Abstract
Despite the potential importance to cognitive psychology of unified theories no attempt has been made to assess concretely the methodological problems that such theorising produces. This paper addresses this issue of unified theorising, and in particular the arguments for unified theories put forward by Newell (1990). Close examination of these arguments reveals that Newell's approach does not adequately counter the difficulties which beset the grand theories of the 1930s, nor the problems of irrelevant specification which arise in modern computational psychological work. These difficulties do not prevent the development of unified theories, but they do pose serious problems, problems which it is argued can only be met by rigorous empirical testing together with extreme methodological sensitivity. The methodological concerns lead us to examine Soar, perhaps the most well-developed unified theory, from methodological, computational, and empirical perspectives. Our conclusions are that, whilst Soar represents an impressive body of research, its methodological foundations are insecure, it is ill specified as a computational/psychological theory, and under empirical testing it does not stand up to close scrutiny as a unified theory. The Soar research programme as it currently stands thus fails to meet the necessary methodological demands imposed by unified theorising.
Collapse