Search This Blog

Follow adrianbowyer on Twitter

My home page

Wednesday 23 March 2011

NaturalIntelligence

There is no point in genetically engineering an organism unless the result is evolutionarily stable.  Usually that stability comes about as a result of engineering in a characteristic that is desirable to people - making a rice that is more nutritious, say.  The new rice will not be able to out-compete its wild rivals because it's wasting energy laying down beta-carotene (or whatever), but it still prospers because of its symbiotic mutualist relationship with its creators.  We protect and nurture it because it gives us something we need, so it doesn't go extinct.

We also put a lot of effort into eradicating pests.  There is a constant human war against pigeons, rats, cockroaches and the rest.  We rarely win of course (except for smallpox), but we manage.  However, suppose we were to change the balance?  Suppose we were to take a pest and to modify it so that it had a characteristic that people valued?  If we were to release that ex-pest into the wild, it should out-compete its unaltered brethren because it would enjoy human support.  We would have won by overwhelming the pest with competitors from among its own kind.

Doing this agriculturally (like the rice) would be useless.  We concentrate agricultural species into a protected monoculture in one place.  In contrast, the ex-pests would have to survive in the wild to compete successfully with the continuing pests, just occasionally gaining support from people to tip the balance.  How could we engineer a characteristic in the ex-pest that would benefit it in just occasional interactions with people?

How about performing for food?  For example, how about stand-up mice?

Artificial intelligence researchers are fond of computer programs called chatbots.  These are formulaic (and quite simple) pieces of software that are able to hold rather restricted conversations with people.  They are a step on the road to programs that can pass the Turing Test.  That's a long road, and chatbots really are only one step, but they are quite interesting to talk with.  They can be set up to tell jokes, to ask people riddles and to conduct all sorts of other, well, chat.

We can't make a genuinely intelligent mouse, of course (well - not yet).  But we might be able to incorporate the straightforward algorithmic processes of a chatbot, and to alter the animal's vocal tract to form the equivalent of a simple train-announcement-style speech synthesiser.  The mouse would come up to you and tell you a joke.  If you gave it cheese, it would tell you another...

Soon, all the mice in the world would be much-loved well-fed comedians...

No comments:

Post a Comment