The most commonly used form of artificial intelligence is neural networking, software that operates in ways that mimic the human brain. But where the software scores is, unlike a human, it can work from hundreds of pieces of data simultaneously.
Credit-reference agencies have used neural networking for 10 years or so. Their databases store information and patterns of behaviour of people and corporations prior to insolvency. The software can then pick up warning signs that predict bankruptcy.
Bradford University is one of the country's leading bodies working on neural networks. Nick Wilson, its professor of credit management, says: "You can monitor a huge range of variables."
To predict a company's impending financial crisis can involve examining a range of factors that might not otherwise be obvious warning signs, says Professor Wilson. These might include changing the auditor, increased audit fees, late filing of accounts, qualifications of accounts, and any resignations of directors, as well as more usual factors such as liquidity ratios, gearing, speed of payment of bills, and profit levels.
Even an increase in turnover may be a warning indicator, as organisations in distress may sell fixed assets as a way out of immediate crisis, while reducing a business's viability in the longer term.
It is only in the past few months that the potential of neural networking has become clear. Improvements in computer processing have allowed sums that once took a full day to take just a minute.
Tom O'Brien, a partner in Andersen Consulting, says: "The biggest application is in the financial services arena, especially for brokers in futures and options, looking at spreads and prices." Auditors might use neural networks to better pick out fraud, distress signals and weak performance. Insurers might use neural computing to improve their marketing, highlighting customers who are covered for car and household insurance, but not for life cover. Banks are adopting the systems to improve their mail shots, to eradicate the practice of encouraging customers to apply for loans that would actually be refused.
"You can use it to match customers to products, and predict demand," says Mr O'Brien. He says it could also prevent poor service, anticipate customer complaints, and apologise for service problems.
A number of banks are looking neural networks for fund management. It may be used alongside share-tracking programmes so that computer software would not simply follow the market, but could predict it as well. But neural networks have limitations in their effectiveness. While they can predict bankruptcy, they are unable to justify their predictions. This is a drawback for credit-reference agencies the lenders who use them.
One solution is to use other forms of artificial intelligence alongside neural networks. So-called "fuzzy logic" can back-up the results of neural networking. And "genetic algorithms" can explain the outcome to outsiders, by examining the results and relating them to agreed criteria.
SearchSpace, which is providing the Stock Exchange's software, is running the three models alongside each other. Konrad Feldman, a consultant with SearchSpace, says: "Genetic algorithms can produce rules which are transparent."
But however good the software is, it does not in cure the problem of illegal share dealing, and may not provide the proof needed for conviction. "Prosecution is not something we are involved in. The burden of proof is a different matter," says Mr Feldman. "We can provide the reasons behind something suspicious, and give the mitigating circumstances."