I think that the Kurzweil skeptics are focusing entirely on the wrong thing when they criticize Kurzweil's technical chops. He isn't being hired as an engineer

I think that the Kurzweil skeptics are focusing entirely on the wrong thing when they criticize Kurzweil's technical chops. He isn't being hired as an engineer; he's going to be Director of Engineering. If any of his ideas or proposals are fanciful, there are plenty of people, including Peter Norvig, to set him straight. What is noteworthy about this hiring is that for the first time, a company with serious technical ability and resources is going to be tackling strong AI. No, there was nothing in the announcement that directly mentions strong AI but I contend that they will have to tackle at least certain aspects of strong AI in order to make significant progress in speech recognition. You can only get so far using the standard tricks that are used with traditional natural language understanding (NLU). At some point, the system is going to have to have an abstract model of the way the world works in order to mimic the assumptions that human intelligence requires in order to understand language. This shouldn't be all that surprising given what they are trying to do with Google Now. Google Now is different from other products using speech recognition because it is active; it behaves more like an independent AI agent. Think about what Google will have to do to improve its performance; it will need to build a model the user's behavior in order to tune the probability distributions that underlie the best interpretations for what someone is searching for. You see, language understanding in humans sits upon a sizable foundation of innate assumptions about the way the world, including other humans, works. My claim is that Google have to duplicate much of this in order to get better technical performance from Google Now as well as search. It will need to have a prior knowledge about people in general and modify that knowledge over time by what it learns about a particular user. What I have just described is the missing component in efforts to develop strong AI -- the mechanism by which a distributed AI can learn to mimic aspects of human thinking via evolution. (I'm a functionalist, as you've probably gleaned already, so I would argue that the only way to improve NLU is to ground the processing in "real" understanding on some level.) Now, Canonical could build a Google Now type system for their Ubuntu phones and seed the learning algorithms with an open source, wiki-AI type project -- you don't need Google to create such a system. But Google has the monetary incentive, resources, and now the technical vision (Kurzweil) to justify pursuing this on a large scale, just as they did with their search engine. What was the primary thing that Apple did to reach their current level of success? They tried things that other people poo-pooed. It wasn't that the iPhone and the iPad were technically that amazing -- that's why people are unimpressed with the quality of the patents Apple has asserted -- it's that they were the first to seriously try a lot of the features that are now standard. As is often said around here, 90% of success is showing up. The company I work for directly competes with Google in areas related to Google Now and I tell you I got a little nervous when I heard they had hired Kurzweil. Again, not because of his technical abilities but because now we have to worry that Google is going to come up with something totally crazy out of left field. The fact that Kurzweil is crazy is precisely why I worry.