[dedication] "What you have to remember about parsing," Merrill said, "is that basically it's a reflex."
[acknowledgments] Scholarship is the process by which butterflies are transmuted into caterpillars
p.2 this is going to be a rather long and rambling story, a fault for which I apologize in advance.
p.5 the notion of computation is intrinsically connected to such semantical concepts as implication, confirmation,
and logical consequence. Specifically, a computation is a transformation of representations which respects these sorts of
semantic relations.
p.18 generally, one remembers what one understands.
p.39 Any mechanism whose states covary with environmental ones can be thought of as registering
information about the world
p.40 what perception must do is to so represent the world as to make it accessible to thought.
p.40 perception is a mechanism of belief fixation par excellence: the normal consequence of a perceptual
transaction is the acquisition of a perceptual belief.
p.49 If you have an eccentric stimulus domain - one in which perceptual analysis requires a body
of information whose character and content is specific to that domain - then it is plausible that psychological processes
defined over that domain may be carried out by relatively special purpose computational systems.
p.52 highly tuned computations are suggestive of special-purpose processors.
p.52 From our point of view, the crucial question is all such examples is: how good is the inference from
the eccentricity of the stimulus domain to the specificity of the corresponding psychological mechanisms?
p.63 one can, and often does, spend hours thinking about a problem in philosophy or chess,
though there is no reason to suppose that the computational complexity of these problems is greater than that of the
ones that are routinely solved effortlessly in the course of perceptual processing.
p.63 It is only in trick cases, of the sorts that psychologists devise in experimental
laboratories, that the perceptual analysis of an utterance or visual scene is other than effectively
instantaneous.
p.64 it may well be that processes of input analysis are fast because they are mandatory. Because
the processes are automatic, you save computation (hence time) that would otherwise have to be devoted to deciding whether,
and how, they ought to be performed.
p.64 Compare: eyeblink is a fast response because it is a reflex - i.e., because you don't have
to decide whether to blink your eye when someone jabs a finger at it. Automatic responses are, in
a certain sense, deeply unintelligent... But what you save by indulging in this sort of stupidity is
not having to make up your mind, and making up your mind takes time. Reflexes, whatever their limitations,
are not in jeopardy of being sicklied o'er with the pale cast of thought.
p.64 Some of the claims that I'm now about to make are in dispute among psychologists, but I shall make
them anyway because I think that they are true. [That's the spirit. Critics are just people who are in the way of your ideas.
Best just to ignore them and hope they go away.]
p.83 reflexes have two salient properties. They are computationally simple (the stimulus is "directly connected"
to the response), and they are informationally encapsulated [JLJ - Fodor believes that basic perceptual modules are partly
closed off from the cognitive background knowledge of the subject: the modules are informationally encapsulated]
p.107 By saying that scientific confirmation is Quineian, I mean that the degree of confirmation
assigned to any given hypothesis is sensitive to properties of the entire belief system; as it were, the shape of
our whole science bears on the epistemic status of each scientific hypothesis.
p.114 as soon as we begin to look at cognitive processes other than input analysis - in particular, at central
processes of nondemonstrative fixation of belief - we run into problems that have a quite characteristic property. They seem
to involve isotropic and Quineian computations; computations that are, in one or other respect, sensitive to the whole belief
system.
p.115 Raphael (1971) comments as follows: "(An intelligent robot) will have to carry out tasks. Since a
task generally involves some change in the world, it must be able to update its model (of the world) so it remains as accurate
during and after the performance of the task as it was before. Moreover, it must be able to plan how to carry out
a task, and this planning process usually requires keeping 'in mind' simultaneously a variety of possible actions and corresponding
models of hypothetical worlds that would result from those actions. The bookkeeping problems involved with keeping track of
these hypothetical worlds account for much of the difficulty of the frame problem"... So far as I can tell, the usual assumption
about the frame problem in AI is that it is somehow to be solved 'heuristically'. [JLJ - or, in my view, by using a "bag of
tricks". One doesn't have to re-invent an action each time it is needed - we see everyday of our lives examples of people wanting
things and executing very simple plans to get what they want which come from what appears to be a bag of tricks
- we just have to observe, infer the rules of the 'game', and then try to do the same ourselves.
Computation (thought) becomes an examination of the "tricks" available
and a custom-generated series of simulations and probabilities of the "tricks" resulting in the accomplishment of the goal,
including margin for fallback positions, uncertainty and resistance, and the presence of other entities trying to reach their
goals, in which you yourself are their obstacle, and where joint action is necessary to make progress.]
