Look at the Huge Difference between Artificial and Natural Selection

August 5, 2014

I’m always saying that we can’t program a true AI, but it must evolve by natural selection. That’s the only way the AI can really be conscious, alive and morally equivalent to a human being.

Green snake in Kinuta Park - June 15, 2013

I met this beautiful green snake who was crossing the jogging path in Kinuta Park. He was more than 2 meters long! (June 15, 2013) (Enlarge)

The problem with this, which various people have pointed out, is that natural selection is out of our control. Something that evolved by natural selection would just be pursuing its own self-interest, and it could even compete with humans. That could get dangerous! An intelligent machine that evolved without human intervention or any kind of tweaking might be “unfriendly AI.” It might declare war and wipe us out.

In short, if something evolved by natural selection, it would have a mind of its own. And that of course is the whole point.

Let’s pause a moment, though, and just consider the difference between natural selection and artificial selection. You know artificial selection is how we got chihuahuas and St Bernards. People controlled the dogs’ reproduction and chose which individuals would mate. With each new generation, the dogs approched more closely the image that the breeder person had in mind as a goal.

With natural selection, by contrast, there is no goal. This is an amazingly important difference, and it matters a great deal for AI development. With natural selection, the breeding is done by death itself, which is pretty awe-inspiring when you think about it.

And death is not impressed by the products of our artificial breeding programs. After all, most of our domestic breeds can’t survive in the wild because they depend on people for food, or protection. If human beings suddenly disappeared, those human-bred animals and plants would also die out very quickly.

Is that the kind of AI we want? Is that the vision of artificial intelligence we are striving to develop? Do we only want a docile machine that we can lead around by the nose, like cows or sheep? Do we want uniform rows of AI servants standing at attention like corn growing in a field?

Somehow I think this is not ambitious enough. I’d rather have a totally wild AI, even if it ends up like a grizzly bear, or like mosquitoes, or like some weird strain of bacteria that we haven’t even discovered yet. I’m not afraid of wildlife. The important thing is that it really is alive.

Perhaps one reason we have not yet invented true AI is because of fear. We lacked the courage to give our computer creations their freedom. We lacked the ambition to build a wild AI – because we were just thinking about chihuahuas.