While I don't doubt that in the future there may be some machine that can crunch logic at a ridiculous pace, Watson still seems far off from the AI the author here is alluding to. Machine learning of Watson's nature (classification problems, feature selection, etc.) is able to take data generated from the world and find interesting patterns to make a prediction. This is nowhere near the creation of new ideas, like Galois' insights into field theory or the work Perelman contributed toward the concept of a Ricci flow. I haven't yet heard how to create a machine whose thoughts "can wander" in a way that isn't hopelessly inefficient. That would be a big leap toward this type of AI.