Exploring Dead Ends

Photo via C.G.P. Grey

Imagine you and I were playing a dice game. You could choose between two sets of rules. In the first set of rules, you would get $5 no matter what number I roll. For the second set of rules, if I roll a 1,2,3,4, or 5 then you would lose $50, but if I roll a 6, then you win $10.

Unless you have an unhealthy belief in your good luck, this is a simple choice - for rational actors, when faced with a choice, it makes the most sense to choose the option with the highest expected value. A sure $5 is much better than the likely loss of $50, even though the maximum gain is less.

One of the reasons that I think evolution-type processes are so powerful and interesting is that they are too dumb to avoid bad decisions. Evolution "chooses" the second set of rules - it explores lots of dead ends, lots of poor designs, but because it is willing to suffer these losses, it also discovers options and makes advances that rational algorithms will always avoid.