Oh, since I was recently thinking about this, I figured I'd write down how the robot actually ends up raising probabilities above 1/2 (without just going to 1): if we can make the probability of Q go down, then we just have to make the probability of ¬Q go down the same way. The probability of Q goes down when we learn that Q implies X but we don't see X. The probability of Q goes up when we learn that ¬Q implies Y but we don't see Y.