Web Scale Reasoning and the LarKC Project (Review and Progress)

"Scaling automated reasoning can be approached from the bottom up, but speeding up extremely simple reasoning over large numbers of data, or from the top down, but harnessing powerful rules and complex inference to broaden the range of problems that data can address. In this talk, both approaches are discussed. For the first, the EU-Funded LarKC project, of which Dr Witbrock is technical director, is described. In LarKC, simple, semantic web inference over RDF is supported, but its reach is extended with parallelism, application of cognitively-inspired algorithms, stream reasoning, and the ability to experiment with novel, even non-logical, inference mechanisms. For the second approach, the Cyc system is described. Cyc, produced by Cycorp in the US and Slovenia, supports first order and higher inference over very large rule sets and highly heterogeneous knowledge bases. It also has a highly developed mechanism for NL query and knowledge capture. By combining the "deep and diverse" approach of Cyc with the "shallow but broad" approach of LarKC, it may be possible to scale automated reasoning towards genuine AI".