumably there are ethical priorities we might want an AI system to have or constrain itself by which are not instrumental for something else, so it’s not obvious to me that something needs to be instrumental to stick around as a value
Under what conditions do you think an AI system would possess values that lead it populate other worlds with animals? If deep ecology values are not locked in, I would think this makes AI’s much less likely to spread biological life to other planets or terraform them to enable this. I could see something more like a “recklessness” where it accidentally spreads life in pursuit of some other goal, but I have a hard time seeing what human-aligned goal set which explicitly excludes deep ecology would lead it to seed other planets with wildlife. Maybe something adjacent like Pet Planets for people to entertain themselves with / other cases where humans want animals around for a purely hedonistic purpose which is orthogonal to their welfare?
I hesitated on how to frame the deep ecology thing, because I think it’s entirely possible that it ends up locked in. I think my thought was something like the following. If AGI gets the values of its builders and then never modifies it, in the current race, it’s unlikely that AGI would lock in deep ecology values: these don’t seem massive in Chinese labs (could be wrong), and people in AI labs in the West are not hardcore ecologists, for the most part, because of political divisions.[1]
I do agree that AI systems could populate other worlds with animals for other reasons. Logically, we can’t cover all of the reasons why systems that we don’t know anything about would do something. The same applies to future humans.
(More broadly, I deliberately under-hedged all of the above. I don’t think we have any action-guidance on AGIxAnimals)
Nice points! A few questions:
Why do you think deep ecology values would not get locked in to an AI system? Presumably there are ethical priorities we might want an AI system to have or constrain itself by which are not instrumental for something else, so it’s not obvious to me that something needs to be instrumental to stick around as a value
Under what conditions do you think an AI system would possess values that lead it populate other worlds with animals? If deep ecology values are not locked in, I would think this makes AI’s much less likely to spread biological life to other planets or terraform them to enable this. I could see something more like a “recklessness” where it accidentally spreads life in pursuit of some other goal, but I have a hard time seeing what human-aligned goal set which explicitly excludes deep ecology would lead it to seed other planets with wildlife. Maybe something adjacent like Pet Planets for people to entertain themselves with / other cases where humans want animals around for a purely hedonistic purpose which is orthogonal to their welfare?
I hesitated on how to frame the deep ecology thing, because I think it’s entirely possible that it ends up locked in. I think my thought was something like the following. If AGI gets the values of its builders and then never modifies it, in the current race, it’s unlikely that AGI would lock in deep ecology values: these don’t seem massive in Chinese labs (could be wrong), and people in AI labs in the West are not hardcore ecologists, for the most part, because of political divisions.[1]
I do agree that AI systems could populate other worlds with animals for other reasons. Logically, we can’t cover all of the reasons why systems that we don’t know anything about would do something. The same applies to future humans.
(More broadly, I deliberately under-hedged all of the above. I don’t think we have any action-guidance on AGIxAnimals)
Maybe it’s good that environmentalists hate AI so much, because it ensures that people in labs are less likely to be friendly to pro-ecology views?