Transcript Com3240 Adaptive Intelligence

Com3240 Adaptive Intelligence
Epigenetic Robotics
Amanda Sharkey
Reinforcement Learning
Eleni Vasilaki
Epigenetic Robotics
• Interdisciplinary: developmental psychology and
robotics
• Overlaps with artificial life, cognitive science,
adaptive behaviour.
• Focus on “prolonged developmental process
through which varied and complex cognitive and
perceptual structures emerge as a result of the
interaction of an embodied system with a physical
and social environment”. (Zlatev and Balkenius,
2001)
• 2001-9 Epigenetic Robotics conferences
• 2003: Connection Science special issue
• 2006: Interaction Studies special issue
• Robotics learning from psychology
– Models of cognitive development
– Biologically inspired
– Just inspired, or can it generate testable
predictions?
• Psychology learning from robotics?
– Building robotic models requires further
specification and implementation of underlying
mechanisms (synthetic modelling)
• “true intelligence in natural and (possibly) artificial
systems presupposes three crucial properties:
• (a) the embodiment of the system;
• (b) its situatedness in a physical and social
environment;
• (c) a prolonged epigenetic developmental process
through which increasingly more complex cognitive
structures emerge in the system as a result of
interactions with the physical and social environment.”
• Zlatev and Balkenius (2001 proceedings)
• Brain, body and environment: reciprocally
coupled.
• Contrast to earlier emphasis on ‘mind as
computer’ , where body is output device for
mind that manipulates symbols representing
the world, in rule-like way.
• Progress in epigenetic robotics? Successive
proceedings.
• “Epigenetic Robotics is a new discipline at the
frontier of developmental psychology, neural-,
and engineering sciences whose goal is to
model the development of cognition in
natural and artificial systems” Metta and
Berthouze, (2006)
• The word “epigenetic”was chosen here in
antithesis to “phylogenetic”, to distinguish this
work from the field of evolutionary
approaches and artificial life, and rather stress
the aspect of postnatal development.
• Metta and Berthouze (2006)
• Phylogenetic development vs ontogenetic
– Previous efforts to create intelligent machines
• Direct programming
• Supervised machine learning
• Evolutionary adaptation
• Developmental robotics
– Autonomous learning
– Greater emphasis on learning own sensor and
action models
– Epigenetic robotics
Weng et al (2001) Autonomous mental development in
robots and animals. Science, 291, 599-600
• The autonomous development paradigm for
constructing developmental robots is as follows:
• · Design a body according to the robot's
ecological working conditions (e.g., on land or
under water).
• · Design a developmental program.
• · At "birth," the robot starts to run the
developmental program.
• · To develop its mind, humans mentally "raise"
the developmental robot by interacting with it in
real time.
Relevant concepts
• Previous courses are relevant –
• John Searle (1980):
• Strong AI: an appropriately programmed
computer really is a mind, can be said to
understand, and has other cognitive states.
• Weak AI: a computer is a valuable tool for
study of mind – makes it possible to formulate
and test hypotheses rigorously
• (Kurzweil (2005) confusingly also uses term strong AI to refer
to a “machine with the full range of human intelligence”)
• Turing – how to achieve intelligence?
• Child machine
– Some innate ability
– Education (punishment and reward)
Some background: Jean Piaget
• Developmental psychologist: constructivist
• Baby and child’s mental structures develop as a result of adaptation
to the environment
• Adaptation to the environment
– Assimilation: assimilating the environment to own mental structures
• E.g. babies sucking objects
• Transforming the world to fit own mental structures
– Accommodation: altering mental structures to fit with environment
• Mental structures changed by interactions with the environment
• E.g. learning to turn head, and finding nipple to suck
– Equilibration
• When external reality does not match internal logical mental structures, more
sophisticated mental structures are developed.
•
•
•
•
•
•
•
•
Stages of development
1. Sensorimotor development (birth-2 years) “
(i) simple reflexes; (sucking, grasping)
(ii) first habits and primary circular reactions; (e.g.
thumb sucking)
(iii) secondary circular reactions; (repeat actions to see
effect, e.g. grasping cord to ring bell)
(iv) coordination of secondary circular reactions;
(v) tertiary circular reactions, novelty, and curiosity;
and (vi) internalization of schemes."
• 2. Preoperational stage (2-7 yrs)
– No logical thought
• 3. Concrete operational stage (7-12)
– Logical thought with practical aids
• 4. Formal operational stage (12 onwards)
– Abstract logical thought
– Each stage is reached as a result of experiences at
previous stage
Epigenetic Robotics
•
•
•
•
•
Is it easy to define?
What are its aims?
What techniques does it use?
What is there of value in it?
What are the main references/people in the
area?
• Is it progressing?
• Assignment: 2500-4000 word essay:
•
•
•
•
•
•
•
What are the aims of Epigenetic Robotics, and how likely are they to be achieved?
Or
What is Epigenetic Robotics, and how likely is it to make a significant contribution
to Artificial Intelligence?
Or
Epigenetic Robotics, Developmental Robotics, Swarm Robotics, Behaviour-based
Robotics, Evolutionary Robotics: what are the differences between them, and how
useful is it to distinguish between them?
or
How easy, and also how useful, is it to distinguish Epigenetic Robotics from the
following: Developmental Robotics, Swarm Robotics, Behaviour-based Robotics,
Evolutionary Robotics.
• Assignment due: Tuesday April 13th (week 7 semester2)
• Strengths
– Sets new (ambitious) challenges and goals for AI
– Broad church – many approaches
– New inspiration from a rich source (child
development)
• New concepts e.g. Ongoing emergence, affordances
• Weaknesses?
• Overambitious? Can autonomous learning be achieved in
artificial systems?
– Sometimes goal described without saying how to achieve it
– Sometimes focus is on method/technique not clearly related to ER
• Broad church ... Lack of focus
– Achievements? Practical? Theoretical?
– How much leeway for new approach?
• Marking scheme:
• Content
– Is it well researched? Does it show good
understanding of Epigenetic Robotics and
associated issues? (aims, examples,
criticism/reviews)
• Argument and organisation
– Well structured?
• Spelling, grammar, references
– Poor writing obscures what you want to say.