I've kept this item in a browser tab for a couple of weeks waiting to give it the proper attention it deserves, and while I never did, I still want to pass it along. There's a whole debate under the surface of the AI revolution on the relevance of knowledge structures from good old fashioned AI (GOFAI) such as ontologies, graphs and namespaces. The suggestion in this article is that new AI (that we see in large language models, for example) will not escape four issues that dogged (and still dog) these activities: open world (multiple 'truths') vs closed world (single 'truth') models; reification (statements about assertions); context graphs and the primacy of time; and the relation between 'authority' and namespaces ("there is no neutral graph"). Many of the issues people have with large language models can be traced directly to these four issues: questions of bias, authority, the meanings of words, and the passage of time create reasons to doubt generative AI, and AI in general. Well, and human knowledge too. Anyhow, this is a great article. Do spend some time with it.
Today: Total: [Direct link] [Share]
