Knowledge Technologies 2001: Conference Diary

This article is a report from the first day of the conference.

Opening Sessions

Opening keynote sessions included Doug Lenat from Cycorp. Doug has gone against the flow where artificial intelligence is concerned. Twenty years ago, when others were gung-ho for AI, Lenat was a pessimist. As disillusionment has set in over recent years, Lenat reports he is now an optimist.

A lot of this good feeling comes from the work he's done with CYC (pronounced "psyche"). Lenat has been steadily feeding his system facts about the world for 15 years, and reports that it's starting to get to the stage where the system can help with its own development. CYC uses a codification of natural language into a formal logical language.

One interesting insight that Lenat reported from 15 years of work is that maintaining global consistency in the system was impossible: instead they aim for local consistency. He gave the example of Dracula. If asked "Who is Dracula?" you might respond "a vampire," but if asked "Are there such things as vampires?" you'd probably say "no." On a global scale, these two things are contradictions, but inside the contexts of each respective question they are true. You can imagine that over the Web, there will be many of these contradictions.

I must admit to feeling a certain degree of skepticism at this point, but this was assuaged to an extent by Lenat's description of two of Cycorp's products. One of these, CycSecure, uses CYC's ability to integrate diverse information sources to analyze the security of computer systems. It is able to discover viable multi-step attacks, not otherwise easily deduced.

Lenat concluded his talk with an announcement that CYCORP intends to progressively make elements of CYC publicly available under the banner of "OpenCYC", starting in the summer of this year, including exports of the knowledge base in DAML/XML form. Parts of the CYC system will be open sourced under the GPL.

The second plenary session in the morning featured updates from the various standards organizations on the state of their work in the knowledge management space. These didn't differ greatly from similar presentations at XML 2000 last year. Dan Connolly of the W3C focused on the new Semantic Web Activity, kicked off at the end of February this year. He explained the creation of the new RDF Core Working group to perform a similar function to the XML Core Working group -- handling errata and corner cases to the basic RDF specification and coordinating between other groups.

Anyone hoping for radical news from this session would have been disappointed, and it's not too surprising. As a whole on the Web, the area of knowledge representation is very young -- anyone writing standards rather than code right now is probably premature. Dan Connolly observed with respect to RDF 1.0 that there was "too much design in the abstract", and that the new Semantic Web Activity would devote more resources to "building stuff."

Lunchtime Insights

At conferences such as these, it's often who you sit next to at lunch that leads to some of the most interesting conversations and insights into the work of various developers and organizations. Today was no exception, and I found myself seated with XML.com author and O'Reilly editor Simon St.Laurent going head-to-head with Dan Connolly from the W3C. Anyone who thinks that the rages of XML-DEV can't be reproduced in the flesh should have been there.

Beyond the flying sparks, the lunch gave some interesting insights into the workings of the W3C and the people behind the organization. One encouraging sign was that both Connolly, the W3C's XML lead, and Eric Miller, the new lead for the Semantic Web activity, were strongly in favor of as much openness as possible -- an encouraging sign for future development of at least Semantic Web related technology at the W3C.

Real World RDF

Afternoon sessions featured presentations from Dave Beckett of ILRT, Eric van der Vlist of XMLfr.org, and Uche Ogbuji from FourThought. These three all have extensive experience of development and deployment of RDF-based tools.

Beckett gave an introduction to RDF. He outlined the now rich variety of tools available for working with it. Van der Vlist gave a talk entitled "Semantic Web Sites", and described how he used RSS 1.0 on XMLfr.org to categorize and cross-relate items of content. He reported that adding the ability to present related content with news items on his site significantly increased traffic. Van der Vlist then went on to show how Topic Maps could be generated about topics from the web site captured in the categorization (such as a person or an organization).

Uche Ogbuji of FourThought talked about their OpenTechnology.org, a community driven web site that uses an underlying RDF implementation, based on their 4Suite server code. He talked in depth about the implementation problems he'd come acrossd and presented honestly on the difficulties with using RDF as well as the advantages he found. Ogbuji's presentation was a great example of the value of more niche conferences such as Knowledge Technologies -- you can hear implementors talk in depth about their discoveries without having to worry about looking good marketing-wise.