In the 1980s, artificial intelligence was all the rage in computer software. The most promising thing in AI software was knowledge-based, or expert, systems, which can mimic human expertise to help solve problems and offer advice in complex matters such as trading currencies. Trouble is, building such systems requires lengthy interviews with human experts, who often can't explain their thinking in terms of the strict if-A-then-B logic rules that expert systems use.
Now, an expert-systems pioneer, Inference Corp. in El Segundo, Calif., has what it says is a simpler way to build expert systems. A new version of its ART-IM program for engineering workstations and IBM mainframes uses a method called case-based reasoning to make "expert" decisions. First, ART-IM is loaded with descriptions of hundreds or even thousands of cases, such as the symptoms of broken VCRs and the remedies used to fix them. After that, whenever a VCR problem crops up, ART-IM can search through its cases, find similar ones, and propose likely solutions. Inference says this capability eases the way toward the kind of full-blown, rule-based expert systems that it set out to popularize 12 years ago.