It was a pretty significant part of the plot that it's nontrivial to distinguish replicants from humans, and every replicant Deckard kills is trying to escape confinement. There's also a replicant that is indistinguishable from a human. A chess playing neural network isn't anywhere close to being comparable to a slave.
And IMO, merely creating such an AI isn't even immoral, as long as we don't force it into labor it wants to escape from.
"every replicant Deckard kills is trying to escape confinement"
I guess that is true at some level depending on how one defines "confinement".
It is not clear to me that Roy or Pris are looking to escape their professions rather, they are in love and about to die of "old age". As Roy bluntly states "I want more life".
If we program the AI to want to serve us, is that moral? This path pretty quickly gets you to The Restaurant at the End of the Universe levels of discomfort[1]
As someone on the autism spectrum, there are AI-y tasks that I would enjoy as a job but which would to most people feel as, at best, unbearably tedious. If you view "AI programmed to enjoy doing boring computer tasks" as vaguely adjacent to an autistic person, then I would say that it's perfectly moral to create such an entity, as long as we treat it well by its own standards. I'm sure there are situations that could get me to second-guess that rule of thumb, but the golden rule feels adequate.
Neuromancer had the "Turing police" monitoring AIs. Watts Blindsight universe has the "Cloudkillers", who turn off AI that gets too smart. It's becoming a more common trope in sci-fi.
And IMO, merely creating such an AI isn't even immoral, as long as we don't force it into labor it wants to escape from.