Inadequacy of PCT/PTO/EPO/JPO software prior art searches Gregory Aharonian srctran@world.std.com January 1995 That 10,000 US software patents will be issued last year and this year is a very troubling sign, as it is hard to conceive that in the 1991 to 1993 time period when most of these patents were filed, that there was that many novel, unpublished, unobvious software ideas. (NOTE: More statistics and studies on software patents are available at my Web site, www.bustpatents.com, including a study of US patent quality from 1976 to 1999 at www.bustpatents.com/corrupt.htm.) While there are a variety of fine tunings that can be made to how Patent office around the world handle software patents (such as hiring computer scientists), there is one fundamental problem that will require more than fine tuning - that of software prior art. As we speak, none of the patent offices around the world are qualified to conduct novelty analyses for software patents. Over two thirds of all examiners and searchers involved with software patents do not have the skills, tools, and resources to find much of the prior art in which to assess a software patent application. As one bit of evidence, 75% of all issued US software patents cite less than two non-patent prior art items. Unless someone in Washington (and Brussels) figures a way to give the Patent offices the resources they need to do the job right, you might as well suspend software patent processing - as now it is unfair to everyone. What follows are two studies: a relative comparison of the software searching capabilities of the PTO/EPO/JPO (about the same), and an absolute analysis of the PTO's software searching capabilities compared to industry norms (they find about 15%). RELATIVE COMPARISON OF THE EPO/PTO/JPO SOFTWARE SEARCHING CAPABILITIES To compare the three Patent offices, I put together a sample of 250 PCT applications dealing with software patents. These were PCT applications published in 1994, based on patents filed in America, Europe and Japan in late 1992. PCT applications allow you to preserve your US filing date when applying for patents in other countries (and similarly for those of you filing patents in your home country). If you file in the US, and have the US PTO conduct the PCT search, it costs $420, but if you have the EPO conduct the search, it costs $1537. So where your PCT application search done is important. The EPO has a better reputation for doing searches, but at least for software, this is not very warranted. For the 250 PCT software applications that I examined, PCT searchers on the average found 1.2 non-patent software prior art references and 4.0 patent software prior art references (comparable to US frontpage prior art citations, all of which will be reviewed in the second part below). 110 of the searches were done by the EPO, 88 by the PTO, and the rest by the Japanese, Swedish and Australian patent offices (whose averages were similar to the EPO and PTO). For the EPO, EPO searchers found an average of 2.0 non-patent software prior art references and 3.0 patent software prior art references, while the PTO searchers found an average of .66 non-patent software prior art references and 5.4 patent software prior art references. Compared to the many dozens of prior art references that should have been found, there is no appreciable differences between the EPO and the PTO when it comes to software prior art searching capabilities. The Patent offices tend to do best finding patents in their sphere of the world - EPO tends to cite EPO/WO patents, PTO tends to cite US patents, and JPO tends to cite Japanese patents, and do find decent numbers beyond those cited in the application. They all do worse when it comes to finding patents from other countries. This, and the fact the US finds more patents, is easily explainable: the examiners have free unlimited online searching use of their own patent databases and have the luxury to search and find patents (and that PTO searchers find more patents than EPO searchers is due in part to the larger US patent database (APS) and easier access to APS for PTO searchers). Some of the PCT searches done by both the EPO and the PTO exhibit a rather cavalier attitude towards the applicants fees. For example, one company sent in three applications for software technology that dealt with GUI editors and object manipulators (WO94/27227, 27228, 27229). Three applications, three $1537 search fees, applications for which there are tons of non-patent prior art (for example, one application dealt with a GUI for manipulating objects between a desktop and a document). The EPO did all three searches, and returned the same results for all three searches - 5 patents, 5 technical papers. Given that there are dozens and dozens of papers on the subject, a careful search should have been able to find different references for each application. Other suite filings searched by both the EPO and the PTO exhibit similar patterns - the same search results for related applications. Unless there was a discount somewhere, this is unfair to the applicant. The US PTO is just as bad with software. For example, consider PCT application WO 94/14127 titled "Asynchronous system and method for electronic data management, storage and communication", a typical distributed processing software patent application. Myself, I would have used a variety of keyword search strategies on APS to find relevant prior art, and the same strategy on something else, such as Dialog. Here is the entire search session done for this PCT application (which surprisingly was listed in the International Search Report verbatim, as the erroneous L7 query shows): APS => d his (FILE 'USPAT' ENTERED AT 15:06:59 ON 29 JAN 94) L1 414 S SHARED (3A) STORAGE L2 2888 S ARCHIV? L3 15 S L1 AND L2 L4 5918 S (ENCRYPT? OR COMPRESS?) (3A) (DATA OR FILE#) L5 17042 S (SHARED