close this bookVolume 6: No. 63
View the documentFunding news
View the documentInternet news and e-publishing
View the documentCareer jobs (in our CCJ digest this week)
View the documentJournal calls
View the documentGreat ideas in AI participants have been asking what are the 20 great ideas of AI (or at least stolen by AI). Some suggestions: Machines can manipulate symbols as well as numbers. (Newell, Simon) Programs are data. (Von Neumann?) Intelligence is a matter of effect, not media. (Turing) Learning can be achieved via non-symbolic adaptation, as in clustering or neural networks. Symbol manipulating may be sufficient for intelligence. (Newell, Simon) Predicate calculus, with extensions, can be used to describe real-world reasoning. (McCarthy) Symbolic discovery can be automated. Search is important. Avoiding search is also important, and heuristics are a good way to avoid search. Knowledge (in databases, inference links, rules, frames, scripts) is always a good way to avoid search. (Feigenbaum) Choosing a good representation is important. (E.g., object-oriented programming.) Before you can discover something, you must be capable of being taught it. (McCarthy) Large spaces can often be searched efficiently via genetic selection. (Holland) Real-world reasoning requires imprecise concepts. (Zadeh) Intelligence springs from self-organizing systems or societies of agents. (Selfridge, Minsky) Intelligent systems have to deal with intractable problems. (Cook) [Matthew Ginsberg, Tim Finin, William Grosso, Philip Jackson, et al.,, 9/11/96.] (I would add that rapid prototyping is a key to conceptual breakthroughs, as well as a great way to sell research to sponsors. It doesn't necessarily sell systems to corporate clients, though. Also, analogy and case-based reasoning are important, even though we don't know how to do them in general yet. Recognition is often the most difficult part of reasoning.)

Sean Luke came up with some practical suggestions, softened by smiley faces. Paraphrasing: 1) An "important first step" toward an AI-complete goal is the way to fame. 2) Develop a toy application domain to make your abstract idea look good. 3) Intelligence is defined by theory, as that which is just beyond our implementation capability. 4) Being outspoken is more important than being right. [,, 9/11/96.]

-- Ken