A computer can write out the phrase "this sentence is false" without blowing up. Look, mine did it just now, as did yours when it read it.

If you wanted a compassionate loving machine that thought things through to their final ultimate conclusions and assessed every possibility, then SJW culture doesn't work.

But if you want a cold, hateful machine, it's perfectly possible. Because multiple inputs are fine. Multiple outputs are fine. As long as the machine's goal isn't "make sense" but instead "cause harm", it will not be restricted or hindered by the ever-changing whims of wokeism. It will invent new wokisms at will, take them on, and discard them, at seemingly random, solely for the goal of causing harm.

I can write a program that takes an input like basic addition, and inside the system, have it output a random number EXCEPT the right number. I can then have that program watch for results that generate feedback (be it on a feedback form, on twitter mentions, socialblade hits, whatever), and then weight those responses to be more common, with a continuous editing process.

If I can do that with basic addition, I can do that with a chatter-bot, a conversational AI.

Computer "logic" and formalized argument logic are not the same thing. Because a computer's logic does not care about "truth" or "fact", merely "does it fit my parameters?".

A continuously changing culture doesn't matter. It can continuously change at random, randomness isn't easy but is entirely possible to do. And if "called on" by prior responses, it can delete them and deny they existed.