Interactive Empiricism: The Philosopher in the Machine

1
InteractiveEmpiricismThe Philosopher in the
Machine

• Ron Chrisley
• COGS/Informatics
• University of Sussex

2
Take-home message

• Philosophy Engineering A two-way interaction
• Some philosophical breakthroughs may only come
  about through attempting to design and build
  working systems (engineering helps philosophy)
• Building complex systems (e.g. an artificial
  consciousness) might require incorporating
  scientists and philosophers into the design,
  modelling
• How they affect the system dynamics
• How they system dynamics affect them

3
Direction 1Engineering conceptual change
4
Conceptual problems

• Not all limitations on our scientific
  understanding are a matter of insufficient data
• E.g., consciousness
• Naturalist intuition consciousness (like
  everything else) is at root a physical phenomenon
• "Zombie hunch" It is possible for there to be a
  creature physically identical to you, but
  nobody's home

5
Conceptual change

• Best diagnosis It is our concept of
  consciousness that is to blame
• One solution change our concept of
  consciousness, so that we no longer suffer from
  the zombie hunch

6
Conceptual conceptual change?

• But it seems unlikely that this conceptual change
  could itself come about purelyconceptually,
  merely by, e.g.
• Acquiring more beliefs
• Philosophical argumentation
• Reading journal articles

7
Non-conceptualconceptual change

• Rather, problems of consciousness seem to require
  a non-conceptual development in our concepts
• (Bad) examples of non-conceptual change
• Getting hit on the head
• Undergoing neurosurgery
• Taking drugs?!

8
Non-conceptualconceptual change

• Better change that is non-conceptual, but still
• Rational
• Justified
• Based on experience of the subject matter
• What kind of change/learning could this be?

9
Concepts as skills

• Wittgenstein What underlies being able to move
  between ways of seeing something (e.g.
  duck-rabbit) is the "mastery of a technique"
• Then (some) concept acquisition is like skill
  acquisition
• Just as one can't read/argue/theorize your way to
  knowing how to ride a bike
• so also with some concepts one must experience
  the phenomenon to understand it

10
Interactive empiricism

• But not just passive experience (normal
  empiricism)
• Rather, interaction mastery of how one's
  experiences of the subject change in the light of
  one's different interventions (interactive
  empiricism)

11
Interaction is essential to

• Perception (O'Regan and Noë Sensory-motor
  contingency theory )
• Consciousness (Hurley Consciousness In Action)
• Cognition (Bickhard "Interactivism A
  Manifesto")
• Mammalian visual development (Held and Hein)

12
Meta cognitive scienceTheorist as subject

• A science of human cognition in general should
  apply to the cognition of cognitive scientists in
  particular
• If the cognitive science is right that cognition
  is essentially interactive
• then doing cognitive science (or AI) should be
  as well

13
Engineering as interaction

• But what kind of interaction?
• Perception of brain states (one's own and
  others') during manipulation (social, physical,
  etc.)?
• Limited
• (compare doing something similar with a computer)
• Better attempt to design and build cognitive
  systems, and observe them working (or failing
  to!) Engineering

14
An aside The Mary problem

• Jackson's Knowledge Argument against a physical
  science of consciousness
• Mary knows everything the physical sciences can
  tell us about colour, but has never seen red
• Will she acquire some knowledge when she sees red
  for the first time?
• Yes, she will learn what it is like to see red
• So there is knowledge of consciousness the
  physical sciences cannot provide

15
Solving the Mary problem

• But science is essentially interactive
• So although Mary may have read every possible
  book about color vision
• she doesn't have all the knowledge involved in
  doing color science
• Or rather, if we assume that she has all such
  knowledge, then it is a contradicition to also
  assume that she has never interacted with redness
  (i.e., seen red)

16
Direction 2The philosopher in the machine
17
We are a part ofthe systems we build

• Just as interaction can have a crucial,
  beneficial effect on the theorist/philosopher
• so also can it have such an effect on the system
  being designed/built

18
We are a part ofthe systems we build

• Q What has been the biggest engineering advance
  in AI in the last 20 years?
• A Kismet's eyebrows (Breazeal et al)

19
Interacting with Kismet

• Kismet could only learn to visually track objects
  if trained on suitable stimuli
• This required a trainer to wave objects in front
  of Kismet at a certain speed, distance, etc.
• How to ensure this efficiently?
• Exploit affective responses in the trainer if
  trainer gets too close, Kismet jumps back, and
  raises eyebrows
• Trainer readjusts without having to be
  instructed, understand physics of the system, etc.

20
Combining directions 1 and 2

• If we are part of the system, then not only can
  we have a beneficial causal effect on the robot's
  performance, but vice versa
• Thus, instead of trying to design an AI/machine
  consciousness in one step
• why not instead design a system S1 so that it
  will prompt conceptual changes in us
• that will enable us to design an S2 that will
  prompt changes in us
• that will enable us to design an S3
• and so on?

21
Frank Herbert's prescience

• In the science fiction novel Destination Void,
  the author of Dune speculated that the best way
  to create a machine consciousness might be to
  design a situation in which
• Carefully engineered people (clones)
• In a carefully engineered technological
  environment (computers, spaceship, neural
  wetware)
• Are manipulated and motivated to find a way to
  create machine consciousness (e.g., they will die
  if they don't!)
• A crucial part of the project is for the
  challenges they face and the technology they
  build to play a role in them figuring out what
  consciousness is (conceptual change!)

22
From fiction to fact?

• Perhaps it is not too far-fetched to suggest that
  something like this could be developing
• Not just work like Kismet
• But also, e.g., the search for creative
  technologies environments, document systems,
  brain wave induction devices etc. that facilitate
  insight
• Synthetic phenomenology interactive familiarity
  with a robotic system as a way of developing a
  means of specifying linguistically inexpressible
  experiential content (e.g., Chrisley and
  Parthemore)

23
Thank you.

• ronc_at_sussex.ac.uk
• http//www.sussex.ac.uk/users/ronc
• Talks available in various media at
  http//e-asterisk.blogspot.com