Greg Detre
Thursday, April 03, 2003
I know you have all been eagerly awaiting assignment 1, and here it is.� It will be due Monday, April 7.
Please read the draft of chapter 7 of the Emotion Machine, which is now available at
http://web.media.mit.edu/~minsky/E7/eb7.html
Then, write a short essay where you (a) criticize a specific theory in chapter 7 that you
disagree with and (b) debug the problem you have
identified, perhaps by making some changes to the theory, by elaborating the
theory by supplying missing details, or by providing an alternative theory of
your own imagining.
For example, you might complain that thinking doesn't
involve a 4-phase cycle, but rather a 7-phase cycle, or you might complain that
the described ways of thinking are missing some especially important way to
think that you use all the time, or you might complain that the critic-selector
model is too simple because there are only two kinds of parts when there should
be three or four or more.� And so forth.
Restrict your responses to up to 1000 words.� Submit an electronic copy (as plain text, a Word document, or a URL to a web page) by e-mail to minsky@media.mit.edu and push@media.mit.edu.
what�s the difference between critics and selectors???
why critics AND selectors??? why not lump both together???
one looks for preconditions, one acts (correspondingly)???
I guess so that you can pair them up differently/combinatorically�???
what level do you find critics???
at every level??? just high levels???
at every level
is all of the brain reactive then???
he addresses this � he kind of says yes, but in such a complicated, modularised, inter-related way that I suppose it doesn�t make much sense to talk in these terms
presumably the Critics� sensitivities and choices of Selectors are influenced by emotion/mood???
I think he argues that they create emotion
� not quite: emotional states are visible Ways to Think
the preconditions for the Critics should be flexible, or influenced top-down in some way
he doesn�t explain enough about the preconditions that activate the agents that notice the preconditions that run the Critics�
e.g. noticing that a problem is familiar is no mean task�
these ideas can�t really be implemented until you have a pretty complete high level system already (in contrast to Minsky�s claim that SoM is the �many shallow approaches at once approach�
that�s maybe not fair � after all, you do get Critics at the lower levels
confusion and competition between Critics is interesting
are Critics being activated at different timeframes continually???
I think so � see end of ch
is there some level at which there are no Critics???
how does the notion of Critics/Selectors fit with the Model 6???
they all have Critic/Selector pairings
Hofst temperature vs Critics as a means of dealing with impasses
but Minskyan Critics serve other purposes as well�
like what???
I�d be happier if he could see emotional states as different in kind rather than degree (of duration) (as in externally noticeable)
critic/selector vs for/backward thinking??? did that idea come from Minsky originally???
do the Critics get criticised???
how do the Critics self-organise???
I suppose in
some ways that Critics are easy to organise � once you have some method of
deciding whether to add a new one, you just add a Critics to a high-order
Critic as though it�s any other agent that�s getting trapped in some problem
type
emotional states ARE WtTs � see 7.3.4
it�s kind of the other way round, right, i.e. WtTs that are high-level, generate cascades, and visible from the outside, are emotional states
can you have Critics of Critics at
the same level???
how do Protectors work??? are they Critics of Critics???
well, bear in mind that Critics
aren�t necessarily inhibitors� they�re better named
Noticers
I think Protectors are things that stop another sub-system taking over from an activec sub-system
do the Protectors pair up with Selectors too??? or do they inhibit the Selectors� actions
I suppose you have the Critic/Selector combination to take different contexts into account when choosing the new WtT �
but couldn�t you add the context-detecting features to our preconditions???
no, but then you�d need a �switch� statement in your Critic/Selector Combo, making it too complicated
I kind of feel that by choosing just one type of Way to Think (the alarm type) and reifying it in terms of Critics is neglecting the other (equally important) agents that make up Ways to Think (e.g. analogy-seeking + mapping agents)
how the fuck would something like this self-organise??? would all the Critics be hard-wired???
this is as much a part of learning as learning in new situations when you have a fully-functional system in place already
I guess you build up from the levels slowly through infancy, starting with instinctive and learning some reactive, then maybe a few deliberative can start to take over
he doesn�t talk enough about how the system could self-organise from nothing
does it start with a random smattering of Critics at each level that get rewarded for noticing impasses and choosing appropriate Selectors to deal with them???
how does it generate new Critics???
