Greg Detre
Wednesday, April 16, 2003
NE43 playroom
|
Minsky |
Brooks |
|
gofai � hmmmm, not really search � again, not so much top-down software abstract symbols |
nouvelle ai behaviour-based, embodied cognition vs perception/action physical grounding robots bottom-up |
astonishingly, he�s giving the lecture as though some people won�t have read the SoM�!
symbol is not in the index of SoM
argues that we can follow Brooks� behaviour-based methodology without actually building robots
Brooks
argued that the old perception/modelling, planning, task execution, motor control, actuators was too slow for neurons to implement
threw all of that out
cognition is in the eye of the observer � it�s emergent � parallel layers � sensors to actuators, with just a bunch of tasks in the middle
argues that Minsky�s Model-6 doesn�t look so different
I�m not so sure � what it seems to come down to is the serial vs parallel � and Minsky�s layers do work in parallel
Push: you�re comparing a Minskian tower of reflection vs Brooks� super-goals
reflective � thinking about recent thinking
self-reflective � thinking about self as a larger entity, goals etc.
shows his agent-based little monster that goes round a world putting things in boxes � argues that it exhibits some of the Brooksian behaviour-based ideas
like what???
there�s no learning yet
I don�t see the kind of k-line he argues for in the suppression module in the subsumption architecture
k-line based programming language
you�re allowed state in a subsumption architecture � that is, each box has
why should I fear your research?
why should I rejoice in your research?
what should I tell my mum about this?
what�s the most interesting thing you�ve discovered?
what�s the most recent thing you�ve discovered?
push: do you really think that you can build a SoM programming language? will it be broad enough? it�s kind of analogous to the way you want to use different programming languages for different problems�
e.g. perhaps you might want to build first-order logic in natively to an intelligent system
Brooks: didn�t really see SoM and subsumption architecture as being incompatible
they reckon that one of the problems of AI is that there aren�t obvious milestones demarcating progress
after all, apparently Bobby Fischer isn�t actually that smart
make the usual, �once we understand how it works, it�s no longer considered intelligent� �
why hasn�t anyone actually built a SoM architecture so far???
Push: refce: carbonell � derivational analogy � directly based on k-line
and it�s had a catalytic role
does it matter that it doesn�t self-organise???
aren�t both systems pretty static??? problem learning/self-organising