AI research seeking new ways

But before some brokers had managed to throw themsleves out of the window, the prices began to recover. The direct causes of ‘The Flash Crash’ were not to be found in humans, but with computers.

The state commission of inquiry, appointed in response to the price collapse, concluded that the culprit was a major player, the algorithm-driven programs had acted with a share over 20 minutes. It was enough to trigger other algorithms in what became a chain of increasingly bizarre transactions, in which Accenture shares were sold for a penny, while Apple shares went for $ 100,000 each (although this could be cancelled).

In recent decades, financial markets are increasingly computerised. In the latest issue of Wired magazine Harold Bradley describes how he, as a manager at a medium-sized investment company in Kansas, devoted several years of the 1990s to creating the most effective computer programs for trading purposes.

Bradley, however, wanted to go further than combining the factors he understood through his experience that govern a stock’s development. In amongst others, like Jim Simons and David E. Shaw’s imitations, his goal was to create a system that could analyse the movement of shares in a fundamentally different – and smarter – way than what humans are capable of. Therefore, he gave the task to AI (artificial intelligence) specialists and financial analysts to develop an algorithmic system, built around the seven most important variables for the prediction of an equity portfolio’s development.

This program was then run against the historical development of ten shares that were randomly chosen from amongst the most lucrative. It then weighs the variables, and these weights are then fine-tuned by additional runs with new selections of shares. On the basis of the ten weights which have shown themselves most effective, 100 new weights are then distilled for new shares, so that the ten most optimal directions could be distinguished, which were to be the basis of 100 new weightings, from which a new selection was made, and so on.

During the 2000s, a number of similar algorithm-based programs took over all major sectors of the financial market. The algorithms are typical of how artificial intelligence is used today. Not to mimic human thought, but rather to approach the market from a mechanical perspective. In much the same way that one, with the help of algorithms, observes the weather system: by creating huge, intricate patterns on the basis of incalculable amounts of data, whose fluctuations are instantly analysed – in which a nano-second delay in the financial market can be the difference between buying and selling.

For modern AI research’s part, the circle seems to have been closed. In the case of computer pioneer Alan Turing it developed in order to crack codes during World War II, then there was the AI research during the 1960s and 1970s in the hope of creating artificial human brains. Subsequently it has been, from the 1980s onwards, more and more concentrated on optimising different advanced mechanical and technological processes, not necessarily modeled on the human intelligence. This has in turn revolutionised the development of search engines, cars, credit cards and medical research.

Perhaps this process, which can be interpreted as AI research, has retreated from the kind of mechanistic and instrumental view of people, which often seem to have been a prerequisite for the visions that artificially simulate human behaviour and thought. Paradoxically, it has been during the historical periods in which the human individual has been at the centre that interest in artificially constructed human beings has skyrocketed.

Not in vain were the first experiments with human robots that took place during the Renaissance, and which were typically performed by one of humanism’s cardinal figures, Leonardo da Vinci. But the field that is artificially constructed people did not really accelerate until the advent of the modern, bourgeois individual. During the Enlightenment, La Mettrie, following Descartes, posited man as a machine; similarly Voltaire, in Philosophical Letter, the soul of a watch; and Jacquet-Droz created tangible human machines.

The reason was of course that a scientific mechanical image of man had replaced an essentialiy religious one. Thus was born in earnest the interest in intelligence of which this issue of Axess Magazine is an offshoot. But that does not mean that religion was obsolete. As a corrective to an exaggeration of empirical reality, religion would, in the existential and the scientific context, prove to have a bright future. Not least in the sense that its reverence for that which is specifically human, was much more tenacious than one might have believed possible in the 1770s and 1970s.