Evolving the reaction set

[joan]

Hi,
Has it been considered to evolve the reaction set itself ? (I mean,
using genetic algorithm on reaction sets)
As reactions are hand crafted and designed, it's probably hard to find
the rules yielding to super rich behavior…

We could consider that a set of reactions that succeed at several
challenges simultaneously is richer than the ones succeeding
separately.

We would start with the complete set of neutral reactions, and
introduce some mutation to alter the rules.
Then we run variations until we have a few sets that succeed at
challenge 1. Then we take these winners, duplicate them and introduce
more variations, and check the new sets against both challenge 1 and
challenge 2.

We can go on like that until we have reaction sets that can win a
number of challenges at once. Not necessarily all the 19 challenges,
but maybe a choosen subset that we know expose interesting properties,
like creating polymers, catalysts, replicators, etc.

Obviously the rule sets found this way would be far from optimal. They
would have a lot of garbage reactions that don't help any particular
challenge. But that's probably ok, maybe these garbage reactions will
have some impact later on.

Any thoughts?
Thanks!