On prediction #4, I think we have to say Kurzweil missed. While voice recognition software is certainly improving (we have pretty good verbal Google searches, decent verbal customer service, and even decent text to speech on PCs), but it would certainly be wrong to say “the majority of text” involves speech recognition. We also certainly do not have anything near “ubiquitous” user interfaces for verbal language. In fact, while they also exist, they are still pretty cumbersome and not nearly universal in their use.
Kurzweil missed this one and I find his miss instructive. He knows perhaps more about this space than most because of his background in building aid devices for people who are blind and deaf. However, it is perhaps his closeness to the subject that caused him to miss this one so clearly.