Greg Detre
February 5, 2003
shame - when you're worried that what you're doing will be disapproved of
by someone like your parent that you want to think well of you - cf Aristotle,
Rhetoric
books on how to feel better, whether you deserve to, or not J
Aristotle - brain theory, cooling the blood -
surface to volume ratio...
problems with translating mind, agent etc.
how did writing the SoM in fragments help with the lack of a 'plot'???
why doesn't he mention 'consciousness' in the SoM???
����� you mention that we have to put
a bracket around which processes to call 'mind' - do you have any new ideas
about how to group those???
vectors - invented by Gibbs - when???
augustine 400s - how does the mind work? and how come it's so hard to do
the right thing?
����� a person is just the memories
you have...???
����� minsky sees his discussion of
trying to remember something by associating components of the idea as a
precursor to the idea of data structures
analogy between the chinese room and viavoice... J
he characterises the chinese room in terms of a processor and a set of
rules, i.e. a von neumann machine - does it help that brains don't
differentiate between processing and rules??? almost certainly not relevant,
since that's only at the level of implementation
����� could you argue anyway that the
neural network has the processing in the activations, and the memory in the
weights ...???
no, because the activation can be the memory, and the weights the
algorithm�
he argues that the main problem with the Chinese Room is that it assumes
its own conclusion, by assuming that because he has the word 'understanding'
there must be something that brains have that the chinese room doesn't
on the one hand, augustine is looking to find out about the soul, while
also talking about the mind as a cavern containing subcaverns in
incomprehensibly complicated ways
early psychologists: wundt, james, helmholtz, galton
Lavoisier's insight about oxygen came in a flash when he realised that the
smell of sulphur could be particles floating in the air, and that these
particles might not even be sulphur
psychology was held back by physics envy
cf skinner, even trying to write his last book on how his few laws could
explain learning language
IMHO: a much better model for how the mind works is the manifest rules of
syntax, rather than physics
minsky points out that while piaget is applauded for his discoveries, made
mainly by interviewing his three children, freud is widely criticised by his
intensive interviews over years with just a small number of patients :)
����� argues that whenever there's
some paper arguing that infants can count or whatever earlier than piaget
realised, then they've just discovered some other mechanism besides the one
that the seven year old is employing
doesn't searle sort of think that something global emerges as a single
consciousness out of the brain???
idea of difference-engines comes from allen + newell
there's no way you can understand how thinking works if you assume that
there's someone thinking it - this is searle's problem
����� consciousness is just such a
single self idea
����� big suitcase of all the
different things we don't understand about the mind
attention is negative - it's about keeping everything else stable
����� parallel beings would realise that
consciousness is not a positive thing, but a suppression
he says that neuroscientists just don't think about the high-level - they
don't know what cache is (hippocampus)
����� grossberg is the only
neuroscientist with a decent theory of cortical columns - minsky thinks they
might be the terminals for frames
the 'emotion' machine - because we want to explain thinking in terms of
ways of thinking
����� as opposed to emotions as
something extra that's added to the brain, like colour
����� e.g. makes you reprioritise, assign
lower importance to long-term goals etc. if you're angry, or infatuated -
emotions as subtractive, rather than additive
hypothalamus - hot/cold
if consciousness is subtractive, doesn't that leave you with panpsychism???
I think he means attention, rather than consciousness
why is the brain connectionist, if the algorithms it's implementing are so
much higher level???
he thinks that the brain uses neurons cos they're digital
he thinks that the brain could be implemented on a pentium one
it's only the way it is because evolution has to stick with what came
before it
����� (might not include visual
processing)
you may need many hundreds of neurons for one bit of reliable memory
you don't want emergence, i.e. a big complicated system that emerges into
something clean and simple - it's one of the worst ideas in our culture - i
don't understand it so noone else can - they've just given up and stopped
thinking
����� there aren't any examples in
science of phenomena that get described as emergent and stay like that for ages
����� understanding is to know how
each part works and how it relates to all the others - reduction ignores these
relations
����� he thinks that all of kaufman's
good examples are of things where we can actually see how it happens
global workspace theory of consciousness (baars) - agents can't crosstalk
����� since the agents all have
different internal languages, you'd be much better off with lots of different
blackboards, rather than one with a high signal to noise ratio and loads of
filter
why are people afraid to work on common sense? this is the reason that his
ideas haven't been built as systems
����� there's something in our culture
that evades such approaches
e.g. what do you do about pollution - reduce the population...
����� lots of things would be better
paranome???
i know that we're talking about separate levels, of implementation and of
algorithms or function, but wouldn't you expect there to be more
neuroscientific evidence to support some of the high-level AI theories that you
have put forward???
how many of the main theories of humour has Minsky embraced???
�� humour as surprise - yes, in that
you're sneaking up on your censors by employing a second frame that it doesn't
notice, right???
�� superiority - yes, because you're
often using it as a mechanism for learning (often through social or
self-rebuke)
�� relief - no, not really???
�� incongruity - is this the same as
surprise???
do we have a separate logic reasoning agent, or is it simply a version of
uniframing etc. that happens to work with more abstract, algebraic symbols???
do we have a uniframing agent, or is it a method that all agents employ???
does it make sense to talk of different types of agents??? are k-lines,
scripts, isonomes etc. all types of agents???
do you really think that entire, or even large portions of, old agents
remain in our brain dormant to this day???
will it ever be possible to tease apart the distributed connectionist
representation of the brain to see different agents and realms???
why does he think we have emotions??? how did they relate to learning???
did he link them to intellect in the same way as damasio???
how do you explain the so-called 'g-factor' noticed in almost all IQ tests,
where children tend on average to cluster towards the high or low end in all
their abilities, rather than their performance in different domains being
almost independent of each other as you might expect if our agents are so
separate???
����� i think he he can easily argue
that he's shown how children who
learn good leaning strategies are going to do better overall in pretty much
every area. moreover, he's also shown how connected and inter-dependent the
different agents are
would you expect to see agent-level information in the genome at all???
����� he seems to imply not (pg 310)
why is our STM so small???
could we build a system now that would learn and benefit from these
number-meaning ideas??? if not, isn't that an indictment of this sort of
free-wheeling speculation??? don't you need to design (or even
reverse-engineer) with a task in mind???
could we actually teach maths any other way??? wouldn't the formal maths be
harder to learn at a later stage???
isn't the polyneme approach wasteful of memory??? (see 19.5)
����� not really, if the properties
each agent memorises don't overlap much
is this more than a semantic network cos multimodal/sensory info is
involved???
could the level/spiralling control be modified like the 'temperature'???
(see 20.6)
memorisers vs recognisers???