New Scientist is reporting that Ramesh Sharda at Oklahoma State University has developed a practical neural network (artificial intelligence) program that 75% of the time can successfully predict, within a fair degree of accuracy, a film’s theater receipts.

The interesting thing about this to me is that Sharda has apparently gotten the neural net to identify seven key parameters that reliably predict film receipts. He fed a large database of hundreds of films into the neural net, effectively “training it” (which is how neural nets work), presumably with initially many more than seven parameters for each film.

While being able to predict a film’s success would certainly be useful information, simply being able to identify a limited set of key parameters for a given activity could itself be highly worthwhile. For example, say a restaurant chain such as McDonald’s wants to improve their real estate selections. Presumably they would have a large database of sales information that they could plug into a neural net, which might then be able to pinpoint which parameters are truly key (e.g. traffic count, demographics, proximity to other restaurants, etc.).

If certain parameters over which the company had control could be found that would yield substantial improvements, then this might better guide improvements to present locations (or processes, products, etc.), while the neural net could be used to analyze future potential locations.

In general, the state of the artificial intelligence field seems pretty obscure nowadays. I’ve been wondering lately whether many practical applications are finally taking hold, and whether these involve neural net technology or some other approach. In the mid-80s, languages such as Prolog and Lisp were being touted as tools for building elaborate knowledge bases, etc., but I wonder if anyone is still working with these.

In the second half of the 80s, neural nets seemed to pretty much eclipse language-based artificial intelligence. Way back then I took a serious look at Prolog, which the Japanese had favored for their research. The notion of declarative programming still fascinates me, but without a background in predicate logic or computer science I didn’t have much confidence in my assessment. For one thing, I never could see how one would go about debugging a Prolog program!