Another strike against the “broken mirror” theory of autism

The “broken mirror” theory of autism, most famously explained by famous neuroscientist V. Ramachandran, argues that autistic people have difficulty understanding other people’s intentions at a higher level because at a lower level, they do not perceive other people’s gestures the same way others do.

"Mirror neurons,” first discovered in monkeys, are neurons that fire both when we perform an action ourselves and when we see someone else performing the action.  They respond to specific actions, and not just patterns of stimuli—one group of neurons will fire for reaching, another group for grasping, etc.

The idea is that we process other people’s actions by mentally simulating them, by generating the same brain state we would have when performing the action ourselves, thus literally “putting ourselves in someone else’s head.”  Since autistic people do not automatically understand other people’s mental states from observing their behavior (they often have to be told what others are feeling), it would seem to follow that something has gone wrong with their mirror neuron system.

Notice that this theory assumes that imagining yourself performing an action is all the information you need to take someone else’s perspective.  But a 2009 study by Stewart Mostofsky and colleagues at the Kennedy Krieger Institute suggests that not only is this information insufficient, it may actually be counterproductive.

Dr. Mostofsky had 14 autistic 10-12 year olds and 13 typically developing 10-12 year olds learn to use a novel device, a robotic arm which they held in their hand and reached with to capture “animals that had escaped from a zoo.”  The robot produced a force field that perturbed the children’s arm movements, and the children had to learn to adjust their movements to control the tool and capture the animals. Notice that learning to use the robotic hand required mastering two sets of cues: visual cues (watching the arm move closer to the animal) and proprioceptive cues (the feeling of the force field and the feeling of the body when in the correct reaching position).  Proprioception is the body sense, the feeling of the position of muscles and joints.  Dr. Mostofsky wanted to know whether autistic as well as typically developing children could learn from both sets of cues.

Dr. Mostofsky explains what a defender of the “broken mirror hypothesis” of autism would predict:
When we observe another person performing a movement, the internal models to execute the same movement may also be activated in our brain… Indeed, after volunteers observe another person reach while holding a robot that is producing a force field, they perform better than naive volunteers if they are tested on the same field. This is consistent with the hypothesis that observation of an action instantiates the same internal models that are required for production of that action.
In other words, when typically developing people watch someone perform an action (visual input), this activates their internal model of performing the action, which includes what it feels like to perform the action (based on proprioceptive input).  But what if autistic people can’t use visual cues to activate their proprioceptively-based internal models?
Because this instantiation relies on visual cues, internal models that place a greater than normal reliance on proprioception, while discounting visual consequences, might place the observer at a substantial disadvantage in understanding other people’s actions and imitating their movements. To test our hypothesis, we looked for correlations between how the children represented our simple reaching task and clinical measures of motor, imitation and social function.
 Mostofsky found that typically developing children generalized from both the visual and proprioceptive input, while autistic children generalized ONLY from the proprioceptive input.  They did so twice as strongly as the typically developing children.  In other words, autistic children were not relying on visual information to learn to use the new tool, but they were relying extremely heavily on their body sense.

Furthermore, the more children generalized from proprioceptive cues, the more impairments they had in general motor function, social interaction, and the ability to imitate others’ gestures.*

(This study would have been stronger if they had had autistic children observe people using the tool before attempting using it themselves.  This would have directly demonstrated whether autistic children can learn from visual cues in the absence of proprioceptive ones, the way typically developing children do).

Be that as it may, these autistic children had a very strong sense of what it felt like to perform these movements, so one would think their ability so mentally simulate these actions would be intact. In other words, their mirror neurons were probably fine.  Yet they had difficulty understanding and imitating other people’s gestures.  So what was missing?  Not mirror neurons, but the ability to learn from visual cues.

Haswell, Courtney C., Izawa, Jun, Dowell, Lauren R., Mostofsky, Stewart H., & Shadmehr, Reza (2009). Representation of internal models of action in the autistic brain. Nature Neuroscience.

