Back to the Basics

Dear Readers,

For many years artificial intelligence has taken primary two directions. The first course was and still is to better understand the dynamics of artificial neural networks, and explore optimizing their structure or accuracy of their application. This has included extensive research in in many of the fields of discrete mathematics, functions of statistical mechanics, differential geometry, and nonlinear regression. The other direction is applied to the more ubiquitous ideal of mimicking human problem solving. To take a step back there are so many application domains that still rely on standard case based specifications, or ad hoc component architectures to build the most basic systems.

If we think of PDA for instance many applications, can store email, lookup relevant news, and get a weather forecast. In the context of AI think of a simple shopping list, many PDAs can store your to do list with out much thought. But if the application developers could only use neural networks and genetic algorithms, things would be allot different. For instance does your shopping list predict what will be on sale next week? Can your email client anticipate the delay of receiving feedback, from an associate that usually takes thee days to get back to you? With neural networks and simple genetic algorithms these applications could exist now.

As a NOLS instructor once informed me the summit is not the only place on the mountain. Building advanced AI shouldn't be our primary goal, it should come secondary to finding basic solutions to the majority of tasks that require so much of our attention. And artificial intelligence need not be used for only the most complicated of questions. It's not that we want smart technology to improve on the past, it's that we need stupid smart technology to improve on the present.