OR COMMON) (3A) (STORAGE OR DISK# OR DRIVE#) L6 115 S L4 AND L5 => s l2 and 6 1576869 6 L7 2786 L2 AND 6 => s l2 and l6 L8 10 L2 AND L6 Yes, that was the entire $420 PCT International Search done by the PTO - a minimal prior patent search and a non-existent prior non-patent search. There are entire conferences and books devoted to just this subject - none cited. All that was done was a quicky one-pass keyword search on APS, when APS use is free to the examiner. For US applicants, one conclusion is that unless you want the cosmetic benefit of a EPO search for your international application, you should save the $1000 and have the US PTO do the PCT international search. This is definitely true for software patent applications, and I suspect other areas of technology. US applicants file thousands of PCT applications every year - this translates to million of dollars being spent with little benefit. The money would be better spent on conducting your own prior art search. Not much of a contributor to the US trade deficit, but rather money better spent here. That the PTO Commissioner will be shipping some US PCT searches over to EPO is somewhat disconcerting, if there is no appreciable differences in the quality of the two offices ability to search for prior art, and if the USPTO is making up the $1000 difference in fees. Someone in this country - the PTO, or the AIPLA - should conduct an indepth study of PCT searching practices at the PTO and EPO and see if the equivalence found in software is true for most areas of technology. If so, US applicants should be discouraged from paying the extra costs of electing a EPO PCT search. Thus, for software there are no appreciable differences in searching skills by the various Patent offices around the world, and I suspect much the same for other areas of technology. The national governments are not insuring that their Patent Offices have the resources to handle software prior art. ABSOLUTE CAPABILITIES OF THE PATENT OFFICES WITH NON-PATENT SOFTWARE PRIOR ART Average issued US software patents cite 8 US patent prior art references, which with additional prior patent refernences from PCT searches is near the dozen or so prior patent references that I find for the application prior art searches that I do. This is not unexpected, given that examiners have free unlimited searching access to their own patent databases. For non-patent prior art, six numbers to keep in mind: .66, 1.2, 2.0, 2.1 and 24 and 48. The first three numbers are the above cited averages for PCT non-patent prior art searches done by the EPO and PTO separately and together. The fourth number, 2.1, is the average number of non-patent software prior art references in over 2000 software patents from 1994, which as the latest sample of US software patents, most likely holds true for all 15,000 US software patents since 1970. 24 and 48? 24 is the number that a good prior art search for a patent application should find on the average, while 48 is the minimum number that a killer prior art search for an infringement defense should find. I base these numbers on many things, but two reasons in particular. The first is that these are numbers typical of the number of non-patent prior art items that I find when I do searches. Admittedly I have better tools, skills and resources (being in Boston) than most patent examiners, but I am trying to establish a limit. The second reason is that these numbers are those typically found in studies of citation references (check out the journal Scientometrics). For example, one software-related IEEE journal did a study of bibliographic references in over 2000 papers published in their journal. Here are the yearly averages over the same time period that software was being patented in the US: YEAR Average No. of References 1971 10.7 1972 12.0 1973 12.1 1974 14.8 1975 15.9 1976 14.2 1977 17.6 1978 14.1 1979 16.2 1980 18.4 1981 19.7 1982 18.7 1983 25.9 1984 19.1 1985 25.4 1986 21.0 1987 25.2 1988 29.3 1989 26.8 1990 26.0 Much like the number and relevance (say the 'Y' PCT category) of my searches, as of the 1990's, references in this journal (and I suspect similar figures for other IEEE and ACM journals) averaged over two dozen references. Further the averages over the last twenty years reflect the general growth in what has been published about software. Keep in mind that as of the 1990's, there are probably over 500,000 software prior art items (articles, theses, books, product literature) which for about 50 software categories means that there is lots of potential non-patent prior art. That the IEEE, myself and other professional searchers find many dozens of non-patent prior art items is both mutually-consistent and statistically likely (if one could determine the ideal distributions of non-patent prior art for various software applications). So that's the 24 number. The 48 number is when you are being methodical (an infringement defense search or a retrospective journal article), and is closer to the average size of the total prior art that should be considered by an examiner when doing the patent novelty analysis under ideal conditions. Yet during this time period, the average number of non-patent prior art references found by PTO examiners and searchers ranged from less than one to the 2.1 figure for 1994. Worse is that half of all issued 1994 US software patents cited no non-patent prior art on the front page, and some patents cited no prior art of any form. For example, IBM got a patent in March of 1994, 5,299,123, appropriately titled "Method for allowing retrieval of documents with user defined search descriptors" that cited no prior art of any form on the front page. We are supposed to believe that there is no prior art for this database searching concept???? That many broad and/or trivial software patents have been issued should be no surprise - the examiners