I think it provides a necessary pointer to the idea that we will need more than prediction to make a viable general AI - that the prediction model is useful, but has fundamental limitations.