Greg Detre
Wednesday, March 12, 2003
when the baby rats go through their behavioural repertoires after being fed � his casual hypothesis is that this is reward-seeking behaviour � �I just got rewarded, let me try and figure out what for��
but they only do it up until 10 days (which is around about the time that they�d start to feed on their own)
I just assumed it�s an undefined result of an unnatural event (a behavioural null pointer exception)
interesting that suckling is decoupled from aversive influences
especially given that rats are usually so good at social learning of whether tastes/odours are good
structural vs functional views of behaviour
initial assumption � suckling serves the same purpose as feeding, so surely it�s the same thing (why have a redundant second feeding system too?)
does the suckling develop into some other useful module???
could it be that it is used by female dams to figure out when to inject milk
in the first few days of its life, a chick�s pecking is totally uncorrelated with feeding/nutritional intake � pecking for pecking�s sake
for the first 48 hours, they�re still living off the yolk
their taste preference is partly set by the taste of the first things they eat (presumably because chickens will normally be born in environments where what they peck at will be good for them)
Bruce: why set this paper?
some innate behaviours bootstrap (like pecking), others don�t (like suckling)
suckling doesn�t lead to feeding, while pecking does
you might have behaviours that satisfy a motivational goal as a kind of side-effect
behaviours might be refined in motivational contexts other than what they�re eventually used for
perfect the behaviours in an environment where your life doesn�t depend on it
the play period is the interim period where you have some subsystems turning off, and new subsystems turning on and taking over
play is where the system is learning how to rewire the new systems up to produce appropriate behaviour
the flexibility of a hierarchical behavioural system is that rather than generating new behavioural primitives, you can just rewire old ones to perform new functions
why do the pups switch nipples?
perhaps for social reasons (because there�s no reason for them to switch if they�ve been receiving milk themselves)
Deb: dissolves the distinction between organism and environment when you consider that the female dam was once a pup
it�s arbitrary which parent/child pair of mechanisms evolve (e.g. dam providing all the milk necessary and pup controlling vs predefined occasional injections and pup on maximum intake)
makes sense to keep the rat pup as simple as possible
Bruce: had always viewed development as gradient ascent (a continuum)
Coppinger: no � neonatal specialist � adult specialist
suckling has nothing to do with feeding, but it can be done very simply
when building adaptive systems, we should consider building specialist early simple highly-optimised boostrapping modules
and specialist modules in general
what does this tell us about human specialist modules?
Cynthia: when human infants suckle, they stop after a while for a rest, and the mum jiggles to get them to restart, and one theory is that this is early teaching the baby turn-taking
comes back to Deb�s point of the parent-child as a coupled system
the fast-mapping of vocabulary requires you to remember complete contexts to associate with unknown words � perhaps you have a special module for remembering new words
or perhaps you only associate a very highly compressed representation of the context associated with the new word
Deb: why do we as system-builders care about these neonatal specialist modules? can�t we just skip them?
this is not a good question
because there may be learning going on under the hood that requires starting from the neonatal
otherwise, you would need to know in advance exactly what inputs and role your adult modules should play
Blumberg�s Hamsterdam model
Timberlake�s system:
system
�general motivation states .. t prime as et of underlying substates and modules realted to ap articular function�
subsystem
�coherent strategies that serve the general function of the system�
provides perceptual priming
mode
�motivational substates related to eh seuqential and tmeporal organisation of action patterns with respect to terminal stimuli in the system
e.g. feeding sequence might contain a general search omde, focused search, followed by consumption mode
module
�predispositions to respond to a particular stimuli with particular response components�
AIBO didn�t have a systems-level
the problem with the �try� behaviour that tries different behaviours to see what satisfies its motivations when faced with a new behaviour is that it doesn�t have anything to do when faced with really new objects
Deb: linking the form of something with its function is very powerful
Derek:
poorly labelled graphs etc.
they don�t adequately present Timberlake�s ideas, and make them seem very simple
Deb: they�re using �emotion� as little more than variable for �difference from goal state�
Don Norman seemed to be using it in a similarly impoverished way � Cynthia: he thinks that emotion is more related to the functional aspect within the system
muddles emotion and motivation
Bruce: thinks it�s a crap paper, doesn�t really run with the interesting ideas that it�s based on, just because it�s based on nature doesn�t mean it�s good
a lot of behaviour boils down to search, approach and do
perhaps we should try to use different action selection mechanisms at different levels
why do they mention �emotion� (the 6 basic ones) at all?
me: because they were trying to map the AIBO�s variables onto human emotions because it�s a toy
Bruce:
it�s clearly not a model of a dog
but if their goal is entertainment, they should be focusing on what�s entertaining for people
from a design standpoint, it�s rubbish
though he likes the electromechanical bits
Deb: but this is confusing the mechanism with a description of the mechanism
we�ve got these 6 basic description terms, and then they�ve gone away and tried to code them up in C (ignoring what�s important going on under the hood)