The strongest argument he makes is that there's an impedance to text entry vs. "tactile computing." Which seems pretty intuitive to me. With a keyboard a user can generate tons more information (as opposed to raw data) than with almost any other input hardware.

But you have to think about what information you're inputting. A GUI provides an ongoing dialog that guides you through the interaction with the computer. But if there's data that's unavailable a priori to the interface developer, a keyboard is still the best way to enter it.