The complex cognitive systems manifesto

Abstract

In a complex system the overall behavior of the system cannot be analytically explained in terms of the underlying mechanism that causes the behavior. This paper argues that the human cognitive system is almost certainly a partial complex system, and that one consequence of this complexity is that if we try to understand human cognition by looking only for the locally-most-optimal models of all aspects of the system, we will generate models that can never converge on a unified theory. This has serious implications for the methodology of cognitive science. To solve this “complex systems problem,” it is proposed that researchers move toward a new, more theoretically intensive research paradigm that shifts the focus away from local models and toward parameterized “generators” of large sets of models. These generators would then be organized using frameworks, each of which is a prototype of a unified theory of cognition, and the research methodology would involve constraint relaxation among the generated models. The paper concludes with a description of a specific framework, based on a generalized version of connectionism, and the suggestion that this new methodology can only be realized if a new class of software tools is built to support it.