Seminar:           University Seminar on Cognitive and Behavioral Neuroscience (603)


Date:                February 20, 2003


Title:                 How the brain comes to think about the mind


Speaker:           Alan Leslie, Ph.D., Rutgers University


Attendees:        Herb Terrace, Co-Chair, Psychology Department, Columbia University

                        Peter Balsam, Co-Chair, Psychology Department, Barnard College

                        Peter Gordon, Department of Biobehavioral Sciences, Teacher’s College

                        Alexandra Horowitz, Cognitive Science Department, UC-San Diego

                        Herb Lurie, Psychology Department, Teacher’s College

                        Francys Subiaul, Anthropology Department, Columbia University

                        Ori Friedman, Cognitive Science Department, Rutgers University

                        Carolyn Ristau, Psychology Department, Barnard College

                        Jeff Brodscholl, Psychology Department, Columbia University

                        Dustin Merritt, Psychology Department, Columbia University

                        Nate Kornell, Psychology Department, Columbia University

                        Tammy Moscrip, Psychology Department, Columbia University

                        Lisa Son, Psychology Department, Barnard College

                        Per Hedberg, Psychology Department, Columbia University

Leslie Davidson, Department of Population and Family Health, Columbia School of Public Health

Michael Studart-Kennedy, Haskins Laboratory

Tina Li, Sergievsky Center, Columbia University

Hedy Kober, Psychology Department, Columbia University

Tamar Kornblum, Psychology Department, Columbia University

Comana Cioroiu, Psychology Department, Columbia University

Jessica Cantlon, Psychology Department, Columbia University


Rapporteur:      Michael R. Drew




Dr. Lesilie began by defining the term “theory of mind.” It is a “sexy” term coined by Premack, who did not intend it to be taken completely seriously.  In actuality, “theory of mind” is not a theory and, in any event, is not a theory about the mind.  It is a set of mechanisms that are illuminated in a classic field experiment called the Sally-Ann false belief (FB) task, in which a child is shown a cartoon of the character Sally.  Sally puts marbles in a basket and then leaves.  While Sally is away, Ann comes and moves the marbles to a box. The child is then asked three questions: Where is the marble? Where did Sally put the marble?  Where will Sally look for the marble?  If the child uses theory of mind, then she will know that the question is about Sally’s mind or Sally’s beliefs, not about the actual state of the world.  (The first two questions are to determine that the child remembers the specifics of the story.) 


Normal 4 year-olds are successful in the task, but most 3 year-olds fail.  There is, however, reason to believe that 3-year-olds do have theory of mind, because children as young as 2 can pass another theory of mind task.  In this task the child is shown a picture of a mother talking into a banana as if it were a phone.  The child is asked what the mother is doing.  By 2 years of age, children understand that the mother is pretending that the banana is a telephone. 


Both of these tasks raise the fundamental cognitive problem of theory of mind: how is the young brain able to attend to mental states, given that mental states cannot be seen, heard, or felt?  How can the child learn about mental states if he does not already have the concept, if he cannot already feel them?  How can the child attend to mental states so she can learn about them?  The answer, Leslie argues, is that children have an innate theory of mind mechanism (ToMM) that comes online during year 2 of life and allows children to represent and attend to mental states.  Specifically the ToMM provides a metarepresentation (or M-representation) that makes explicit 4 kinds of information:


1.      the agent (e.g., mother, sally)

2.      the attitude (e.g., The mother pretends, Sally believes.  This can be a class of attitudes toward the truth of a proposition)

3.      the anchor; what the attitude is in regard to (e.g., the banana, the marble).

4.      the content; what you are asserting about the telephone or marble (e.g., the marble is in the box). 


The core characteristics of the ToMM are as follows:


1.      It is a specialized component of social intelligence that operates in real-time, online interpretations of behavior.  The ToMM will be active during a conversation but not, for example, when one is reflecting on past behavior. 

2.      It is a post-perceptual mechanism of attention that is domain-specific.  It tracks mental states and is subject to dissociable damage.

