I haven't had the time to invest in researching and processing the relevant materials for this idea yet!
My wife and I have a daughter who just turned one year old; it seems that the way she learns is by being wrong. As if her development is optimized to find new experiences which will prove the existing knowledge base incorrect.
Existing implementations of computing solutions, to my knowledge, optimize for the correct answer to problems.
A study by a BYU professor on how to produce optimal solutions from groups of humans, those with counterfactual participants performed better, is something that may relate to this idea.
Article HTML from Google Scholar: http://bit.ly/hY8bIi...
For HTM we could take that a step further: optimize for the expected error.
Website visitors place an item in their carts
They checkout at a high rate
Optimize HTM to search for ways to decrease the checkout rate
HTM figures the factors which decrease
HTM tests inverse