Last week's article featured several large-scale collaborative projects harvesting both networked computer power and human brain power in an effort to build AI on the Web. I compared them with Cyc, an immense, broad, multi-contextual knowledge base developed over the last seventeen years. Maybe it was an unfair comparison - after all, Cyc acquired almost a cult status, with all controversies, hype and secrecy that go with it. However, it turned my attention to the new developments at Cycorp and give me a chance to compare the project status now and more than a year ago.
For an introduction to the project that already consumed an estimated 500 person-years and $50 million in investments, please refer to the earlier article on the same topic. Some numbers have changed - for example, the system today encompasses more than 1.4 million hand-entered assertions - but the approach used to capture the common sense information remains the same. The set of available products and components grew significantly, enabling a variety of knowledge-intensive products and services. Besides the Cyc knowledge base, inference engine and representation language, new additions include natural language processing components, variety of development tools and the semantic integration bus that turns data from structured (databases), semi-structured (spreadsheets, Web pages), and unstructured (text files) sources into usable knowledge.
So far, everything related to this project sounds impressive, but it was almost impossible to try its performance under the real-world conditions. The announcement of the open source version of the Cyc technology put the smile on many faces in the AI world. However, subsequent news were not so good, stating that "in order to ensure a high-quality release, the first release of OpenCyc, originally scheduled for July 1, 2001, will be delayed until late summer 2001.". However, I thought that it is better to get ready on time: Cycorp already set up an independent organization, OpenCyc.org, to disseminate and administer OpenCyc. According to them, the release 1.0 of OpenCyc will offer 6000 concepts - an "upper ontology" for all of human consensus reality - and 60000 assertions about those 6000 concepts. Besides that, it will include a compiled version of the Cyc Inference Engine and the Knowledge Base Browser, a suite of tools for rapidly extracting knowledge from domain experts, a specification of CycL and Cyc API, etc. Some may argue that this is only a crippled demo application - after all, it contains only 5% of the total number of assertions and the source code for the core applications is not available. On the other hand, the contents of the OpenCyc knowledge base (written in CycL ) are open source libraries (LGPL). Also, the FAQ implies that the source code for the inference engine may be released later. Qualified parties can obtain a free license to a substantially larger subset of the Cyc Knowledge Base known as ResearchCyc, which is for R&D; use only. Year by year, each assertion in the latest version of Cyc will migrate to a subsequent release of ResearchCyc, and each assertion in ResearchCyc will migrate to a later release of OpenCyc. Updated versions should be released twice a year.
So, either way, the AI community could get a powerful tool without a hefty price tag. Cycorp hopes that the release of OpenCyc will help lay the groundwork for the massively parallel effort to rapidly grow the its knowledge base, establishing Cyc as a standard for knowledge representation, knowledge management, data base integration and similar uses. It remains to be seen how successful it will be, but I'm sure many of us will download the OpenCyc and test it in various natural language understanding, database integration, Web search, document filtering and similar apps. Stay tuned for more news and developments related to OpenCyc later this year. In the meantime, visit other AI-related projects at SourceForge: you'll be surprised to see so many of them...
