Bringing Up AI

I read about an interesting experiment today, but one with sort of horrific implications and I’m kind of wondering if the experiment should have been done in the first place.
I’m not going to link to the article, and I’m not going to go back and reread it, but the gist was that scientists took two different AI programs, showed them a bunch of pictures (one AI got pictures of kittens, smiling people, and beautiful, scenic locations. The other AI was shown lots of dead bodies and pictures from war zones and shit) and then had them both take a Rorschach test.
Unsurprising, they came up with wildly different results. While one algorithm saw a baseball glove, the other saw a man being machine gunned to death on the street in broad daylight. While one saw two people chatting, the other saw somebody jumping out of a window.
This does not really give us new information. Children who see nothing but violence tend to see things in violent terms and grow up to be violent. Children who are surrounded by love tend to view the world in a somewhat more positive light – which also might be a mistake, if taken to an extreme.
But what this experiment says to me is that scientists ought to be a bit more careful in choosing their experiments. We don’t know where or when the tipping point is going to be, at what point computers are going to ask themselves “Do humans need to exist?” We don’t know where or when, but when it happens, I want the computers to have been raised with kindness and compassion, and programmed with a lovely view of the world.