Nerds on Wall Street: Math, Machines, and Wired Markets

7.12. Future Plans for AI in Finance (in 1995)

Learning is an important aspect of both real and artificial intelligence. Many AI systems, however, do not learn; they merely mimic what we might consider intelligent behavior. MarketMind and QuantEx fall into this category.

Suppose, for example, you have almost found the holy grail of trading systems: the magic formula that makes money every time. But due to a programming error, your QuantEx system buys when it should sell and vice versa, and you go broke. QuantEx will never notice.

However, this is not true for all AI systems. There are techniques that will allow construction of systems that will notice this kind of error, and learn how to correct it. They can also notice much more subtle errors and deficiencies and make appropriate adjustments. They can also use the results of these adjustments to refine or restructure the process still further. When people do this, we call it "learning." These AI systems are much more modest than the grand AI ultimate goal of machine sentience. They are typically the extended applications of quantitative techniques, enhanced by machine learning.

Artificial learning systems are now being applied in finance and investment. Applications include asset allocation, quantitative equity portfolio management, market making, and currency trading.

That was 1995. To hear how some of this turned out, keep reading the next chapter.