High-Level Connectionist Models

John Barnden, (1986, 1987, 1988), New Mexico State University (NMSU), described the current status and future plans for his CONPOSIT system. Implemented on NASA’s Massively Parallel Processor, the system makes use of a two-dimensional array of registers constructed of simple neuronlike processors; it can represent complex symbolic structures using adjacency between registers as a form of variable binding. Barnden demonstrated how production rules are currently implemented and timed and proposed how a large set of such rules could be managed. He also showed that his system could reason syllogistically, directly implementing Johnson-Laird’s notion of a mental model (JohnsonLaird and Bara 1984). Lawrence Birnbaum, Yale University, spoke about hard problems in planning and understanding that would not “just go away” for connectionists. He gave several examples of abstract schemata that require long chains of inferences in order to be recognized and situations which require such recognition in order to be understood (Birnbaum 1986). One example, a variant of one of Eugene Charniak’s (1983) marker-passing stories, involved a detective following a suspect into a hardware store. After being spotted by he purpose of this workshop was to bring together a small group of computer scientists to focus on the interaction between AI and connectionism. The two fields are often posed as paradigmatic enemies, and a risk of severing them exists. (See Dreyfus and Dreyfus 1988 and other papers in the recent issue of Daedalus.) Few connectionist results are published in the mainstream AI journals and conference proceedings other than those sponsored by the Cognitive Science Society, and many neural-network researchers and industrialists proceed without consideration of the problems (and progress) of AI. The workshop was assembled by invitation, and we strove for both geographic and ideological breadth. The questions posed to the participants involved the application of connectionism to traditionally symbolic tasks, approaches to overcoming some of the limitations exhibited by connectionist models of the past several years, and opportunities for connectionism to make contributions to the study of cognition. The participants were Wendy Lehnert, Michael Dyer, and James Hendler (testers of the connectionist waters); Garrison Cottrell and Jeffrey Elman (representatives of the hard-line Parallel Distributed Processing school); Lokendra Shastri and Joachim Diederich (the Feldman school); Mark Derthick and David Touretzky (the Hinton school); Lawrence Bookman (the Waltz school); and Lawrence Birnbaum (the loyal opposition). In addition, Jordan Pollack, John Barnden, and Roger Schvaneveldt represented their own school, New Mexico State University. Participants were given exactly 1 hour to present their views and manA workshop on high-level connectionist models was held in Las Cruces, New Mexico, on 9–11 April 1988 with support from the American Association for Artificial Intelligence and the Office of Naval Research. John Barnden and Jordan Pollack organized and hosted the workshop and will edit a book containing the proceedings and commentary. The book will be published by Ablex as the first volume in a series entitled Advances in Connectionist and Neural Computation Theory. T