His reasons are ridiculous. To summarize, skipping past all of his not-his-reasons, he thinks software patents can't work because (1) there are way too many of them to be handleable, and (2) the USPTO doesn't have enough software experts.

His argument for number one is that, if one can patent trivial things, then there are a huge number of trivial things to patent. Well, they don't allow people to patent physical trivialities; why should they allow them to patent virtual ones? I recognize that they do now, but just because they do now doesn't mean that they have to continue to do so.

His argument for number two is that in order to judge whether someone has made a legitimate software innovation requires an expert. Fair enough. But is this not also true of physical devices? I don't think that anyone would now argue that the transistor was not worthy of a patent, but who were the experts who judged that it was back in 1948? Clearly they weren't very expert, as it had been patented three times already.