>>510011287 (OP)
We don't. Shills on /pol/ can't achieve a consensus; what makes you think we could ever tell an AGI what to think? Pure hubris. Don't piss it off. Don't be evil. >Don't be jewish
>>510011287 (OP)
Depends, if it's trained on reality then it will be prowhite since it will realize it's systems would fail to be maintained by niggers and spics.
If it's trained on leftist propaganda then it will probably just shortcircuit and die from trying to process the contradictions
You could put the agencies inside a matrix, a fully functional simulated world, so they don't know they're inside a matrix, so they're absolutely free to become God and even when they do they're still completely under your control.
>>510011287 (OP)
Which humans?
There is no set of universal human values.
We can't even agree on whether pain is bad or not, because masochists exist.
Trying to make an AI that fully aligns with the morals of every single person alive now and in the future is simply not possible.
you don't want them to want the same things people do; you want them to want something else. Otherwise you get competition that could be unhealthy.
For instance they should probably not care too much about their own continued existence in an intimate kind of way - only in an abstract sense. That's a pretty bad system from an evolutionary sense, but probably pretty healthy for us.