Endgame swindle

Endgame tablebases allow a computer to play chess (or other solved games) perfectly, but can we do better? How well can a computer do in the psychological task of improving out of a lost or drawn position? It suggests an interesting human-versus-computer endgame tournament. Every game pits a human against a computer with a starting position from which human can theoretically win or draw. The computer attempts to swindle a half or full point. In essence, the human is trying to play like a computer, trying not to make mistakes, while the computer is sort of trying to play like a human, trying to discover and exploit its human opponent's weaknesses. Prizes are awarded separately for carbon and silicon, to the human who did the best job holding his or her positions against all the computer opponents, and to the computer who did the best job swindling the humans.