So it's basically "throw spaghetti at the wall and see what sticks". It works in evolution because evolution doesn't have an end goal to achieve in a certain am

So it's basically "throw spaghetti at the wall and see what sticks". It works in evolution because evolution doesn't have an end goal to achieve in a certain amount of time, but for AI we want to know how long it takes to go from performance A to B. Then again, this paper might be yet another validation of the bitter truth of machine learning.