On Inferring Mechanisms In Cognitive Science

One of the things that cognitive scientists do is look for, identify, and describe mechanisms. For example, cognitive scientists are interested in our ability (or proclivity) to ascribe mental states to others things and creatures. So, some posit a “theory of mind” mechanism. But, intuitively, there will not be a mechanism for every one of our abilities or behaviors. For example, it would be surprising if there were mechanism for driving a car. But if that is right, then we need principled reasons to think so. Or, at the very least, we need a story about why some of our abilities have mechanisms and others don’t. In this post, I’ll briefly consider four such stories. One of the take-aways will be that it is not obvious why some abilities (like driving a car) do not have mechanisms. Another take-away will be that it is not obvious what scientists mean by ‘mechanism’.

1. Evolution

Perhaps the abilities that have corresponding mechanisms are those abilities that plausibly (or, better, demonstrably) play an important role in primate evolution. For example, perhaps we should look for mechanisms of our abilities that have increased the chances of natural and/or sexual selection in primates.

The advantage of this view is that it successfully sorts the abilities to read minds and drive cars: one could have played an important role in primate evolution and the other could not have.

One problem with this view is that it might not allow us to say that humans’ linguistic ability has a mechanism. Given that so few primates have linguistic ability, it is not obvious that we can tell a compelling story about how linguistic ability played an important role in primate evolution. And so it would be difficult to give an evolutionary story for the intuition that language has a mechanism.

2. Cognitive/Neural Architecture

Perhaps the structure(s) and dynamic(s) of the networks that realize an ability will somehow determine whether we should attribute a mechanism to the ability. For example, if inhibiting or disrupting the structure and/or function or a cognitive/neural network mitigates or eliminates an ability, then perhaps we could understand a mechanism of that ability in terms of the structure and dynamic of that network.

This kind of network approach to mechanisms might be operative when scientists discuss the role of the r-TPJ in theory of mind. When the r-TPJ is disrupted, people attribute mental states differently (Young et all 2010). So, on this view of mechanism, the r-TPJ is (at least) part of the theory of mind mechanism.

A benefit of this architecture view is that it allows us to count language as a paradigm case of a mechanism-worthy ability even though it’s evolutionary history is relatively brief. After all, there are ways to disrupt neural/cognitive function such that language is also disrupted.

A potential problem for this view is that it might count almost every behavior as a mechanism. After all, there is probably a way to disrupt any behavior by disrupting neural/cognitive function. If that is right, then every behavior—e.g., driving a car—thereby has a corresponding mechanism. Even particular parts of behavior—e.g., changing lanes while driving a car—might have mechanisms. And that might violate some peoples’ linguistic intuitions about ’mechanism’. (Whether that is a good/bad reason to reject the new concept of ’mechanism’ is a separate question, of course.)

3. Causation

Another option is to say that something is a mechanism (or part of a mechanism) for some phenomena if it causes the phenomena (e.g., Glennan 2009). It need not solely cause the phenomena; it might only cause the phenomena in concert with other causes. But the key is that causation is the mark of mechanism(s)—or vice versa.

A benefit of the causation story is that it can explain why scientists would think that a theory of mind mechanism involves the r-TPJ: because the r-TPJ is among one of the causes of theory of mind.

A problem with the causation story is similar to the problem with the architecture story: it might yield the counterintutive conclusion that every behavior has a mechanism (since surely every behavior has a cause). A related problem for this view might reduce mechanisms to causation. The worry is that if mechanisms are just causes, then one begins to wonder if ‘mechanism’ has any unique explanatory contribution over and above language about causation (see section 2.3.2 of the SEP article on Mechanisms in Science for more on that).

4. Relativism

Another option is that the status of mechanisms are determined by the inquiry. That is, whether something counts as a mechanism (or part of a mechanism) will depend on researchers investigation—its scope, background asumptions, etc. There is no fact of the matter, so to speak, about what counts as a mechanism. Rather, what counts as a mechanism is relative to each investigation (or investigative paradigm).

So, for instance, when scientists find that our attributions of mental states change when transcranial magnetic stimulation disrupts the r-TPJ, they might think that that part of the brain is part of the theory of mind mechanism. If, however, they also find that the noise of each magnetic pulse causes most subjects to blink, they would not necessarily think that noise is part of the blinking mechanism. Why not? Because the researchers are trying to explain behavior at the level of neural function, not extracranial noises. So extracranial noise simply isn’t within the scope of their investigation. And so it is not a candidate for being part of their mechanism. If, however, researchers had been measuring the impact of noise on autonomic responses like blinking, then they might have included the noise of the magnetic pulse in their mechanistic explanation of the behavior that they observe.

A benefit of this relative view is that we might be able to make sense of why any particular phenomena counts as (part of a) mechanism: it has to do with the investigation of that phenomena. But a potential problem with this view (as it has been stated here) is that it is entirely unclear how we systematically determine whether any phenomena should be considered a candidate (part of a) mechanism. Exactly how are mechanisms relative to their investigation? Or, in analytic philosopher speak, how do we finish the next sentence? There is a relationship between investigation and mechanism such that [X] is a mechanism just in case its investigation….

Conclusion

The intuition that not all abilities have a corresponding mechanism was not supported by this brief reflection on the matter. But if even brief reflection fails to yield support for the intuition, then we can at least conclude that if there is support for the intuition (besides the intuition itself), then that support is not obvious. Another take-away from this might be that the promise of each story will depend on how we complete the details of each story—most of which are missing because I limit myself to about 1000 words for each post.