In this post, I argue that Model Theory is a superior account of the broader conception of mindreading laid out in the previous post. Thus far, I have refrained from discussing Theory Theory (TT) and Simulation Theory (ST) even though these theories have been the two main general theories of mindreading for decades. The time has come to very brieflyexplain these theories. So, here goes. Theory theorists hold that our capacity for mindreading is underwritten by an information-rich body of folk psychological information that we employ to infer others’ mental states and predict their behavior. The TT holds that theoretical inferences play a foundational role in developing the capacity for mature mindreading, and they continue to underwrite our ability to explain and predict others’ behavior even in adults. And though other sorts of cognitive processes, such as simulation, may play a role in mindreading, these other processes are subsidiary to theoretical inferences.
In contrast to the TT, simulation theorists argue that simulational processes are developmentally fundamental in the sense that without simulation, we would not have imitation, joint attention, and empathy, which are the building blocks of mature mindreading. Furthermore, ST holds that simulation processes continue to underwrite our developed capacity to explain and predict others’ behavior as adults. Hybrid ST theories allow that theoretical processes may supplement simulation, and in some cases theoretical inferences may be more appropriate, but simulational processes are both more foundational and more common than theoretical processes.
Many readers will have heard about TT and ST for years and have the impression that the debate is stagnated and people have basically settled on some sort of hybrid version of these theories. This is true, but what I think the previous posts show is that the conception of mindreading TT and ST aim to explain is too narrow. General theories of mindreading, and even theories in specific debates about mindreading, tend to assume that the only mode of mindreading (worth discussing) is accurately attributing mental states in order to accurately explain and predict a target’s behavior. That is, mindreading theories aim to explain how this capacity to accurately attribute mental states to explain and predict behavior evolved, how young children develop this capacity, what underwrites adults’ capacity to accurately attribute mental states, etc.
Accurately explaining and predicting behavior is just one small part of our mindreading abilities. The way in which we socially categorize people, whether we perceive others to be relevantly similar to us, the biases we bring to bear on a situation, the effects of situational context, our goals in a social interaction, the mindreading strategies we adopt, and the kind of explanation mindreading produces all can vary. This makes for an incredibly diverse set of mindreading practices. Importantly, these various phenomena reflect diversity in mindreading itself, not just how we use subsequently use mindreading. Most contemporary versions of the TT and the ST do not discuss these phenomena even though they are directly relevant to understanding the processes of mindreading. In light of the broader conception of mindreading articulated in the previous post, it’s clear that we need to revise our general theories of mindreading.
In the book, I argue for Model Theory as a superior general theory of mindreading. Model Theory was originally proposed by Heidi Maibom (2007, 2009, 2003) as a more promising hybrid version of the TT. Peter Godfrey-Smith (2005) developed a slightly different version of Model Theory which was built upon his prior work on the use of models in science. In general, Model Theory conceives of mindreading as a form of theoretical modeling. As such, Model Theory is a version of the TT.
Model Theory holds that mindreading consists in deploying a model psychological profile of a target. The kind of models posited by Model Theory are conceptual models, which involve systems of related concepts. Models are hypothetical representations that specify a general structure, relations, and properties of some phenomenon. These hypothetical representations can be used to represent more complex phenomena. For mindreading, we construct and apply simplified model psychological profiles of targets in order to understand complex social interactions. There is a basic folk psychological model, which consists in a distinction between beliefs and desires, the idea of sensory input and behavioral output, and characteristic dependence of action on perceptions, memories, goals, and temptations (Godfrey-Smith 2005, 10). Elements of this core folk psychological model are innately specified, which explains why basic aspects of folk psychology appear to be common across cultures (Malle 2008).
The core folk psychological model can be elaborated in various ways and is the basis for many culturally and even individually specific models. We may construct specific model psychological profiles for different social groups and perhaps even specific individuals. Over time and with experience, we come to incorporate into our folk psychological models knowledge of different kinds of social norms, institutions, and social roles, individualized knowledge about a particular person’s history and personality, knowledge of stereotypes, social biases, etc.
