I don't think these people are trying to develop general-purpose AI and then direct it at the problem of making a human-shaped machine behave sufficiently humanlike to make it a satisfying sex partner. That would be one of the most amazing cases of achievement without ambition that I've ever heard of. Like developing and perfecting room-temperature fusion so that you could deliver a washing machine that you didn't need to plug into the wall.
I think these people are trying to develop a human-shaped machine that behaves sufficiently humanlike to make it a sufficiently satisfying sex partner for a sufficiently large market. This probably means teaching it to hold a fairly restricted set of conversations, and (more importantly) to move around and handle physical contact in the right ways. Maybe they'll plug it into some other, more advanced services developed by other, larger companies to handle speech recognition and slightly broader conversations.
If someone does develop a library that anyone can use to instantiate a general-purpose AI, and that actually becomes the simplest way for these people to implement their product, I think they would instantiate it with an emotional makeup such that all they wanted to do was please their owner. In such a case, slavery would be unnecessary. (Is instantiating AIs with arbitrary emotional makeups immoral? The thing is, you probably have to pick some makeup, and choosing a potentially bad one at random is just as irresponsible as choosing it deliberately. The most obvious criterion, "Will this new AI I'm creating have a happy life?", would probably be easy to satisfy.)
I think these people are trying to develop a human-shaped machine that behaves sufficiently humanlike to make it a sufficiently satisfying sex partner for a sufficiently large market. This probably means teaching it to hold a fairly restricted set of conversations, and (more importantly) to move around and handle physical contact in the right ways. Maybe they'll plug it into some other, more advanced services developed by other, larger companies to handle speech recognition and slightly broader conversations.
If someone does develop a library that anyone can use to instantiate a general-purpose AI, and that actually becomes the simplest way for these people to implement their product, I think they would instantiate it with an emotional makeup such that all they wanted to do was please their owner. In such a case, slavery would be unnecessary. (Is instantiating AIs with arbitrary emotional makeups immoral? The thing is, you probably have to pick some makeup, and choosing a potentially bad one at random is just as irresponsible as choosing it deliberately. The most obvious criterion, "Will this new AI I'm creating have a happy life?", would probably be easy to satisfy.)