From Autism to Expertise: Connecting Neural to Cognitive

Understanding of Learning

Terry BossomaierCRiCS (Centre for Research in

Complex Systems)MIke Harré, Allan Snyder

Overview

• One of three talks:– This morning: expertise and cognition– This afternoon: complex systems– Wednesday: serious games

• This talk:– Concepts in the brain– Patterns and expertise– Cognitive transitions

Grand Challenges

To build human-friendly artificial creative thinking systems which scale to arbitrary size

To understand how social and organisational systems foster, or frustate, human creativity and how organisations can become themselves adaptive and creative.

Evolution and Learning

• Adaptability properties of entities (agents) in complex systems

• Evolutionary forces and strategies• Phase transitions in systems and populations• Complexity of agent intelligence and

relationship to evolutionary dynamics

Tipping Points

• Phase transitions and catastrophes• Second order transitions

– very long correlation lengths– critical slowing down– increased variance

• Mutual information peak– almost universal indicator

Rain Man

• Film starring Dustin Hoffman and Tom Cruise• Features high functioning autistic (DH)• Exhibits striking savant abilities

– counting, subitisation• Problems with human relationships

– theory of mind often a problem in autism

Reorganisation

Manacled by Mindsets

• Concepts block access to detail (LATL)• Release savant skills with TMS, tDCS

• Centre of the Mind (Allan Snyder)– Numerosity (inspired by Rain Man)– Change blindness– Proof reading– Absolute pitch…

• Building a better brain?

The quick brown fox jumps over the

the lazy dog

Patterns and the Brain

• Autistic savants– exceptional detailed pattern memory– eidetic imagery– numerical (casino) skills

• Expertise– 10,000 hours, 50K – 200K chunks– human expertise dominated by pattern

memory– subtle. Not eidetic.

Go• Most difficult known game for computers• Interesting problems in local-global order• Huge search space – intractable• Human expertise different to computer

– Strong use of pattern memory (we think)– Marvin Minsky conundrum

• People get better the more they know, machines get slower.

Avalanche Joseki

Capture

Liberties

A Single Eye

Suicide

Capture

Two Eyes

Avalanche Joseki

Studying Go Patterns

• Use Go knowledge to select key patterns– Joseki and fuseki

• Study variations from expertise– Two levels (amateur and professional)– Up to 9 dan levels in each– 9 Dan Professional, effectively grand master

• Find probability distributions on moves

Move Distributions

• Ten moves found to be enough• 9 Dan tend to be a bit less diverse in move

options• Middle ranks in between beginner and 9 Dan

Local Global Order

• Comparing sparse positions in game with all– Early positions involve global judgement

• The divergence measure between each player rank and 9 Dan Professional shows no change until 1 Dan Amateur

• Implies very little global understanding before several years of serious tournament play

Transition to Expertise

• Measuring the divergence between ranks shows a peak around 1 Dan Professional

• Since performance is increasing uniformly without any sharp changes, it implies this is a reorganisation of knowledge rather than the learning of new techniques or strategies

• See M. Harre , T. Bossomaier, C. Ranqing, and A.W. Snyder. ́The development of human expertise in a complex environment. Minds and Machines, 21:449–464, 2011.

Game Tree Analysis

• 8,500 starting corner positions– About 2,000 games

• Compute game trees 6 pli deep• Compute entropies on

– Ordered sequences of plays– Unordered (static positions)

• Compute Mutual Information– Real indicator of phase transitions

The Phase Transition

• The game tree analysis shows a peak in Mutual Information at 1Dan Professional.

• This is a strong indicator of a second order phase transition.

• See M. Harre , T. Bossomaier, A. Gillett, and A.W. ́Snyder. The aggregate complexity of decisions in the game of Go. European Physical Journal B, 80:555–563, 2011.

Perceptual Templates

• To further understand the phase transition, a large number of game positions and moves were used to compute a Kohonen Self-Organising Map.

• The maps were thresholded to create a set of several thousand perceptual templates

• The amateur and professional templates are substantially different

• See M. Harre , T.R.J. Bossomaier, and A.W. Snyder. The ́perceptual cues that reshape expert reasoning. Nature Scientific Reports, 2(502), 2012.

•

Creativity

• Many forms – replacement (eg Dali Lobster phone)– random acts (Dadaism)– bottom up (Jackson Pollack)

• Deep creativity changes the foundations– Bach (equal temperament), Einstein (relativity)

• Strong parallel between expertise and deep creativity

The Autistic Genius Idea put forward by Grandin, Fitzgerald, Baron-Cohen and others, that great thinkers and creative minds of the past may have been autistic/Asbergers

…It seems that for success in science or art, a dash of autism is essentialAsberger (cited by Baron-Cohen)

Asberger Geniuses?

• Science: Einstein (Nobel Prize)• Poets: Yeats (Nobel Prize)• Philosophy: Wittgenstein• Computation: Wiener• Politics: Keith Joseph (Cabinet minister)

From Michael Fitzgerald:Autism and Creativity

Words strain,Crack and sometimes break, under the burden,Under the tension, slip, slide, perish,Decay with imprecision, will not stay in place,Will not stay still. -– T.S. Elliot

The Paradox of PoetsHow can an autistic without a theoryof mind be a poet? But poets workwith sound and rhythm.

Complexity and Mindquakes• Fundamental changes in the way we think may arise from low level play

– Tinkering with the building blocks• Complexity theory emphasizes

– unpredictable emergent phenomena– big system outcomes from changes at low level

Tinkering with the Foundations

• Music: – Bach (equal temperament)– Wagner (chromaticism)– Schonberg (12 tone serialism)

• Physics:– Einstein (speed of light)– Planck (quantisation)

• Art: Breton, Dali (surrealism)

Computers and Creativity• Support for human creativity

– Simulating upwards from low level changes– Searching for counter examples

• Computer creativity– Building modular hierarchies with interchangeablility – Teaching software agents to play– Music synthesis for computer games– Scenario modelling for security etc.

Games with (more) ToM

• Go involves very little Theory of Mind (ToM)– Bridge, Poker require judgements about players– A lot of online work in Poker (gambing driven)

• Video games (MMOGs?)• Real life

– Transitions in medicine– Financial trading

Acknowledgements

Michael Harré was funded by an Australian Research Council Discovery Grant, DP0881829, to Snyder, Bossomaier and Harvey

Envoi

• Expertise goes through tipping point in Go– a general characteristic– applicable to ToM tasks too?

• The savant brain has advantages– can we get the best of both worlds?

• Next generation AI?