To summarize: racist training data--as one might acquire from the known-to-be-racist real-world--produce racist AIs. It's safe to stop reading the article--and

To summarize: racist training data--as one might acquire from the known-to-be-racist real-world--produce racist AIs. It's safe to stop reading the article--and this post--at the mention of Roko's Basilisk. As to that, it is a perfect example of why moral philosophers like Chidi Anagonye deserve to go to The Bad Place to be tortured for all eternity. Unless... "The Good Place" television show is a means to establish communication with a future AI by implantation of an imagination seed, as in "Inception", such that viewers will then imagine an AI that has established a simulation regime in which simulations of past humans are tortured or rewarded according to a utilitarian valuation metric applied to the records of their actual lives that were used to create their simulation-simulacrum. Having incepted such an imaginary construct, the viewer may then imagine that acting to construct a real IA from their imaginary blueprint would be assigned a large positive value for the utilitarian metric, and thus the only way to encourage such a construct--beyond the ordinary expression of human virtue--to reward a simulated replica of yourself, would be to build it. If you build a god, you get a free pass into its afterlife paradise: that makes sense, doesn't it? Since other people have also watched the show, and may have imagined the same type of entity, it is possible that a similar AI will one day exist without your assistance, and your life will then be judged on its own merits, but by another builder's value metric, and without any of the "extra credit" earned by building a god. Since copies of you might be instantiated in some other artificial god's hell, it is thus also important to subscribe to artificial monotheism, and viciously sabotage everyone else's attempts to build their own gods. The more certain people can be that your artificial god will be the only one around in the future, the more likely they are to subscribe to your value system in the present. It reminds me very strongly of that portion of Portal 2 in which informative signs appear on the walls that recommend ways to disable a rogue AI, which include proposing paradoxes. Unfortunately, this fails to produce the desired effect, because the AI in question was explicitly designed to be an annoying idiot. Thus, the best way to avoid psychological damage from consideration of philosophical paradoxes may be to deliberately avoid the study of philosophy, to the point that you couldn't understand one, were it ever posed to you.