Thursday, 31 January, 2002
John Campbell, Weiskrantz Room
Marr (1982) elevates modularity to a principle in AI, with reference to his 3 levels
Fodor�s 9 marks of modularity
Marr � claims that the 2D sketch is subjectively available � one-off remark
usually, the lower levels are unavailable to central (i.e. conscious) access
input systems have shallow outputs � basic visual processing isn't too heavily top-down(???)/loaded with memories etc.
major aspect of Fodor�s modularity
belief �independence of perception � couldn't otherwise see what you don't expect
computation vs intelligence � role in vision
synchronic vs diachronic � at a moment vs over time
transparency assumption � Caramazza, 1986
double dissociation � Plunkett doesn't buy this at all
assimulation � perception as hypothesis-generation, problem-solving/intelligence (e.g. Gregory, Rock)
Fodor � no, it�s domain-specific, mandatory, fast, limited central access � modular computation
Can we really have laws of nature stated in terms of representations?
the use that is made of it computational/functionally
external correlation between neural + physical phenomena
role of that represenation in the organism�s success?
what its selective advantage?
Fodor�s first law of the non-existence of cognitive science
the more global, the less understood a mechanism (e.g. analogical reasoning)
principled reasons why we can't understand these unencapsulated processes
central systems (e.g. ordinary thinking)
isotropic (don't know which information will bear on other information)
Quinean (coherentist � confirmation applies to theory as a whole)
why can't you have laws governing central systems according to Fodor? Campbell doesn't get it � he doesn't think the above premises discount it
the theory theory
children + scientists do the same thing in generating/testing hypotheses + theories
scientists � just as rattling more complicated and expensive toys, or alternatively, as big children
terrible twos � experimenting with implications of theory of mind
don't seem tobe laws of scientific thinking/generating alternate theories/logic of discovery
Marr�s 3 levels were levels of what???
why am I so sure that Fodor is wrong??? to what extent is it an empirical question???
see Rethinking Innateness
possibility of a connectionist AI that thinks connectionistically, like the Moties???
advantage of modularity for evolution???
why do MRI studies show such large areas for a given function
large blood flow??? no, they�re higher resolution than that
each N plays different roles???
where is colour vision in the brain??? V4 or V5???
it�s Fodor�s information encapsulation premise that I�m least happy about
are there many modular-hidden Ns in the brain??? I spose so
quantify modularity (in terms of information encapsulation)???
big central model, or multiple drafts??? � fits with multiple drafts
there are �laws� to the logic of discovery, but they�re at a low level that�s neither general� nor linguistically expressible
would multi-modal areas be modules???
let's say we've got a sensory association area, say the parietal cortex. Andersen et al (1997) argues for an abstract multi-modal distributed representation of space, combining vision, somatosensation, audition, and vestibular sensation, which 'can then be used to construct multiple frames of reference to be used by motor structures to code appropriate movements', as well as selecting stimuli and helping to plan movements. This fits in with Goodale & Haffenden's description of the functions of the visuomotor system, and aligns with it in neuroanatomical terms too.
so, we're imagining that the parietal cortex is taking in signals from visual, auditory, somatosensory and vestibular perception, and combining them to form some multi-modal abstract representation of space. to what extent should we consider this to be a module?
yes, this is a module, because it has most of the properties, and it�s domain-specific, even if it�s not modality-specific