Philosophical Foundations of Cognitive Science - PowerPoint PPT Presentation

Loading...

PPT – Philosophical Foundations of Cognitive Science PowerPoint presentation | free to download - id: 7670e-YmNhN



Loading


The Adobe Flash plugin is needed to view this content

Get the plugin now

View by Category
About This Presentation
Title:

Philosophical Foundations of Cognitive Science

Description:

A colony, with each termite following simple, local rules, can ... of the arches, an initial part of the construction of the full termite mound... – PowerPoint PPT presentation

Number of Views:61
Avg rating:3.0/5.0
Slides: 23
Provided by: Michael2205
Learn more at: http://www.cogs.susx.ac.uk
Category:

less

Write a Comment
User Comments (0)
Transcript and Presenter's Notes

Title: Philosophical Foundations of Cognitive Science


1
Philosophical Foundations of Cognitive Science
  • Robotics, Dynamical Systems, Minimal
    Representationalism

2
Background
  • There are several empirical reasons to believe
    that we dont have the rich inner representation
    of the external world that naïve introspection
    would lead us to believe
  • This week we will look at various thought
    experiments, and actual experiments, designed to
    show that rich inner representations really
    arent required to build real cognitive systems
    (contrary to the expectations and assumptions of
    GOFAI)

3
European Philosophy
  • As you will see next term, this new approach to
    artificial cognition (outlined this week and last
    week) actually has a lot in common with the
    European philosophical school of Phenomenology,
    e.g.
  • Merleau-Ponty
  • Heidegger
  • Husserl
  • Ideas developed principally during the first
    two-thirds or so of the last century
  • Emphasis was on the necessity of an assumption of
    an embedded, embodied interaction with the world
    as a precursor to any analysis of that nature of
    mental life
  • Until recently considered by most Anglo-Saxon
    scientists as quite opposed to any scientific
    analysis of cognition, but now considerably less
    so

4
Dynamical Systems
  • This week we are looking at thinkers squarely in
    the Cognitive Science tradition who have adapted
    these ideas (often knowingly) in order to claim
    that we require a paradigm shift if we are to
    make further progress in understanding real
    intelligence
  • Particularly (this week)
  • Rodney Brooks
  • Timothy van Gelder

5
Dynamical Systems
  • But also (here at Sussex)
  • Inman Harvey
  • Ezequiel Di Paolo
  • And, here and elsewhere,
  • Many, many others (philosophers and scientists)
  • In fact this is now an extremely widespread
    movement, almost the new orthodoxy
  • It has not (yet?) replaced the old orthodoxy
  • They co-exist, uneasily!

6
Timothy van Gelder
  • Philosopher, at Melbourne
  • Author of many papers on the correct analysis of
    cognition in a connectionist, distributed,
    dynamical world (see his website for more),
    including
  • What might cognition be, if not computation?
    Journal of Philosophy 91 (1995) 345-381
  • Dynamics and cognition in Haugeland, Mind
    Design II (1997)

7
Dynamics and Cognition
  • In these two papers, van Gelder compares the
    conceptual frameworks offered by
  • The classical computational framework
  • The connectionist framework
  • The dynamical systems framework
  • This week we are contrasting dynamical systems
    with a more traditional symbolic AI approach
  • Van Gelders favourite, and famous, example

8
The Watt Governor
9
Rodney Brooks
  • Rodney Brooks also proposes a radical, new
    framework for evolving intelligence
  • It was a manifesto, or call to arms, for the new
    embodied robotics approach to understanding
    cognition
  • We cant understand an insect, so why should we
    think we can understand a human?
  • Lets start small
  • How? Well, heres one way, says Brooks

10
Subsumption Architecture
  • Multiple layers
  • Each new layer built on (i.e. at a minimum
    assumes existence of) older layers
  • Brooks doesnt really emphasise this, but we can
    see that his architecture is inspired by the
    apparent conservatism of natural evolution

