The Irony of 'Natural Selection' in Robots

There is now such a thing as “evolutionary robotics,” if you can believe it. An article published in Wired Magazine a few months ago called Robot ‘Natural Selection’ Combines Into Something New describes how this new branch of scientific research is attempting to improve the design of robots by simulating the process of natural selection. For example, one of the things that’s proven tricky to program robots to do is walk on various types of terrains. But researchers in Australia have shown how using a simulated natural selection process leads to unusual-looking but more effective robot leg designs.

Of course, in this case, it is computer code rather than genetic code that is being reproduced, and the variations caused by mutations in DNA reproduction are simulated by “adding ‘noise’ to the marriage of data.” As one of the researchers puts it:

The beauty of evolution is that it stumbles upon bonkers ideas all the time. No one, for example, designed a fungus to invade ants’ bodies and mind-control them around the rainforest — that unusual strategy emerged thanks to generation upon generation of random mutations and natural selection.

Maybe it’s just me, but I find such appreciation for the process that led to the parasitic mind-control of ants somewhat disturbing. I can only assume the researcher would heartily approve of the evolutionary horror show that is the tarantula wasp’s method for raising its young by laying their eggs in the bodies of still living and feeling but paralyzed tarantulas.

Just a Splash of Mutation, Please

The article was also interesting for the unabashedly optimistic tone the author had for these endeavors. Toward the end of the article, he claims

The resulting designs may be surprising, but they won’t be detrimental to our species, unless a designer tells the robots to evolve to kick as much human ass as possible. Unlikely, of course. …[This] could also unleash a new kind of engineering creativity, and almost certainly won’t lead to the machines taking our place as the dominant species on Earth.

Am I the only one troubled by the use of the qualifier “almost” in that last sentence? We are talking about the possibility of having robots overthrow and possibly enslave us, right? Is that something we want to enable even a tiny chance of happening? Also, asserting that the introduction of a mutation that might cause robots to try and control us is unlikely because humans are in charge of the mutations makes me think the author must have never interacted with actual humans before. 

Plus, it sounds like doublethink to say, on the one hand, we’re going to improve the design of robots by replicating natural selection, while on the other hand claiming we will somehow remain in control of it, so it doesn’t lead to anything too bonkers?

Silly Robot, Natural Selection is for AIs!

Finally, saying that only humans will forever remain in charge of governing the extent of the mutations is naive. It’s only a matter of time before AIs will be used to manage all sorts of scientific research. They’ve already far surpassed us in complex, “perfect information” games like chess and Go, and have reached the top echelons of complex, “imperfect information” games like StarCraft II, where you can’t always see everything your opponent is doing. For the former type of games, AIs are inventing novel strategies no human player has ever tried. Plus, they develop the capability to play at this level, starting from scratch, in a matter of hours rather than years.

Now, some AI researchers are also experimenting with combining deep learning algorithms with principles lifted from evolution to develop increasingly novel solutions (Computers Involve a New Path Toward Human Intelligence, Quanta Magazine). So it’s quite likely that the process of modifying code, be it DNA or Python, will very soon be done much better and faster by computers than people ever could.
 
Strangely, the process of natural selection we are starting to model our technology after appears to have produced a foundational dissatisfaction in ourselves with that process — at least as it applies to us. The old software engineering joke says, “it’s not a bug, it’s a feature,” but when we consider ourselves we seem to lose our good humor about the situation. You’d think we’d be nothing but pleased and grateful to have had the good fortune to stumble our way into being the most sentient species on Earth, but we aren’t. Instead, we fixate on all the things we don’t have and can’t do.

As a result of this pathological need to fill our perceived gaps, we’ve launched into activities like cutting and pasting our genes for whatever reasons we see fit. Today it’s to cure diseases for specific individuals, tomorrow perhaps a germline edit to ensure no one has to suffer the ignominy of dying of old age. So I can’t help but find it ironic that we’re employing natural selection to build better robots. If that’s the best process for developing them, why is it so inadequate for us?