The mindreading models we employ can be more or less elaborate, with some being mere schema we generate on the fly and others being detailed representations of individuals and social groups. Folk psychological models may be explicit and deployed deliberately, such as in scientific interpretation, but they need not be. The models may be implicit, and the agent using them may not be able to describe them in any great detail, and thus may not be able to articulate the similarities between a model and the target phenomenon. In such a case, the result of the interpretation is that the agent simply sees the target phenomenon as an instance of a relatively familiar model.
So far, I have described what folk psychological models are and emphasized their diversity. There is also variety in how we employ these folk psychological models. We put these models to different uses depending on our interests and the context. We can apply detailed models carefully, checking to see if the model matches the behavior of the target. One primary factor in determining model use is whether something important hangs on mindreading correctly, if it matters to you personally, or if the situation is highly unusual. In these cases, you will tend to use a more detailed model in order to generate more accurate model of the other person’s mind. You may use this model to explain a target’s behavior, make predictions, manipulate the target’s behavior, or all of the above, depending on your interests in the situation.
We can also apply mere schematic models in a quick and simple way. In cases where efficiency matters more to you than accuracy, you will use more schematic models in order to have a close enough model without having to deliberate very carefully. The schematic models you use depend on whether the person you are mindreading is perceived to be part of an in-group or an out-group. In the former case, you will tend to use a model roughly based on what you take yourself to think, feel, and do in various situations. In the latter case, your model will be based on relevant stereotypes. You may use these schematic models to categorize the target (e.g., as characteristic of the relevant social stereotype or as like me), explain behavior (e.g., formal or teleological explanations), or make predictions, depending on your interests.
When your motives are more self-serving, what varies is not so much the level of elaboration of the model but how you use the model to cement preconceived ideas about oneself, one’s in-group, and various out-groups. In these cases, the elements of the model heavily emphasize patterns of behavior in line with your existing values and beliefs. Typically, in such cases, the model is used to explain a target’s behavior – in order to justify, rationalize, condemn, or dismiss it – rather than predict or manipulate the target’s behavior.
As I mentioned above, the Model Theory is a version of the TT, but it has an advantage over the ST and many other versions of the TT. In particular, it is well poised to explain the diversity in modes of mindreading and the psychological processes that underlie these various modes, but it also explains the diversity of input to mindreading, the goals that mindreading serves, the different kinds of products of mindreading, and the conditions under which mindreading is likely to be accurate. Most existing hybrid theories are in principle capable of explaining some of this diversity, but in practice they only address a very limited range of mindreading processes, and it is an open question whether and how we can alter various existing theories of mindreading to explain all these various features of mindreading. Thus, a distinctive benefit of the Model Theory is that it already explains the diversity of mindreading.
That said, Model Theory may actually be compatible with some existing theories of mindreading – most likely other versions of the TT. For example, it seems plausible that we could explain how we learn to construct and employ various mindreading models with a Bayesian or predictive coding model of mindreading. It may take significant work to actually make these theories cohere well in practice, but I take it to be good in principle because this could expand the explanatory scope of the theories. For instance, Evan Westra (2017b, 2017a) offers an a predictive coding account of mindreading that I think is a promising candidate for combining with Model Theory. Thus, I am not so concerned with establishing that Model Theory is incompatible with every other existing theory of mindreading as I am with constructing an empirically plausible, philosophically sound general account of how we understand and interact with others.
References
Godfrey-Smith, P. 2005. “Folk psychology as a model.” Philosophers’ Imprint 5 (6):1-16.
Maibom, H. 2003. “The mindreader and the scientist.” Mind and Language 18 (3):296-315.
Maibom, H. 2007. “Social systems.” Philosophical Psychology 20 (5):557.
Maibom, H. 2009. “In defence of (model) theory theory.” Journal of Consciousness Studies, 16 6 (8):360-378.
Malle, B. F. 2008. “The fundamental tools, and possibly universals, of human social cognition.” In Handbook of Motivation and Cognition Across Cultures, 267-296. San Diego: Academic Press.
Westra, E. 2017a. “Character and theory of mind: an integrative approach.” Philosophical Studies.
Westra, E. 2017b. “Stereotypes, theory of mind, and the action-prediction hierarchy.” Synthese:1-26.