3.      It is the specific basis of our capacity to acquire theory of mind.

4.      It is damaged in autism, resulting in core signs.

5.      It has a genetic basis.


The next section of the talk focused on a possible role of ToMM in autism.  Autism is a  relatively rare neurodevelopmental disorder correlated with specific genetic abnormalities.  Autism is diagnosed through a triad of related, core behavioral symptoms: behavioral incompetence, communicative impairment (delayed language acquisition, impaired use of gestures), and imaginative impairment.  These symptoms are displayed at all IQ levels, including normal levels.  To examine whether autism is caused by an impairment in the ToMM, Leslie compared the performance of autistic and normal children on the Sally-Ann task.  The groups were counterbalanced for mental age.  Twenty percent of autistic children passed, compared to 85% of normal 4-year-olds.  The effect was replicated when actors were used in the Sally-Ann task.  Autistic children also performed poorly on the “Smarties” FB task, in which a pencil is placed in a labeled Smarties candy container (in view of the subject), and the subject is asked to predict what another naďve child will think is inside the container. 


To determine whether the poor performance of autistic children is due to a specific impairment in theory of mind, Leslie constructed a control task that very closely resembles the Sally-Ann task but does not rely on theory of mind.  In the control task, the children watch an actor photograph a cat that is sitting on a chair.  The cat then moves to a bed, and the subject is asked to predict what the photo will show.  The control task is formally similar to the Sally-Ann task in that both tasks require children to infer the contents of a representation that does not reflect the current state of affairs.  The photograph task and a standard FB task were given to normal 3 and 4-year-olds and mental-age-matched autistic adolescents.  The autistic children performed slightly better than the normal 4-year-olds on the photograph task but worse than normal children on the standard FB task.  The normal 3-year-olds failed both the FB task and the control task.  The autistic children also outperformed normal children on another control task.  In the second control task, the position of a cat was marked on a map.  The cat then moved, and children were asked to predict the location of the mark on the map.   The results indicate that autistic kids can meet the general problem solving demands of false belief tasks, but normally developing 3-year-olds cannot. 


Leslie next turned to the issue of why normal 3-year-olds fail in the FB and control tasks.  His hypothesis is that their performance reflects a failure of inhibition or selection processing.  To respond correctly, children must inhibit the tendency to act based on “true belief,” or their beliefs about the true current state of affairs (e.g., “the cat is on the bed,” or “the marble is in the box.”).  To test this hypothesis Leslie compared autistic children with normal 3-year-olds who also failed the FB task, seeking to determine whether the children fail for the same reasons.  Leslie took advantage of a previous finding that adding the word “first” to the Sally-Ann task directions (i.e., asking “where will Sally look first for the marble,” rather than “where will Sally look for the marble?”) enhances task performance.  Normal 3-year-olds perfomed significantly better on the “look first” task than the standard task, but there was no improvement in the performance of autistic children.  The pattern suggests that autistic and normal children fail the standard task for different reasons.  Leslie hypothesizes that the addition of “first” helps normal 3-year-olds by increasing the salience of the false belief and decreasing the inhibitory demands of selection processing.  


If inhibitory processing is essential for success on the FB task, then it should be possible to induce failure in 4-year-olds, who usually succeed, by increasing the inhibitory demands of the task. To manipulate inhibitory demands Leslie created new FB tasks by factorially varying “desire” and “belief.”  Desire refers to whether the actor wants or does not want the object.  Belief refers to the veracity of the actor’s belief about the location of the object.  According to Leslie’s model of task performance, the subject’s prepotent desire is to identify the current location of the object.  Because the task is to predict what the actor will do, subjects must inhibit this desire when the actor either does not want the object or has a false belief about the location of the object.  There is no need for inhibition when the belief and desire of the subject are congruent with those of the actor (i.e., the actor desires the object and has a true belief about its location).  The condition with the highest inhibitory demand is that in which the actor wishes to avoid the object and has a false belief.  This condition requires “double” inhibition: the subject must inhibit her desire to pick the true location and her resultant desire to pick the false location.  Normal 4-year-olds performed progressively worse as the inhibitory demands increased.  Performance was worst on the “double inhibition” task.  The addition of “look first” to this task improved the performance of 4-year-olds, as it improved performance of 3-year-olds on the standard FB task.  That “look first” helped 4-year-olds indicates “look first” does not simply clarify task demands; the performance of 4-year-olds on the standard FB task indicates that they already understand the task demands.  “Look first” must help the children through another mechanism, such as by increasing the salience of the false belief and thus decreasing the inhibitory demands of selection processing.


