David Rumelhart
(1942-2011)
David Rumelhart was a psychologist at UCSD San Diego and an early cognitive scientist who worked on symbolic artificial intelligence.
With Donald Norman<, he wrote the 1975 book Explorations in Cognition and in 1977 with Andrew Ortony he wrote the article "The Representation of Knowledge in Memory" in the book Schooling and the Acquisition of Knowledge published by Lawrence Erlbaum.
In the latter he wrote...
While originating from the senses, knowledge is not a blind record of sensory inputs. Normal people are not tape recorders, or video recorders; rather, they seem to process and reprocess information, imposing on it and producing from it knowledge which has structure. The human memory system is a vast repository of such knowledge. Some of this knowledge seems to be in the form of specific memories of particular events which we have experienced; some of it seems to be
in the form of more general abstractions no longer tied to any particular time, place, or source. It is one of the tasks of a theory of the representation of knowledge to provide a characterization of the way in which knowledge is structured so that progress may be made toward answering other important
questions...
The progress to which we refer can be regarded as the focal point of a new
emerging discipline called Cognitive Science. The research of an increasing
number of people working in artificial intelligence, cognitive psychology, and linguistics on problems concerned with the representation of meaning and the structural and processing aspects of knowledge, reveals a substantial convergence of opinion on the essential components of systems for representing knowledge. Recent papers by Bobrow and Norman (1975), Minsky ( 1975), Norman, Rumelhart, and LNR (1975), Rumelhart (1975), Schank and Abelson (1975), and Winograd (1975) all attest to this convergence.
Inspired by Frederic Bartlett's classic 1932 work in experimental psychology Remembering, Rumelhart (and Ortony) identified those "components of systems for representing knowledge" with "schemata." The ancient Greek concept σχημα, meaning figure or pattern, was adapted by Immanuel Kant to describe his 3x4 "architectonic" or arrangement of twelve pure concepts of understandng.
Rumelhart was the first author of a highly cited paper in Nature 1985 which was co-authored by Geoffrey Hinton, the "father of artificial itelligence." In 1987 Rumelhart moved to Stanford University, serving as Professor there until 1998. He received many prizes, including a MacArthur Fellowship in July 1987 and was elected to the National Academy of Sciences in 1991.
Perhaps his most extensive work was the 3-volume reference series Parallel Distributed Processing, co-edited with Jim McLelland and the PDP Research Group at UCSD San Diego.
In the preface to volume 1 Rumelhart wrote (p.x)...
Our interest in these topics began in
earnest, however, during the period when we were developing the
interactive activation model of word perception, in 1979, ·shortly· after
Geoffrey Hinton began a postdoctoral fellowship at UCSD. Geoffrey's
crisp explanations showed us the potential power and generality of
models created from connections among simple processing units, and
fit together nicely with our own developing conviction that various
aspects of perception, language processing, and motor control were best
thought of in terms of massively parallel processing.
In our information philosophy model of the mind as an Experience Recorder and Reproducer, the brain does not computationally "process and reprocess information." There is no central prcessing unit, let alone multiple parallel processors. There is no digital data storage, no addressing system to access such stored knowledge, and no communication of the knowledge to the prcessors.
The experience recorder and reproducer (ERR) is simpler than, but superior to, the computational models of the mind popular in today's neuroscience and cognitive science, the mind as "software in the brain hardware."
Although we agree that the mind is, like software, immaterial information, we think that man is not a machine, the brain is not a computer, and the mind is not "processing" digital computer information.
Of course the mind/brain does "process" information. It also creates new informaton, as part of the cosmic creation process, despite the second law of thermodynamics that demands the destruction of information as disorder and entropy necesarily increases.
As to Rumelhart's "representation of knowledge in memory," the ERR mind simply recalls the knowledge in related past experiences that are similar in some way to a current experience so as to provide context and meaning for a current experience, including the feelings and emotions of those similar past experiences that help to evaluate the worth of the current experience.