one of the most interesting ideas to
emerge from cognitive science in the last 50 years is that systems which
apparently embody complex rules (at different levels???) can self-organise
is this an idea from cogsci??? Hebb was a psychologist, right???
how do the ideas relating to B- and C-brains fit with the Model 3 or 6 architecture???
attack the Model 6 architecture
once you have agents that can operate on other agents (i.e. second order agents) you don�t need to add any special further abstractness to generate arbitrarily high-order agents
does he do enough to talk about the different kinds of problem types???
how are censors etc. different from Critics???
he describes a problem as hard if
none of our current methods have worked on it � I just don�t think that quite
covers it
there�s a mental effort feeling/variable of difficulty that can be high even if we feel that we do more or less know how to do a problem, or if we�re simply having trouble keeping it in our heads
this is more of a feeling of cognitive limitation
his discussion of a kind of inter-agent currency doesn�t fully fit the bill either
what about emotions as totally flat emotional affect, i.e. modular, e.g. in Damasio???
see Minsky�s discussion of this in the middle
does it make sense to describe higher-level or Critics of Critics using the same name as just Critics, or Critics at the first level???
would there be a need for Critics at the bottom-level???
yes
but not censors, right???
is there
anything wrong with the account of emotions???
selfish gene theory, e.g. honour � such emotions are hard-wired
moreover, they aren�t in response to cognitive difficulties, e.g. disgust
why do we
have inhibitors rather than excitators???
�
problem of
credit assignment as related to the problem of comparing unlike with unlike???
of the
Society of Mind architecture, but it seems a particularly important problem
here. In fact, there may be two problems here:
Perhaps
even prenatal, perhaps neonatal, regarding hard-wired perceptual features that
are considered salient, the kinds of problem types that we�re born looking for
etc., perhaps even how the first Critics are hewn and honed from the �booming,
buzzing confusion� (James, 1890).
There are
also other ways in which Hofstadter�s approach is clearly influenced and
similar to the Society of Mind ideas. In later work (e.g. Metacat), he does
introduce added levels of self-reflection. Secondly, we could see the low-level
codelets that notice problems and continually evaluate the temperature in the
background as being like low-level Critics.
,
since presumably when we have no Critics that appear to be useful for a new or
unusual problem type, a massive cascade of stochastic dismantling of current
representations goes on, during paradigm shifts and brainstorms. We might term
such cascades curiosity, agitation, excitement or anxiety, depending on the
perceived urgency/danger of our situation.
critics vs selectors
protectors
B- and C-brains fitting in with Model 6 and Critics/Selectors
I�m a little confused about where the discussions of B- and C-brains
fits in with the Model 6 Critic/Selector model.
see footnotes in reactions file
Thinking about thinking
How do we choose what we think about?
The critic-selector model of mind
What might useful Critics be like?
What are some useful �Ways to think�?
Emotional thinking
Critics as mental economists
Positive critics and self-control
Appreciating the value of failures
Emotional embodiment
Some processes of everyday thinking
Poincare�s unconscious process
Is human thinking micro-cyclical?
Of course,
the inherent heterogeneity of the Society of Mind is, in a very general sense
indeed, a paradigmatic example of a weak method.
, implying
that the earlier work ignored such explicit high-level reflection out of
expediency
Secondly, one could see the low-level codelets that calculate changes in
temperature as simple critics. However, there are no high-level critics.
How could this be integrated???
This adds a complication, because it requires the entire state of the
system to be evaluated and summed up by some temperature/satisfaction variable.
This may or may not be easy for a given problem type.
Moreover, although Hofstadter�s letterstrings domain is rich in the
types of concepts, analogies and representations it supports, and the kinds of
examples and impasses it deals with are impressive, Copycat is pretty much
solving just the one type of problem, that of finding a mapping between
pairs of strings and applying it to a new string.
If we
decide that each sub-system/collection of resources in the mind has a separate
temperature to reflect its satisfaction with its own progress in its own
domain, then there has to be some way to weigh the constraints and demands of
different sub-systems, i.e. some way to compare temperatures across
sub-systems.
, My first
reaction on hearing the idea of problem-specific Critics as the mind�s means of
dealing with dead ends and problems was to think of
I found the
discussion of creativity and the unconscious interesting, but at this stage too
vague and speculative.