Is there anyone approaching the patent problem from another angle? Say by trying to productize research on prior art discovery? Or anything that would help patent officers become more productive (as in any of: do more without loss in quality in equal time or the same with increased quality in equal time or do more better in less time)?

Sure a human will vastly outperform the state of the art algorithm on any one patent but when talking about thousands of applications week in week out and increasing, some kind of automation to filter the cruft would be useful. Surely as bad as things are now, the worst that could happen is no effect? Would certainly be a better application of machine learning than classifying the mood of movie ratings or calculating social media "klout".