Theory of Mind

It’s Spring(!) which means it’s officially time to begin deciding whether to apply to grad school some time in the next few years. My plan was to start with the Saxelab reading list, but it is short and unsurprisingly heavy on theory of mind. At this stage, I would rather survey a wide range of topics I’m interested in before getting too deep into any one. That said, it’s always hard to know where to jump into any field, and I already had a curated list, so for now I picked the oldest theory of mind paper, Domain specificity in conceptual development: Neuropsychological evidence from autism by Leslie and Thaiss (I was going to link to it, but I don’t want to anger the academic publishing Gods. You definitely can and should find it yourself). I’m realizing how rusty I am at reading academic papers, so bear with me.

Although I’m sure no one ever said this explicitly while I worked in the Saxelab, I sort of assumed that Rebecca Saxe was the first person to propose a neural mechanism specifically for theory of mind following her fMRI studies. Leslie and Thaiss proposed the existence of “specialized cognitive mechanism” for theory of mind prior to doing any brain imaging. But hey, people also knew the Earth was round before Columbus. And like Eratosthanes, Leslie and Thaiss deserve a lot of respect for a resourceful way of showing it.

The concept of Theory of Mind, that people can infer and reason about the mental states of others, has been around for ages (probably since before Eratosthenes, knowing those Greeks). But according to Leslie and Thaiss, around the 90s, the leading model was Representational Theory of Mind (RTM) where people (or at least preschoolers) imagine others’ mental states as representations of the world, like photographs or diagrams that can be true or false. It’s an alluring model because it plays nicely with false belief tasks, the most common ToM test. In false belief tasks, like the famous Sally-Anne test, one agent usually places an object somewhere and leaves the scene, then a second agent moves the object, and the participant is asked where the first agent thinks it is. It’s a convenient setup because the participant’s answer reveals whether they’re considering the agent’s limited knowledge.

Leslie and Thaiss gave groups of autistic and non-autistic 4-year-olds these false belief tasks along with “false photograph” tasks, where a photo was shot of the world in one state (with a puppet sitting on a bed) and later the world changed (i.e. the puppet got off the bed) so that the photograph represented a false state of the world. It was known that autistic children are slower to develop Theory of Mind since (according to Wikipedia) Baron-Cohen et al. 1985 (cousin of Sacha Baron-Cohen). (This is one of the main parts of autism). If RTM was true, then we would expect autistic kids to lag non-autistic kids in the false photograph task as well.

But this is not what happened! As expected, autistic kids were worse at the false belief tasks, but they outperformed the other group on the picture task. Leslie and Thaiss infer that this means there is a cognitive mechanism separate from an understanding of representation that is dissociably affected by autism.

In the discussion, they really get into the shortcomings of common-sense RTM. “The temptation is to surreptitiously supply the RTM with the concept of belief: to say that the picture-in-the-head causes behavior because the agent believes what the picture says!” (Emphasis and excitement from the original). Different people with identical representations of the world could have vastly different relationships to those representation. That is, I could believe representation X while you hope X and someone else pretends X. Even one person could believe that it’s raining but hope it’s not. And you could recognize that when considering their mental state.

On the whole, it seems like RTM is a theory born from an overuse of false belief tasks. There are tons of ways to think about other minds including inferring emotion and intent. Maybe false belief was just much easier to test before brain imaging. Or maybe it wasn’t even clear that these would all be handled by the same region of the brain before imaging. (This whole post is just a front to get people to watch this lecture).