Obviously, this went horribly wrong in this case, which brings us to the lesson:
Optimization technology can generate lots of designs that score well against some set of metrics. You still need someone with decent 'taste' to select which solutions you'll put into production. It doesn't matter if you're optimizing stock trading strategies, jet engines, or t-shirts.
Creating a decent UI for selecting good designs from a giant batch of autogenerated designs is a hard and interesting problem, and almost always needs to be custom built for the domain. Picking the best jet engine needs a totally different UI than picking the best stock trading strategy.