[jstrebel]

I see this article as a piece of its time - it tries to bridge the divide between connectionists and symbolists. And that's exactly my problem with it: of course, you need a non-symbolic approach to sensor data processing, fusion, aggregation, category discovery etc. but it does not necessarily mean that a neural network is the only option. In the end, an ANN implements a mathematical function and as such, you as the system designer should be able to choose any mathematical framework for your lowest level.