3 ms·
USC researchers use software to mine blogs for cause-and-effect data
- _delirium 17y agoAndrew Gordon has a history of pretty interesting work this area, and is probably one of the best researchers in the overlap of symbolic knowledge representation, data-mining, and pragmatic approaches to commonsense-reasoning problems--- areas that often like to resist overlapping. This sounds like the sort of work he's as well placed as anyone to do. As a commenter on the linked article points out, it's "epistemologically unsound" to mine data in this way, and not even humans really understand causality. And yet, as that comment misses, humans are able to make routine use of casual inferences, even though we don't "really" understand causality, and we acquire our causal information in ways that are almost certainly epistemologically unsound (people do not usually use careful scientific experiments in building their everyday models of how the world works). Even uneducated and very young humans can do stuff along those lines that'd be amazing for AI to be able to do! It'll be interesting to see if approaches like this one can help computers pragmatically acquire similar levels of proficiency about causal relationships, as an alternative to, say, large databases of consistent, carefully hand-authored facts like Cyc. His page: http://people.ict.usc.edu/~gordon/ http://people.ict.usc.edu/~gordon/
- elblanco 17y agoDupe: http://news.ycombinator.com/item?id=1186429 http://news.ycombinator.com/item?id=1186429