*This study clarifies a puzzling earlier Mostofsky study I blogged about, which found that the more participants relied on proprioceptive input, the worse they did on a motor learning task.  Most likely, proprioceptive ability does not interfere with performing motor tasks (that would be ridiculous).  Rather, large amounts of it seem to accompany poor visual ability, and that lack of visual ability interferes with performing motor tasks. (Perhaps the unusually high proprioceptive ability compensates for the lack of ability to learn from visual cues?)


  1. Autism and compensatory are not normally used in the same sentence. That in itself is novel. Thanks.

  2. Sorry, I meant post. Nobody can see how far my son has come. He has to be enormously strong in character. They look at the outside and judge. If autism is a disability, there must be compensatory strategies. No one looks for them.

    I'm so tired of the crap that comes out of cambridge. This is SBC's next obsession: Will Autistics have a moral foundation? Only Simon can tell us. How do I know? I copied this directly from the volunteer link I was given. He finds his subjects on the internet. That's pure science, baby.
    Moral Foundations Questionnaire
    How do you decide when something is right or wrong?
    Moral Foundations Sacredness Scale
    How much would we have to pay you to make you do something?
    Levenson Personality Questionnaire
    See whether you agree or disagree with these statements
    Sympathy Tasks
    Tell us to what extent you have sympathy for these people

    He is such a bully.

  3. Thanks for commenting, glad you liked the post! I didn't 100% understand your 2nd comment but I'll take a stab at it, and you can tell me if I missed anything. :)

    So glad to hear your son is learning to compensate. My brother has done the same. It frustrates me that when people like them haven't compensated enough, people focus on their disability, but if they compensate successfully, people say they can't possibly be disabled. Kind of a no-win situation, isn't it?

    SBC is trying to study morality now? ::cringe:: Is he studying NTs too? Let me guess, he thinks people with "male brains" are moral for rule-based reasons ("It should be this way," "these are my principles") and "female brains" are moral for empathy-based reasons ("it makes people happy" or "it helps people?"). I wonder if he'll go as far as to say women are utilitarians and men are Kantians? 0.0

    Many psychology researchers do legitimate research online (MechanicalTurk through Amazon is a particularly cheap and popular source of participants from around the world). I wouldn't trust SBC to do ethical online research, though. (But then, I can't think of a single area of research I'd trust him to conduct ethically).

  4. I enjoyed the post but I just want to clarify the aim of the Haswell study (as I interpreted it).

    Haswell et al. compared generalization in extrinsic (hand-position, roughly) vs. intrinsic (joint-space) coordinate reference frames. They did not compare reliance on visual vs. proprioceptive cues and so, unfortunately, we cannot infer the childrens' relative preference for proprioceptive versus visual information from this study.

    E.g. "The training took place in the left workspace (target 1; Fig. 1a) while a velocity-dependent field pushed their hand perpendicular to the direction of motion. [They] quantified generalization in the right workspace in the intrinsic coordinates of the arm (target 3, identical joint rotations as compared to target 1), and in the extrinsic coordinates of the task (target 2, identical hand motion as compared to target 1)."

    In effect:
    (1) Children with ASD showed weak generalization when the hand path was identical but joint rotations differed from the learning phase (compared to control group)
    (2) Children with ASD showed strong generalization when hand path differed but joint rotation was identical to learning phase (compared to control group)

  5. Thanks for clarifying!

    You're right, the study only discusses generalization to either intrinsic or extrinsic coordinates, there was no difference in visual input, and the major input during training was proprioceptive (a forcefield). Someone--either myself or the authors--associated intrinsic coordinates with proprioceptive processing and extrinsic coordinates with visual processing. It makes sense to think of extrinsic coordinates as requiring processing other than proprioceptive, because it involves perception of things outside our bodies, and by definition, proprioception only involves our bodies. But you're right, going from "internal vs. external coordinates" to "visual vs. proprioceptive learning" is a generalization.