Metacognition and the social construction of realityy Ch i F Chris Frith ih All Souls College, Oxford Aarhus University, y, Denmark University College London

What it Means to be Human New Perspectives on the Boundaries of Human Nature British Academy, 8 March 2012

Cognition: Thinking Metacognition: Thinking about thinking A high-level cognitive process that monitors and controls lower-level cognitive processes

Metacognition while typing Skill d ttypists Skilled i t slow l ddown after ft making ki an error

Monitoring: the error is detected Control: typing speed is reduced to avoid further errors

An evil demon (psychologist) inserts errors the typist has not made and corrects errors the typist has made

Illusions of authorship in skilled typists

Behaviour: the typist does not slow down after inserted errors, but does slow down after corrected errors Awareness: the typist believes she made inserted errors and believes she did not make errors that were corrected

Two kinds of metacognition Implicit: rapid rapid, automatic, automatic without awareness Explicit: slow, deliberate, prone to error

Explicit metacognition is prone to error b because we hhave little littl or no conscious i access g pprocesses to our cognitive Nisbett & Wilson, Psychol Rev 1977

We still justify our choice even if we didn’t didn t make it Which face is more attractive?

choose

explain l i choice h i

On 20% of trials the face was switched between choosing and explaining 74% of these deceptions were not detected Now shown for jam as well as faces Johansson et al. Science 2005

E li it M Explicit Metacognition t iti We seem unaware of how little conscious access we have to our cognitive processes So we are happy to make up stories about these processes

What is explicit metacognition good for? Do we gain any advantages from reflecting upon our actions?

We have an experience of agency which hi h li links k actions ti to t outcomes t Event 1

250 msec

Tone

Physical time

Voluntary action

Tone

Mental time

Sham TMS

Tone

Involuntary action

Tone

Intentional Binding

Haggard et al., 2002

Wh t iis meta-cognition What t iti goodd ffor?? Our experience of agency is constructed after the action has occurred Libet et al 1983 Haggard et al 2002

Action A i is i linked li k d to outcome, creating i the h experience p of agency g y and responsibility p y (although this can be erroneous) Wegner 2003

Facilitated communication can lead to d l i delusions off agency

The facilitator believes that the child is determining the actions

What is meta-cognition good for? Reflection R fl ti on action ti has social consequences It can generate a feeling of agency and responsibility for self and others we only o y pu punish s peop peoplee co considered s de ed to be responsible for their actions e.g. Singer et al, Neuron 2004

What h is meta-cognition goodd for? f ? Reflection on action permits the discussion of motivations and strategies t t i for f decisions d ii which affect future behaviour

Discussions of the experience of action alter lt subsequent b t behaviour b h i Limited resource: “Working on a strenuous mental task can make you feel tired such that you need a break before accomplishing a new task” Non-limited N li it d resource: “Sometimes, “S ti working ki on a strenuous t mental t l task t k can make you feel energized for further challenging activities”

Job et al, Psychol Sci, 2010

Through Th h our attempts to monitor i our actions we can justify our behaviour to others Our experience of action can be altered through g discussion IIn thi this way we can d develop l group (cultural) norms for making decisions

Our understandingg off concepts p like free will and will power h have cultural l l aspects acquired i d through h h discussions with others such discussion depends upon metameta cognition: our ability to reflect upon our actions and discuss their origin with others

What else is meta meta-cognition cognition ggood for? f Reflection on sensation It p permits the discussion of sensory y experience as well as action

By sharing their experiences people can acquire a more accurate account of the world

Perceiving the world together

Interval 1

First S Second d

Joint D i i ? Decision?

Individual decisions declared

group decisions required

Correct Correct Wrong

Interval 2

Individual decisions made

feedback

Bahrami et al. Science 2010

Are 2 heads better than 1? Yes The joint decision is better than the best individual decision

How can subjects resolve discrepancies in their perceptions? 1) Toss a coin: performance of dyad is mean of individual partner performance 2) Choose the response of the more competent partner: performance of the dyad is equal to that of the better partner 3) Choose the response of the more confident partner on a trial by g confidence f sharing): g) pperformance of trial basis ((weighted the dyad is better than either individual ((confidence is estimated from the individual ppsychophysical y p y curves)) Bahrami et al. Science 2010

Weighted Confidence Sharing gives i the th best b t fit

sdyaad / smax

1.6 1.4

WCS

1.2

DSS

1

Experiment 1

0.8

E Experiment i t2 Regression fit

0.6 0.4 0

0.2

0.4

0.6

smin / smax

0.8

1

Participants described to each other their confidence in what theyy had seen (translated from the Danish)

certainty Group 43 to be sure to know

to believe

intuition

t guess to kind of fifty

saw it well saw it alright saw think I see it thought I saw doubt that I saw anything didn’t see it don’t think I saw anything couldn’t see didn’t see shit seriously, I didn’t see anything don’t see anything didn’t see any difference I see nothing didn’t see a thing only saw a blink

N = 16 pairs Data: 1470 short interactions ≈ approx. 20 hours of video

Group 21 sure not 100 % sure more sure almost sure 55 % sure sure – a little a little uncertain not quite sure not sure unsure far from sure damn – I’m not sure very unsure fucking unsure too unsure totally unsure

uncertainty

to see to know to think to believe

in doubt difficult fift fift fifty-fifty to be lost

Kristian Tylén & Riccardo Fusaroli, Aarhus University

Thee be benefit e t of o discussion d scuss o occurred occu ed even eve when the partners were not told whether th were right they i ht or wrong

