KMWorld 2023 sees a sea change
Two terms, “generative AI” and “hallucinations,” dominated the 2023 KMWorld Conference (kmworld.com/conference). They arrived like Venus arising from the sea. Both expressions were entirely absent from previous KMWorld conferences. Their sheer novelty and dominance heralded 2023’s degree of disconnect with all previous KMWorld conferences. Also quite notable, but not so novel, was the futurity of the presentations, which overwhelmingly emphasized the potential and possible pitfalls of generative AI, specifically Chat GPT. It is the combination of these factors that drives the conclusion that foreseeing a “sea change” is not at all out of place or overstated.
It is quite remarkable that a capability that appeared only 1 year ago—and was previously almost unknown—has had such a broad and dramatic impact so quickly. Remember, OpenAI released ChatGPT only 3 weeks after KMWorld 2022 ended.
Most of the papers presented at the 2023 conference did not report on what had changed. Instead, they assumed and predicted that there would be substantial change moving forward. To a modest degree, they speculated on what that change would consist of and what consequences would follow.
In future years, someone looking at a collection of previous KMWorld conference programs, but with all dates removed, would have no difficulty in recognizing the 2023 program and the seminal moment it represented. It is pretty safe to predict that 2023 will long be seen as a, or perhaps the, watershed year for the KM community. Perhaps KMWorld 2023 should be dubbed “the Chat-GPT KMWorld.”
Search on the upswing
An area of consensus was that search methodology, having been in the doldrums for a while, is becoming more powerful, thanks to generative AI (GenAI) and related technologies. A long dormant consideration in the KM community, now reappearing in the context of enterprise search, is “vector,” as in Gerry Salton’s long established vector search methodology, which he liked to refer to as the SMART System (Salton’s Magical Analytical Retrieval Technique). This surfaced in discussions of possible uses for GenAI enhanced retrieval, particularly during panels in the Enterprise Search & Discovery conference that was once again colocated with KMWorld.
A topic that appeared in 2022, in a very prescient KMWorld presentation by Tim Powell, was zombie knowledge. Zombie knowledge is material that is dated, or no longer correct or relevant, which clutters up or degrades your KM system. One of the concerns that arose frequently in 2023 was that, with GenAI, the need arises to recognize and correct hallucinatory information— the false and misleading information that it creates and propagates. Such “knowledge” can not only be misleading, but even dangerous. It was seen as obvious that dealing with hallucinations is a nontrivial problem, and that it will require the same techniques that are appropriate for zombie knowledge.
Put another way, GenAI can be expected to enormously expand the domain of and the problem with zombie knowledge. Powell referred to the potential of GenAI to be an industrial-strength disinformation machine, potentially creating reams of zombie data and purported knowledge, which would need to be recognized and dealt with.
Hallucinatory information and knowledge
One theme, consistent across the conference, was the issue of hallucinatory knowledge created by GenAI, what to do about it, how to cope with it, and how big a problem it would become. The realization that hallucinatory information is functionally simply an extension of the concept of zombie knowledge came almost automatically, as did the realization that the problem could be very much larger and more difficult to cope with when driven by near ubiquitous GenAI. Even before KMWorld proper commenced, KAPS Group’s Tom Reamy, in a preconference workshop, focused on the problem of hallucinatory information.