Scientific American describes self-repairing computers

Our group of research collaborators at Stanford University and the University of California at Berkeley has taken a new tack, by accepting that computer failure and human operator error are facts of life. Rather than trying to eliminate computer crashes--probably an impossible task--our team concentrates on designing systems that recover rapidly when mishaps do occur. We call our approach recovery-oriented computing (ROC).

They began by looking at why computers fail. They found: We were a bit surprised to find out that operator error was a leading cause of system problems. Traditional efforts to boost the dependability of software and hardware have for the most part overlooked the possibility of human mistakes, yet in many cases operators' miscues accounted for more downtime than any other cause

Given this, they are exploring ROC computer design with the principles of speedy recovery, better tools for pinpointing problems when they do occur, support for 'undo' functions for all programs, and the ability to inject test errors