Commercial artificial intelligence software suppliers have faced hard times recently. The 1980s explosion of interest in the technology – fuelled by scare stories about Japanese plans for world domination via its Fifth Generation project – led to such public displays of faith as the UKP250m UK Government-sponsored Alvey programme. In its heyday in the early 1980s you had to be the first user on the block with your very own team of knowledge engineers, forward and backward chaining like billy-o, changing the very nature of data processing with the application of symbolic oodjmaflip. This generation of commercial artificial intelligence was based on successful lab systems mostly developed at Stanford University in the 1970s, when the technology revitalised itself by tackling domains like spectroscopy analysis where there was a well-defined body of specialist encodable knowledge. Alas, so-called knowledgebased systems, though useful and the basis of several working products, proved not to be the World’s Best Mousetrap: at the high end, a market for essentially untried software was quickly overexpanded by credulity-stretching hype; worse, you had to buy dedicated, expensive hardware to run what were at the end of the day stand-alone systems; and at the low end, limited functionality MS-DOS shells proved too fragile to deploy working applications.

Neural networks

Artificial intelligence went very quiet. Then we had a new technique – neural networks, which, as hundreds of TV science programmes never tired of telling us, worked not by emulating human intelligence but by learning the same way insects do. The problem with this approach was that no-one – sometimes not even the computers themselves – could work out what the hell was going on witness the system that seemed a whiz at picking out which aerial photographs had hidden tanks in them until in one final trial they tried it again on a sunny day when all the pictures with tanks in them had been taken on a cloudy day – and suddenly it saw no tanks. Now we have a new artificial intelligence white knight: Case Based Reasoning. Nothing to do with software engineering with the redundant computer-aided on the front, Case Based Reasoning is based on the idea that people solve problems by matching current states with past cases until they find a like enough match in their experience, rather than deducing from first principles. Knowledge based systems use rules and if given a problem that falls outside the scope of the rule or knowledge base, can’t cope: such systems don’t learn from mistakes, as it were. Case Based Reasoning systems, on the other hand, search their memory of cases to find the nearest match in their experience: if they find a very close match, fine, we’ve got a solution; otherwise a human helps out by modifying the offered, partial, solution, which the machine adds to its knowledge base, hence learning something useful.

By Gary Flood

Case Based Reasoning has been around in one form or another since artificial intelligence researcher Roger Schank outlined the approach in 1982, but only now are commercial products appearing, though trial systems have been developed for large users such as the US National Aeronautics & Space Administration, Martin Marietta Corp, GTE Corp Digital Equipment Corp, and Boeing Co. The technology has been hailed as the latest hot artificial intelligence topic, with one exemplar product, CBR Express from El Segundo, California-based Inference Corp. Case Based Reasoning Express, an MS-DOS-based product, has so far been aimed primarily at help desk applications. Of the four biggest surviving artificial intelligence companies Inference, Intellicorp Inc of Mountain View, California, AICorp of Waltham, Massachusetts and Aion Corp of Palo Alto, California – Inference was the only one with a Case Based Reasoning answer and a current product. Now Aion, best known for its IBM mainframe Aion Development System, marketed in the UK by Hitchin based Software Generation Ltd, is catching up, announcing a technology and marketing deal with a comp

any Schank set up, Cognitive Systems Inc, of New Haven, Connecticut, and its ReMind Case Based Reasoning system. ReMind is the fruit of a three year $4.3m research contract from the US Government’s Defence Advanced Research Projects Agency awarded in 1987. Written in C++, running on Macintosh, MS-DOS with Windows 3.0 or OS/2 Presentation Manager with at least 3Mb RAM, the product is made up of 11 different tools to enable end users to represent, index, retrieve and analyse cases. Steve Mott, Cognitive’s president, claims ReMind offers greater functionality than the Inference offering through a wider range of search algorithms CBR Express uses the nearest neighbour only, while Re-Mind includes induction and prototype indexing capabilities. ReMind is effectively a Case Based Reasoning shell, with no user front end, unlike CBR Express, which is targetted as a self-contained MS-DOS application along the lines of Lotus Development Corp’s 1-2-3. Anyone thinking of taking on the ReMind system would have to consider integration and development of a front end, but this doesn’t bother Mott, who says trial users such as American Express, Nestle, Barclays and British Airways are integrating ReMind in the form of delivered libraries of C code into existing knowledge based systems. The UK and European market is about a year and a half behind the US right now, but one thing in common is that artificial intelligence isn’t such hot news right now. But everyone can understand Case Based Reasoning, and we’re not selling it as heavy duty artificial intelligence but as a solution, says Mott, who adds that the company carefully scrutinised the remaining artificial intelligence players and selected Aion as the most commercially sound. Sometime this year – no timetable is given – the Aion Development System and ReMind will be integrated. Some interesting technological hurdles remain to be climbed, mind you – Re-Mind is written in C++ so cannot run in native mode on the MVS mainframe which is home to the Aion System.

Mainframe C++

ReMind’s development machine is the Apple Macintosh – even further removed! – but by using Colorado based XVT Corp’s application programming interface tool the company has been able to implement the system successfully for other environments, so he doesn’t anticipate delivery problems: Someone with a C++ compiler on the mainframe would clean up right now, he says. The advantage to Cognitive of a link with Aion is clear: access to a claimed customer base of 700 companies worldwide, with a multi-user, mainframebased expert system offering object-oriented programming relatively successful in banking, insurance and finance – Norwich Union is one enthusiastic UK user. If Case Based Reasoning takes off, we can expect the next artificial intelligence technology on the horizon – genetic algorithms – will become the next hot topic, if only for TV researchers beavering away on a Horizon meets the computer industry special.