7.2. AI People Can Use

Meanwhile, back at the AI ranch, the vendors who sold all these tools couldn't help noticing that their financial customers were less than happy. Some introspection was in order. Why was this happening? The answer was being repeated over and over again at the proliferation of AI conferences held around the country. These systems were too isolated and too hard to use. You might have three different flavors of truth maintenance to choose from, but simple communications and mathematical functions were hopelessly tedious and complex.

In mid-1987, Dale Prouty and I founded a small firm, Integrated Analytics, and tried to solve these problems by designing an expert system tool specialized for trading. This was called MarketMind. The top-level requirements for the design were:

  • Detect the kinds of patterns and events in the market that traders look for themselves, but do this for hundreds or thousands of securities simultaneously.

  • Be a record player, not a record; support a language that expresses a wide variety of trading techniques.

  • Work right out of the box—connect to market data feed without any systems work.

  • Support real-time calculations using real-time and historical market data.

  • Use quantitative logic. Be able to distinguish between barely true and whoppingly true.

  • Run on standard hardware, with no nasty waits for garbage collection.

  • Provide a standard, simple user interface appropriate to trading problems.

  • Use only as much AI as the users can digest. Hide it, ...

Get Nerds on Wall Street: Math, Machines, and Wired Markets now with the O’Reilly learning platform.

O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.