Bahrami et al. Science 2010

Discussion also improved p individual performance, especially without feedback

Senssitivity [Slope]]

„INTERACTIVE NO FEEDBACK N 30 N=30 „ INTERACTIVE

N=28 „ NON-INTERACTIVE

N= 50

Block number [64 trials]

Karsten Olsen Aarhus University

The advantages of explicit metacognition Discussion of experience enables groups to achieve more accurate estimates of the state off the h world ld This advantage occurs even in the absence of independent evidence about the world Discussion of experience also improves the abilities of individuals

The nature of explicit p metacognition g Since we have no direct contact with our own cognitive processes, the contents of explicit metacognition are extremely malleable and responsive to social factors.

At the beginning of our life, the content of explicit meta-cognition is a blank slate on which we learn to write our experiences. What we write there is determined by discussions with others. others We develop shared views of the world and of ourselves, which develop within each life time and which evolve across the generations to form cultural norms and beliefs.

The social construction of reality By sharing our experiences we can create explicit models of our physical and our mental world. Throughh our interactions Th i t ti with ith the th world, ld the th models can become increasingly g y accurate but we can also create mass delusions

What does it i mean to be human?? Humans can develop new communication systems on the fly These communication Th i ti systems t are tools for jjoint action and change our perception of the world

What does it i mean to be human??

O perception Our i off the h world ld is i shaped h d by b social interactions involving reflective discussions that depend on explicit metacognition Explicit p metacognition g is a component p of social cognition th t is that i uniquely i l human h

Queen Square Bloomsbury

All Souls, Oxford

Peter Latham

Geraint Rees

Bahador Bahrami

Karsten Olsen

Thank you!!

Andreas Roepstorff

Kristian Tylén

Riccardo Fusaroli

Mind Bridge Project

42

What we say is not what we do Each face in this set is composed p of 9 continuously varying features Volunteers sorted the faces into groups Statistical analysis gave an objective measure of which features were used The features the volunteers said they used did not always correspond Our lack of knowledge does not deter us from explaining and justifying our behaviour Frith & Frith, J Exp Ch Psychol, 1978

Rev Th Bayes 1702-1761 p(A|X) = p(X|A)*p(A) p(X) (X)

P(2ndd)

Data from two pairs

S1 S2 Joint

Contrast difference (C2 – C1)

M t Metacognition iti 2 Mentalising: Thinking about other people’s thinking Monitoring the mental states of others Manipulating the mental states of others

We attribute mental states to others what h t th they k know, thi think, k feel, f l desire... d i

this helps us to predict what they are going to do Theory of Mind

Mentalising

Machiavellianism

Sally (mistakenly) believes that the ball is in her basket

Explicit Mentalising At age four children track the beliefs of others, others they are aware of differences in beliefs and can explain how this determines behaviour in themselves and others Wimmer & Perner, 1983

Mentalising is not just for deception It also l enhances h joint j i t action ti For successful team work we need to keep track of the current knowledge, intentions and beliefs of the others in the team This keeping p g track can happen pp implicitly

We automatically track the beliefs of others Agent appears

Agent present Belief formation

Agent absent Reality change

Agent returns

condition c: false belief The reaction times of adults and the looking times of 7-month-old i f t are automatically infants t ti ll affected ff t d by b the th belief b li f off the th agentt

Kovács et al. Science 2010

We automatically track the knowledge of others

consistent

avatar

rectangle

inconsistent

It takes longer to say how many dots we see if this is different from what the avatar sees

Samson et al. JEP 2010

But honeybees (and ants) can also communicate experiences to achieve optimal group decisions During swarming scouts go out to assess the d i bilit off potential desirability t ti l nesting ti sites it

How honey bees integrate information from different sources l goodd site less it

goodd site it

Competition within a limited pool of scouts. More weight is given to the more confident fid t scouts. t

Seeley (2010) Honey Bee Democracy

H Honeybees b ((and d ants) t ) can communicate i t th the d i bilit off a new nestt site desirability it to t others th They y know when to stopp collectingg new evidence (quorum sensing) (q g) They combine information from many individuals to reach optimal group decisions

If bees can make d democratic ti decisions d ii then what is special p about human group decisions?

Effects of testosterone on group decision making

Testosterone impairs group performance, but does not effect individual performance p

Testosterone reduces relative trust in others

We automatically represent the goals of another when hen working orking together

Left key for green, right key for red (pointing direction is irrelevant)

330 msec

450 msec

Individual 2-choice

Individual 1-choice

Joint 1-choice Sebanz et al., 2003

Combining information from vision and touch in a single individual enhances perception

Ernst & Banks, Nature 2002; Alais & Burr Curr Biol 2004

What is the communicated confidence?

Global Linguistic Convergence predicts joint task performance The degree to which a dyad converges on a single set of confidence expressions rather th indecisively than i d i i l drifts d ift between b t numerous types t off expressions: i The % proportion of the most frequent confidence expression type in of tokens of the dyad.

relation to the overall sum