Leslie concluded by summarizing his model of performance.  People act to satisfy their desires in the light of their beliefs.  They assign initial confidence levels to candidate belief contents (“the marble is in the box,” or “the marble is in the basket”), with highest confidence going to true beliefs.  They then adjust confidence levels in light of specifics of the task (“did Sally see the marble in the basket?”), and, if necessary, inhibit their desires or beliefs.  They then select the most subjectively likely hypothesis about what Sally believes.  This model describes the performance of 4-year-olds.  Three-year-olds are limited in their ability to adjust confidence levels in light of specifics.  Because inhibition is in limited supply, increasing the inhibitory demands of the task can cause even 4-year-olds to fail.




Herb Lurie: How do you explain that 20% of autistics pass the FB task?

Alan Leslie: Any disease process occurs by degree. This includes breaking your leg.  You can get a simple fracture or your leg can be completely crushed.  The ToMM can presumably be impaired to varying degrees.  The worst cases of autism cannot be tested in this paradigm because the children are completely mute. Also, as the age of the autistics goes up, they are more likely to pass the FB task.


Herb Terrace: Would you consider joint attention as example of a theory of mind?

AL: Absolutely.  But joint attention is not simply you and I attending to the same thing at the same time.  Maybe last night we both watched the BBC news.  That’s not joint attention.  Joint attention is my knowing that we both are attending to the same thing.  For me to know that we are attending to the same thing I need the concept of “attending.”

HT: Is there any information about autistic children and joint attention?

AL: One of the first symptoms that parents notice is this failure of joint attention.  Autistic children do not understand pointing, for instance.


Michael Studart-Kennedy: What is cause of the delay in language acquisition in autistic children?  Is it related to the ToMM?

AL: Autistics are markedly impaired in vocabulary.  Joint attention is necessary for development of a normal vocabulary.  Derek Baldwin showed that children must understand joint attention to identify the objects of reference in speech.  For instance, if the mother is looking at something and says “this is a blicket,” the child must understand that “blicket” refers to what the mother is looking at, not necessarily what the child is looking at.

MS-K: How do autistic kids compare to children with Williams syndrome?

AL: The Williams syndrome children are more like children with Down’s syndrome.  They do not show the same impairment in ToMM tasks. 


Alexandra Horowitz: Given that people with autism don’t engage in much social play, do you think this is the result of their deficit in theory of mind? Or conversely, is the ability to engage in social play dependent on theory of mind? What does this say about non-humans who engage in social play?

AL: The answer to the first two questions is Yes.  The last part is about animals.  Pretend play is a special kind of play, and I have theory about that, but that is a talk in itself.  There no compelling evidence that animals engage in pretend play as I define it.  But I have no axes to grind here.  If other animals can be shown to engage in social play, I will be perfectly happy for them.  But I have not seen compelling evidence of this.

AH: Can you describe briefly how social pretend play is distinguished from social play.

AL: Not all play is pretend.  Basically all pretend play is social, except sometimes you pretend on your own.  The remarkable thing about pretend play is that as soon as the child is able to pretend by herself, she is able to engage in pretend play with someone else.  This is what tell us that pretend play engages theory of mind.  If this weren’t true –and you had to engage in pretend play by yourself for some time before you could involve someone else—then there would be a completely different account of pretend play. 


Peter Gordon: Why do the autistic children treat people different than photographs? 

AL: Because theory of mind is not the whole of social intelligence.  There are all sorts of bits of social intelligence that do not have to do with representing subjective states.  For instance, there’s face perception, eye gaze detection.

PG: So they do represent the other person’s desire to retrieve the object?

AL: Understanding goal-directedness or desire seems to be the most intact part of theory of mind in autistics.  An understanding of goal-directedness seems to emerge in babies during the first year of life.  That has led me to suggest that it develops earlier, is more robust, and is phylogenetically more widespread.  So I think it is less impaired in autistics than other mental state concepts. 

PG: In the Sally-Ann task, could you enhance performance by showing Sally looking at the basket?

AL: I think you are referring to explanation tasks where children see Sally looking in the wrong place and they are asked to explain why.  There is a lot of work on this.  My guess is that this helps a little to decrease the inhibitory demands, but it doesn’t help very much.