Prior Knowledge and Neurobiology ?
mckee at starbase.neosoft.com
Sun Mar 5 10:30:26 EST 1995
Edward Simon Dunstone (ted at wraith.cs.uow.edu.au) wrote:
: Does anyone know of references that describe the role that
: prior knowledge plays in learning, from a biological rather than
: cognitive or computer science perspective (ie at the neuronal or dendritic
: level). Would I be correct in assuming that so far there has been
: little work in this area ?
I think you'd be correct in saying there's been little work in this
area, and I'd argue that people who are avoiding this topic are doing
so for a good reason. Thinking about knowledge at the level of neurons
or smaller is what philosophers call a "category error". In spite of
the ravings of some artificial neural network theorist about "the
hedonistic neuron" (I've conveniently forgotten his name), neurons
don't "know" anything, they're just pumps for potassium, calcium, and
It's only when neurons become organized into a
brain with representational power that the "system" can be said to
have any knowledge, be it prior, current, or subsequent. Although
the most interesting organism (us) obviously contains representations
and knows things, the brains of most organisms don't, they're too simple.
It's only via a psychological projection that flowers "know" to develop
flowers in time for the bees in springtime or spiders "know" how to build
At the abstraction level where the word "knowledge" is meaningful,
it doesn't matter whether you're a biologist, psychologist, or
computer scientist, which is why the interdisciplinary field of
"cognitive science" exists, to cover all these bases.
I've found the work of Allen Newell usually quite enligntening,
particularly his magnum opus (literally "big book"; actually it's only
500 pages) "Unified Theories of Cognition". I also found Henry Plotkin's
"Darwin Machines and the Nature of Knowledge" worthwhile in understanding
"where does knowledge comes from" in an evolutionary framework.
If you're in a philosophical community where truth is less important
than having a good argument, this can be a highly productive field,
publication-wise. If you're interested in doing hard, data-based
science, what we need is a knowledge-based expert system that "grows"
neural nets "in calculo" based on statements tagged to the literature.
This will take massive amounts of compute power and even more work
entering the facts. Just barely reachable within today's state of
the computational art, I think.
- George McKee
: Thanks in advance,
: Ted Dunstone
: CITR Neural Network Group
: University of Wollongong
: NSW, Australia 2500
Internet: mckee at neosoft.com
Voice: +1 713 890 8122
More information about the Neur-sci