11
Subsumption Architecture
  • In Brooks version of this layered architecture,
    individual layers operate largely independently
    of each other
  • Decomposition by Activity
  • We can look at the agent Allen
  • Lowest layer avoids walls, stops if it collides
    with something
  • Next highest layer gives an innate
    predisposition to wander
  • Next highest layer again actually allows robot
    to explore
  • Each layer does not have to be concerned with
    tasks of the next lowest layer
  • Each layer has a different world in which it
    can operate, when compared to the world
    available when designing (or evolving) the lower
    layers

12
Herbert
  • The more complicated robot discussed in this
    weeks paper, Herbert, has
  • No internal state maintained for longer than 3
    seconds
  • No communication at all between separate
    behavioural layers
  • It has separate
  • Wander behaviour
  • Laser based can orienting behaviour
  • Can pick-up behaviour (infrared beam)
  • Nevertheless it produces coherent behaviour in
    the world, wandering around Brooks office,
    picking up coke cans
  • In fact, this separation allows additional
    opportunism

13
Earlier work
  • W. Grey Walter (The Living Brain, New York W.
    W. Norton, 1953) a British Cyberneticist
    physically produced simple embodied agents with
    no explicit representation which actually did
    things in the world
  • Similar work was rediscovered, in thought
    experiment, by Valentino Braitenberg Vehicles
    experiments in synthetic psychology, MIT Press,
    1984

14
Philosophical Implications
  • Brooks says that he is not particularly
    interested in demonstrating how human beings
    work and not particularly interested in the
    philosophical implications
  • We dont, perhaps, have to take him at his word
  • Other people certainly are interested in these
    implications of his work

15
All kinds of e
  • Next term we will go further into the philosophy
    of embedded, embodied, enactive, etc. cognition
  • We can already see the beginnings of a working
    alternative to the representation-rich,
    symbol-systems view of cognition

16
Minimal Representationalism
  • Andy Clark, for one, is not happy with the idea
    that we should through away representations
    completely, just now
  • But (even for Clark) this is a very different
    view of cognition
  • Representation has a much less central role
  • Perhaps this kind of behaviour, not rational
    reasoning, lies at the centre of cognition
  • None of this looks anything like a systematic,
    compositional, propositional Language of Thought

17
More examples
  • Clark discusses yet another example of a
    distributed, complex system, which seems to
    produce emergent intelligent behaviour
  • A termite colony
  • No individual termite can build a termite nest
  • A colony, with each termite following simple,
    local rules, can
  • Clark discusses the creation of the arches, an
    initial part of the construction of the full
    termite mound

18
Evolution
  • We are not covering this in detail this term, but
    it is worth understanding, when reading papers in
    A-Life, that because intelligent behaviour is
    emerging from such simple rules, it is often
    possible to engage in Artificial Evolution of
    these behaviours
  • Clark (in Mindware) mentions this on p.99, with
    some references to some user-friendly guides,
    if you want to understand more, but

19
Evolution
  • You should at least be aware that this kind of
    simple artificial cognitive behaviour can be
    successfully evolved
  • With an artificial genome (data in a program!)
  • And an artificial phonotype (thats the
    creature)
  • Often the evolution is all in a simulated world
    (which Brooks wouldnt like, of course), but
    sometimes the fitness of these robots really is
    evaluated by actually building them, and seeing
    how they do!

20
Comparison to Human Intelligence
  • We can reason logically, but were really rather
    bad at it
  • If you just want to get the steps right,
    computers are better
  • For choosing which steps to take, we still win,
    hands down
  • We still make (far) better mathematicians than
    computers do

21
Chess
  • Computers (eventually) beat all of us at chess by
    being very good at something were very bad at
    (the logical reasoning massive, lightning fast,
    purely rule-based look-ahead in this case)
  • But it was hard to make them beat us, even so,
    because we are very good at things they are still
    very bad at
  • Pattern recognition in particular, but this is
    just part of
  • General competence to act successfully in the
    world

22
General Competence
  • Flies and frogs have that, too
  • So what is it?
  • How can you build it?
  • And where and when does propositional reasoning
    come in to the picture?
  • The first two questions are A-Life, now
  • The third is still (at the moment) philosophy,
    even for those trying to answer the question from
    an A-Life point of view!
About PowerShow.com