The mortgage industry needs applied artificial intelligence, argues Paul Hunt, managing director of Phoebus Software
John McCarthy, the man who laid the foundations of artificial intelligence, stated before his death that he was saddened by the course of the development of computer science. He said the utility of programmes could have been far greater if more time and effort had been spent on ‘representation and reasoning’ rather than simply the pursuit of ‘high speed search on very large databases’.
This chimes with a similar and rather surprising statement by Alan Turing in 1950. The great codebreaker claimed computers were fast enough already and that the greatest problem to solve was ‘mainly one of programming’. When one considers how computing has developed, it’s astonishing the words of these two denizens of the science have been pretty well ignored in the path taken by computing technology in the last 60 years.
In October, I said Moore’s law, the principle that computational speed will increase exponentially, is ultimately bound to fail. Given the performance limitations on the computers of the 1950s, it’s hard to argue there was no good reason to concentrate on increasing computing speed. But the rush to comply with Moore’s law may well have denied the world the advances Turing and McCarthy thought possible by the year 2000.
One of these advances was that computers should be able to operate so they are indistinguishable from humans. The Loebner prize offers a prize of £100,000 to the first programme to achieve this. Today, much of the mainstream AI community has boycotted the prize, claiming it’s focus on the Turing test is pointless. But the boycott probably also has more to do with the fact that 61 years after Turing predicted a machine would pass his test, nobody has managed to build anything that comes even close. It’s an embarrassment. So far are we from even a semblance of artificial intelligence that many developers aren’t even willing to take on the challenge.
Until developers actively engage with the challenge posed by the Turing test, the development of computers will be shackled. Trying only to improve the speed and performance of computers is rather like racing a drag car at Monaco. Cars designed purely for straight-line speed have their uses and are certainly impressive to observe, but outside a very limited window of performance, they are almost completely useless. What makes useful cars from mechanical curios are the subtleties of design that go beyond how many bhp they can produce. The same should apply to computer systems.
In the sphere of financial services systems, Phoebus strives constantly to apply this subtlety of design. Regardless of the power of a server, companies require programming which maximises efficiency and utility. The new requirements of regulatory compliance mean more pressure than ever before is on financial services systems to operate intelligently. TCF requires servicers to identify and inform borrowers likely to fall into difficulty with payments as soon as possible. Doing this minimises the risk of long term arrears and, ultimately, repossession. But it requires a system which is able to look at borrowers as individuals and assess the complex financial risks they each face. Meeting this challenge requires not an improvement in computational power, but complex and innovative programming.
It’s unlikely Professor McCarthy would have picked the mortgage industry as a prime first example of the need for applied AI, but the types of programming advances that are his legacy are in fact essential to the systems the industry uses today. Although the passing of the pioneer of AI may seem like confirmation of the folly of trying to create intelligent programs, doing so is now more important than ever.
