The idea that we can predict human behavior and thus invalidate the concept of free will is that there exists an algorithm that takes less time to run than the future human behavior it predicts will happen in. However, certain biological processes, like protein folding, take exponential time to compute, so we're going to have to make due with approximations for now. Thus, there will always be some aspect of the unknown with regards to human behavior until we get those algorithms to run faster.