Most of the 4569 software patents last year were in what I consider "hardware" or systems areas: image processing and video (623 patents), networking and telecommunications, operating systems, process control, and graphics; also multiprocessing and parallel programming. Then come applications: health care, engineering, transportation, GUI, signal processing, databases, CASE, security, circuit design, financial management, office automation, CAD, word processing, physics, and navigation; also biology (71), chemistry (51), geophysics (47), education (33), spreadsheets (8), and biotechnology (6). Most of the patents went to IBM (396), Hitachi (189), and other large companies. Fewer than two patents per week were issued for applications in each Communique-related technology: speech recognition and synthesis, robotics, neural networks (89), distributed processing, pattern recognition, compression, AI (78), music, NLP, numerical analysis, character recognition (62), algorithms, OOP, games, fuzzy logic (47), simulation, vision, and virtual reality (26). [Gregory Aharonian , Internet Patent News Service. comp.ai, 1/30/95. David Joslin.] (Shouldn't you be reading your two patents per week?)

Greg Aharonian says that 10K software patents will be issued last year and this year, at an average of expense of $5K, nearly all with inadequate reference to prior art. With over 500K books, dissertations, articles, conference papers, and product descriptions published, there should be no shortage of references for any new discovery. (The applicant must cite prior art to show how the new work differs.) Aharonian's own searches routinely turn up 24 relevant non-patent references, and he would seek out at least 48 for use in a litigation. The 24 figure is similar to the average number of citations in an IEEE software journal article. Yet searches by the patent offices in the US, Europe, and Japan turn up an average of about one non-patent reference and four related patents. (Each patent organization achieves a similar hit rate, but with emphasis on it's own database for prior patent citations.) For this search, the US PTO charges $420 and the European EPO $1537. Aharonian says that the searches are amazingly superficial, and that your money would be better spent on a private search (e.g., through Dialog, or through his service). Patent examiners can search Dialog, but must pay for both the search time and $10 for any article ordered -- so it is likely that they work from abstracts alone to determine relevance. [srctran@world.std.com, misc.int-property, 1/31/95.]

(If you'd like to help the PTO, they would gladly accept donations of professional books -- especially those that are now out of print. Contact Michelle Crecca .)

In less than 20 years, people can freely use the ideas now being patented. In the meantime, "if someone sues you for using their little patented subroutine, tell them you'll start a re-exam of their patent, threatening with all of this obvious, prior art." Barry Chapin , misc.legal.computing, 1/31/95.] (There have already been such cases, but the process can be expensive. Patent infringement insurance costs about $50K/year, according to Greg Aharonian.)

Mitch Kapor suggests that software protection should last 2-3 years, as is true for semiconductors (via the Semiconductor Chip Production Act of 1984). [Information Week, 2/13/95, p. 26. EDUPAGE.]