It's been a while, but if I recall Penrose basically attacks the notion of "stong AI". Others, e.g. Scott Aaronson, have ripped his argument to shreds. But not to dismiss Penrose's argument too quickly, Aaronson did devote an entire university course to presenting the counter argument http://www.scottaaronson.com/democritus/. I think there is a case to be made against strong AI along the lines of "just because a posteriori you can model anything digitially does not make it so", but apparently Penrose failed. (And don't get me wrong, I'm not the guy to make the case either.)