I don’t think there’s much crossover. I hope you know that there are lots and lots of incentives for active deception and responding to deception in various parts of the natural world and evolutionary psychology—if you’re interested in the workings of and responses to deception, definitely read more about it. Like, the argument you make for females being interested in “people over things” could also explain the reverse—males are incentivized to deceive females, which you can do better the better you model people, right? I think you are observing something real about relevant preferences, but if that’s the extent of your understanding, I’d learn more about evolution and alternate explanations e.g. cultural pressure towards taking on emotional labor.
Anyhow, this example is narrow and specific to a human problem. As you say, the concern about AGI is mainly about intelligence significantly past humans, that do not share a basic substrate or set of biological imperatives. Like, even a person who I think might be lying to me can be modeled as fundamentally human—having limited amounts of information, limited physical strength, needing to eat, fearing death, etc. Heck, if I’m looking for a partner and am concerned that the partner is going to try to deceive me to get sex or whatever from me, I’m already aware of the threat!
The current environment you’re asking about people’s experience in is also pretty damn different from the ancestral environment evolved for—in as far as resource constraints, information ability, and I guess most other things—so I doubt that this example applies much.
Nice link on the Wikipedia article, thank you for that. “Koko, a female gorilla, was trained to use a form of American Sign Language. It has been claimed that she once tore a steel sink out of its moorings and when her handlers confronted her, Koko signed “cat did it” and pointed at her innocent pet kitten”. That animal, Koko, was just incredible. Having watched her on a few videos, I find that story perfectly plausible...
Humans are pretty clearly very especially generally intelligent, and so will display far more of the problems with aligning a general intelligence than displayed in animal interactions.
Like, the argument you make for females being interested in “people over things” could also explain the reverse—males are incentivized to deceive females, which you can do better the better you model people, right?
Males are hypothetically less incentivized to get alignment. So the knowledge about alignment would hypothetically be more concentrated in females. It would still be relevant to understand how males (or anyone) deceives others, specifically for understanding deceptive alignment.
Like, even a person who I think might be lying to me can be modeled as fundamentally human
Yes, I agree it’s much easier of a problem, e.g. for the reasons you list. It’s a very common tactic when dealing with an impossible seeming problem, to focus on easier but still very nontrivial versions of the problem.
I don’t think there’s much crossover. I hope you know that there are lots and lots of incentives for active deception and responding to deception in various parts of the natural world and evolutionary psychology—if you’re interested in the workings of and responses to deception, definitely read more about it. Like, the argument you make for females being interested in “people over things” could also explain the reverse—males are incentivized to deceive females, which you can do better the better you model people, right? I think you are observing something real about relevant preferences, but if that’s the extent of your understanding, I’d learn more about evolution and alternate explanations e.g. cultural pressure towards taking on emotional labor.
Anyhow, this example is narrow and specific to a human problem. As you say, the concern about AGI is mainly about intelligence significantly past humans, that do not share a basic substrate or set of biological imperatives. Like, even a person who I think might be lying to me can be modeled as fundamentally human—having limited amounts of information, limited physical strength, needing to eat, fearing death, etc. Heck, if I’m looking for a partner and am concerned that the partner is going to try to deceive me to get sex or whatever from me, I’m already aware of the threat!
The current environment you’re asking about people’s experience in is also pretty damn different from the ancestral environment evolved for—in as far as resource constraints, information ability, and I guess most other things—so I doubt that this example applies much.
Rewritten more abstractly: https://www.lesswrong.com/posts/ypvs4asdFq7riDWmd/interpersonal-alignment-intuitions
Nice link on the Wikipedia article, thank you for that. “Koko, a female gorilla, was trained to use a form of American Sign Language. It has been claimed that she once tore a steel sink out of its moorings and when her handlers confronted her, Koko signed “cat did it” and pointed at her innocent pet kitten”. That animal, Koko, was just incredible. Having watched her on a few videos, I find that story perfectly plausible...
Humans are pretty clearly very especially generally intelligent, and so will display far more of the problems with aligning a general intelligence than displayed in animal interactions.
Males are hypothetically less incentivized to get alignment. So the knowledge about alignment would hypothetically be more concentrated in females. It would still be relevant to understand how males (or anyone) deceives others, specifically for understanding deceptive alignment.
Yes, I agree it’s much easier of a problem, e.g. for the reasons you list. It’s a very common tactic when dealing with an impossible seeming problem, to focus on easier but still very nontrivial versions of the problem.