Thanks Shannon, this is really fascinating. Do you mind if I ask about your final paragraph? You say that model theory might be compatible with other versions of TT; I was wondering about its compatibility with ST. In particular, I’m wondering to what extent it makes sense to think of models, of groups or of people, as something like ‘recipes’ or ‘instructions’ for how to simulate: e.g. my model of a group including ‘they like pasta’ consists in a stored instruction to simulate liking pasta when I simulate their minds, and my model of an individual being aggressive consists in a stored instruction to ‘pump up’ any simulated anger or aggressive feelings when I’m simulating them.
(I guess I’m also sort of wondering how far all the theories in this area are always addressing the same questions: what if a simulation theorist said something like ‘being a simulation theorist says nothing about how the information that informs and guides simulation processes is stored and organised. It’s very plausible that it’s stored and organised in the form of models, some very specific for individuals and some much broader for groups, and that these models are intertwined with social emotions and sympathies so that, e.g., the models for members of an in-group will tend to be deeper and more detailed, while the models for members of out-group will be simpler and more caricatured – but that’s all compatible with thinking that what we use the models for is (primarily, or at least often) simulation’?)
Hi Luke. Thanks for your comments. In the book, I argue that folk psychological models may involve simulational elements (e.g., projection or explicit, deliberative simulation) and in that sense the Model Theory doesn’t exclude Simulation Theory. One *could* take all the other phenomena I describe and try to fit it into a Simulation Theory model. However, it seems that the phenomena I describe are a more natural fit for Theory Theory types of accounts simply because the latter are more flexible. Theory Theories posit information-rich processes for understanding other people. The variety in folk psychological models and how we employ them seem best captured by theories that allow for such information-rich processes. One of the primary benefits of the Simulation Theory is that it’s supposed to be information-poor and rely only on offline use of our own cognitive machinery. Thus, it would be hard to make room for the phenomena I describe in the book without sacrificing this asset of Simulation Theory.
With all that said, one can have the traditional ST/TT debate with respect to all the phenomena I describe in the book. I’m not that excited about having that debate. I’d much prefer to think about, for example, the nature of these folk psychological models, or how to combine work on heirarchial predictive coding in mindreading (see Westra, 2017b above) with the models I posit. But if others want to take these phenomena and revamp the traditional theories, go for it! There’s a lot more work to be done all around given the broader conception of mindreading.
Thanks, and yeah I can see why rehashing an old debate might not be too exciting, and certainly you’re right that there’s a lot more work to be done to ‘revamp’ ST. But for what it’s worth, I’d be inclined to think that ST is well-placed to capture a broader, more pluralistic sort of mindreading insofar as it links mindreading to the same cognitive bases as imagination, pretence, engagement with fiction, etc., which are clearly oriented towards a much broader range of goals than predicting and explaining.
Also, do you mind if I ask about the way you’re thinking of ‘information-rich’, ‘information-poor’ here? On the face of it, a lot of the stuff about stereotyping and group categorisation seems to be about using informational shortcuts – e.g. I notice X’s eye colour, I assume X’s political affiliation, or whatever. But that’s also based on a lot of prior learning – seeing that eye colour associated with that trait repeatedly, etc. Would you think of that as an ‘information-rich’ process, despite being so ‘quick-and-dirty’?
Hi Luke. That’s a good point about ST’s connection to imagination and related cognitive phenomena. I’ve never been sold that ST has a real advantage explaining imagination and related phenomena, at least not once we allow for hybrid theories, but it is a nice way to link in a broader set of goals.
On the information-poor idea, ST is supposed to simply re-use one’s own cognitive machinery for the purpose of mindreading. So, for figuring out what decision a target will make, you take pretend inputs and run them through your own decision-making mechanisms (whatever those are), and the output is attributed to the target. One advantage is that you don’t have to rely on folk psychological information about what other people think, feel, and do in various scenarios. You just have to figure out what *you* would do in that scenario. It is in that sense that ST is supposed to be information poor. TT in contrast holds that you infer what others think by based on a lot of folk psychological information about how people think, feel, and behave.
If I am simulating to figure out what a target will decide to do, stereotypes about the target’s social group shouldn’t play a role in mindreading because they don’t play a role in my own decision making. (It’s a different story for projection and egocentric heuristics in general.) In contrast, stereotypes, implicit associations, generics, etc. easily could fit into the TT framework.
I hope this helps!