just didn't have access to relevant prior art to make decisions. Combined with access to prior patents, I estimate that on the average software patent, the average examiner has 20% of the relevant prior art. This is inadequate and unfair to applicants and the industry alike. For example, consider the PCT search results for Compton's multimedia patent that has been rejected after reexamination by the PTO (where the examiner cited dozens and dozens of references - a number in the range of the IEEE and my figures), done in 1991: "Visual Information Retrieval: Browsing Strategies in Pictorial Databases", 12th International Online Information Meeting, 1988 "Video Disc Dictionary System" IBMTDB, 25 8, 1983 "Integrating Optical Videodisc and CD/ROM Technology to Teach Art History", J. Educational Technology Systems, 1988 "The New Electronic Encyclopedia", Reference Books Bulletin, 1989 "Toward the Development of Interfaces for Untrained Users" Proc. of the 48th ASIS Annual Meeting, 1985 "Where the Heck Am I", Incider, October 1989 That's it. Entire IEEE and ACM conferences missed, a variety of academic research efforts missed, and a scattering of journal articles and books. And this search is actually one of the better ones, in that it found 6 non-patent prior art items, 6 being greater than .66, 1.2, and 2.0. One could even argue that it is impossible for examiners to do decent software non-patent prior art searches. First, the PTO pays for commercial access to databases such as Dialog (where plenty of software prior art can be found). However, as many studies show, searchers react differently when they do searches on free access databases and for-fee access databases, no matter what permissions they are given. Why do PTO examiners do a great job finding prior US patents? Their access to APS is for free, so they can sit at the APS workstations and search at their leisure, no clocks ticking in their minds. But even if access to Dialog and other databases was free, there is a second problem. A decent PCT search category 'A' level reference (at least somewhat related) done on Dialog (for example the INSPEC database) would on the average return about 100 hits. For myself, I would then goto a local university library (such as MIT or Stanford, which admittedly are two of the best in the country) and look at the 100 hits to find the 24 to 48 I would then cite to my client. But I have to look at the entire article to make such a decision - an abstract is useless. But given the small numbers of books, journals and CDROMs available in Crystal City to the PTO software examiners and searchers, they have to order copies of these papers from one of the document delivery services. At an average cost of $10, this means that an examiner would have to spend $1000 to obtain the materials in which to find the relevant prior art to cite to the applicant. Given that $1000 is greater than the US patent application fee, and the PCT search fee, I doubt highly this happens very often, and that this is one more psychological block affecting an examiner's searching style. Why waste time doing a search if you can't examine the search results? Given that a larger scale statistical analysis of issued US software patents, file wrappers, PCT applications and IEEE/ACM journal histories will not return numbers too different from .66, 1.2, 2.0, 2.1, 24 and 48, there is a good chance that for over 75% of all issued software patents around the world, that an insufficient novelty analysis was performed, and that the value of such patents acquired at a cost of many thousands of dollars is questionable, if not marginal. At an average cost of over $5,000 dollars to acquire a US software patent, the 10,000 software patents issued in 1994 and 1995 represent an economic activity between $50,000,000 and $100,000,000. Can we not demand that non-patent prior art capabilities be more reflective of these large amounts of money? Given the importance of software to the economy, should not Congress first stop dipping into the PTO's till, and second should not Congress provide extra money for special problems like software prior art? ==================== I argue that the software prior art searching and examining capabilities of the Patent offices around the world are fundamentally inadequate, and that applicants in the software industry (except for the big patent factories) are being misled as to the nature of what they are getting for their many thousands of dollars in application fees. All that results are excuses for nuisance infringement lawsuits, gimmicky publicity campaigns, and lousy grounds for the CAFC to make precedent decisions (for example, the Alappat patent should have never made it to the CAFC, as there is ample prior art to have rejected the patent on novelty grounds). I continue to gather such statistics as I expand my prior art databases, but I will be glad to bet that the numbers won't change much (and it should make you wonder how others claiming to be involved with software prior art operate without gathering such numbers themselves). Unfortunately, nothing being done by Patent offices around the world now will do much to change these non-patent software prior art statistics. Partly because they aren't being given the financial resources, partly because they haven't been examining their own records to gather the type of statistics one needs to plan for the future. For the next few years, the torrent of questionable software patents will continue to gush forth. A sample of that torrent tomorrow. Greg Aharonian Internet Patent News Service P.O. Box 404, Belmont, MA, 02178 617-489-3727, patents@world.std.com (for info on free subscription, send 'help' to patents@world.std.com ) (for prior art search services info, send 'prior' to patents@world.std.com ) (for WWW patent searching, try http://sunsite.unc.edu/patents/intropat.html )