I was fortunate to have Gerry teaching my section of 6.031 (the predecessor of 6.001). I remember a story he liked to tell about why he thought Marvin Minsky was very smart. Gerry was building some kind of connectionist system -- a neural net or something, I don't recall exactly -- and mentioned to Marvin that he was initializing the weights randomly "so it won't have any preconceptions". Marvin replied, "Of course it will have preconceptions! You just won't know what they are. "