Of course, predicting the future with 100% certainty is impossible. But that doesn't mean that making predictions is a mug's game. Predicting the future, with appropriate levels of uncertainty, is a very sensible thing to do with your time. I, for instance, predict that if I wander down to Cheeseboard in twenty minutes I'll find that they're selling delicious pizza. And I predict that if I eat that pizza, then it won't poison me. These are all useful predictions, which may be wrong, but are useful nonetheless.

It's only when you start sticking inappropriate error bars on your predictions that it becomes a problem. Kurzweil predicts things which are unlikely or perhaps impossible as having probabilities near 100%.