I've been thinking somewhere along the same lines, but eliminating bias is incredibly hard to do, even with machine learning. Hopefully we'll see some experimentation in this area here on HN.