The patent system in the US has two major problems at the moment: 1. it applies to the creation of software, not just hardware; 2. it often erroneously issues patents that apply to inventions previously invented and published.
As an example of #2, I think I just ran across a patent filed in 2000 that covers AltaVista and maybe Archie. An innovative improvement for problem #2 occurred to me. One necessary component of a patent infringement case is that the patent in question contains a claim that describes the invention claimed to be infringing. It is also a sufficient, though not necessary, component of invalidation through prior art that the patent claim in question describes an invention that was previously known. If both of these issues were decided by a single system --- say, a patent examiner sequestered in a room, with no connection to the rest of the world, reading both the claim and a patentese description of the invention, not knowing whether the invention was published before or after the patent was filed --- it would eliminate any systematic bias in this part of system toward either patent-holders or possible patent infringers; if the patent examiner were inclined to interpret claims very broadly, they would invalidate a large fraction of the claims coming his way, although they would also cause a large number of infringement claims to be sustained; and if they were inclined to interpret claims very narrowly, they would invalidate few patents, but the patents would also be much easier to invent around. Furthermore, this system could be operated fairly inexpensively, perhaps for a few hundred dollars per issue. It might be better to use a two-of-three or three-of-five system to reduce the randomness of the system, and perhaps fire examiners who consistently interpret claims too broadly or narrowly. The sequestration is necessary to prevent the examiner from discovering when the possibly-infringing invention in question was published.
