I almost objected, but you're right. The human has finite time with which to calculate. The predictor can easily be assumed to have more time. There's no problem calculating if a program will halt in finite time.