Philosophy Research Papers on Cognitive Architecture

Eric Lormand, University of Michigan

My work on cognitive architecture is about the hypothesis that human cognitive processes employ a language of thought (LOT)--a system of mental representation which supports syntactically complex mental symbols, physically realized in brains.  Connectionism and holism are my focus with regards to this issue.. Some philosophers have thought that inferential holism (and related phenomena) present serious theoretical difficulties for "classical" research programs in artificial intelligence and cognitive psychology—this is the root of the philosophers’ "frame problem"--and propose "connectionist" approaches instead.

List of cognitive architecture papers on this site

Name

Description Date/Status
Classical and Connectionist Models Offers a formulation of the LOT hypothesis which eliminates various obscurities in standard formulations, and discusses some interesting conceptual issues for philosophy of mind presented by connectionism—about what makes something a representation or a computation, about what makes a representation language-like or syntactic, about what makes a process an inference rather than an association, etc.  (73K) Chapter 0 of Classical and Connectionist Models of Cognition, MIT Ph.D. dissertation, 1990
Connectionist Languages of Thought Suggests directions for empirical comparisons of classical LOT models (including many connectionist models) and those connectionist models which do not implement a LOT.  Argues that nonclassical connectionist models are unlikely to succeed as general accounts of cognition, but that they have promise as part of an account of the (alleged) inferential processes guiding skillful activity, which are unconscious, rapid, and holistically sensitive to a vast range of potentially relevant conditions. Argues that Fodor and Pylyshyn's argument against radical connectionism fails, and provides a better argument.   (101K) in preparation, from Chapter 1 of Classical and Connectionist Models of Cognition, MIT Ph.D. dissertation, 1990
Connectionist Content Shows how representations in nonclassical connectionist models, despite having no syntactic or semantic structure, can realize genuinely propositional attitudes (and can therefore undergo genuinely inferential processes). Develops a notion of a "simple proposition" as the content of simple connectionist nodes.  (103K) Chapter 2 of Classical and Connectionist Models of Cognition, MIT Ph.D. dissertation, 1990
The Holorobophobe's Dilemma Argues that classical models can themselves be applied to holistically sensitive inference, in the face of various objections which philosophers have advanced under the name of the "frame problem".  Shows how the frame problem can be disarmed without serious revisions to LOT research programs, and in particular without appeal to "connectionism" as an alternative.  (76K) The Robot's Dilemma Revisited, Ford & Pylyshyn (eds), from Chapter 3 of Classical and Connectionist Models of Cognition, MIT Ph.D. dissertation, 1990
Frame Problem Very brief and introductory overview of the problem and solutions.   (7K) MIT Encyclopedia of Cognitive Science--Frame Problem entry

 

Relations to meaning papers

Some nuances in "Connectionist Content" have been superceded by the papers on meaning holism and meaning atomism, though the basic idea remains.   Also, the main reason that contents are holistically interconnected is that inferential dispositions are holistically interconnected (and that inferential role is a component of content). Another issue about cognitive architecture and inferential holism that relates directly to content is the distinction (if any) between perception and cognition. This is important for distinguishing among various "modules" or "systems" relative to which something is a semantic "atom" (see section 4 of "Atomist").

 

Relations to consciousness papers

The distinction (if any) between perception and cognition is also important for distinguishing inner perception from inner-directed thought (see section 6 of "Stopgap").

 

I hope you enjoy this stuff; please send comments to lormand@umich.edu.