Situated Adaptive Design: Toward a New Methodology for Knowledge Systems Development

In recent years there has been much criticism of knowledge systems technology and of artificial intelligence (AI) in general. In addition to arguments concerning brittleness and lack of noise and fault tolerance the main thrust of the criticism has been, in essence, that cognition is “situated” while most AI systems are not. “Situated” means that an intelligent system can only be understood in its interaction with the real world in a particular situation in which it has to act. Traditional non-situated AI systems which are strongly based on the notion of models tend to suffer from the notorious frame problem. In the first part the basic criticism and the issues involved are introduced and reasons for the pertinent problems of AI are given. It is then argued that by taking the criticisms seriously we cannot only gain theoretical insights but we can also derive guidelines on how to build better knowledge systems. This will lead to a methodology of “situated adaptive design”. The methodology is illustrated with two successful examples. It is concluded that (a) the application of this methodology promises to resolve some of the basic problems, and (b) that research efforts in the area of knowledge systems should be re-directed.