Comments on “From Human Reasoning to Belief”
Aliya Rumana
The holding power of dual-process theory in experimental process psychology is mysterious if we don’t consider the basic constraints on statistical testing. First, every study aims to reject the null hypothesis (H0) in favour of the alternative hypothesis (HA). Second, H0 claims that there is no difference in the cognitive process underlying observable behaviour—i.e., single-process theory (SPT)—whereas HA claims that there is. Thus, statistical testing puts the burden of proof onto dual-process theory (DPT). Rejecting SPT in favour of DPT is the most informative outcome. It’s no wonder, then, why DPT is so popular.
Failing to reject H0 doesn’t mean that H0 has been confirmed. For one, the data might have been easy to explain, such that it wasn’t much of an achievement for H0 to explain it. For another, H0 might have the capacity to explain a lot of data, such that the data wasn’t challenging for H0 to explain. Thus, failing to reject H0 only confirms H0 if it was difficult for H0 to explain the data. Making it a challenge for SPT to explain the data is also better for dual-process theorists, because it makes it easier for them to reject it. As a result, experimentalists of both persuasions have reason to challenge SPT.
There are two ways of challenging SPT: subject it to constraints (decrease model capacity) or get it to explain more complex data (increase data dimensionality). Increasing data dimensionality requires increasing the number of experimental conditions, which is costly, so the favourite approach is to decrease the capacity of SPT. There are two ways to do this: (a) assume the single-process model has fixed parameters (since free variables can more easily explain data) and (b) assume the function that maps its parameters onto observable response variables is linear (since a more flexible function could more easily explain data). From a philosophy of science perspective, a charitable interpretation of both assumptions is that they are distorting (idealising) for the purposes of statistical testing.
Mugg’s book defends a Soundboard Account, which claims that human reasoning is a complex dynamic system of functions and variables. I think it would be foolish to deny that this account is a more accurate, more literal representation of human reasoning than the process models that experimental psychologists test. After all, any model that represents human reasoning as a static system of low-capacity functions and fixed parameters plausibly distorts it. But this distortion is necessary if we expect confirmation or disconfirmation from single experiments. Mugg’s model has such high capacity that it is impossible to reject and hence, to confirm. Thus, accurate literal representation comes with a steep price.
By itself, this isn’t a problem for Mugg’s account. High-level theories like his are always impossible to confirm or disconfirm at the level of individual experiments. Instead, they get confirmation or disconfirmation at the level of entire literatures, e.g., from how well they can integrate lower-level explanations. Consistent with this, Mugg seeks to defend his account by saying that it provides the best explanation for various results in research on belief formation, implicit attitudes, and belief-like attitudes. That’s all well and fine.
What is a problem for Mugg’s account is that his target in the book seems to be a literal interpretation of the low-capacity DPTs. To be clear, I don’t deny that many experimental psychologists are guilty of taking their low-capacity models too literally. Still, philosophers of science should recognise that these models are distorted (idealised) for statistical testing. When Mugg relaxes the capacity constraints on a single-process theory, he should allow the dual-process theorist to do the same. Thus, the proper opponent for Mugg’s account is the sophisticated DPT that claims human reasoning has two separate soundboard architectures and maintains that low-capacity DPTs are idealisations of these dual-soundboards.
I wish we saw more work in Mugg’s book to articulate a worthy opponent like this. Without one, I worry we aren’t seeing a fair fight. To be fair, we do get glimpses of a worthy opponent throughout the book. In Chapter 4, Mugg assumes that it is impossible for a system to simultaneously token beliefs with contradictory contents. A dual-process theorist could avail themselves of this assumption to distinguish between two soundboard architectures: if token beliefs with contradictory contents were simultaneously tokened, the impossibility claim would entail that they must belong to two different soundboards. This would suggest that the reasoning process itself is fragmented—not just our web(s) of belief.
Once we have a worthy contender to Mugg’s single-soundboard account, it seems like this contender can co-opt any explanation Mugg gives for his account. At the end of Chapter 4, for example, he says that the Soundboard Account explains delays in negation processing by appealing to compositional operations. Likewise, at the end of Chapter 5, Mugg says, “the Soundboard Account explains why implicit measures are easily manipulated: researchers can influence mode of operation (especially the vehicle of representation) and the fragments that are activated” (p. 98). Obviously, experimental DPTs can’t claim this: they need to limit capacity by appealing to one-shot linear functions and fixed parameters only. However, dual-soundboard accounts face no such constraints. In fact, because they posit two soundboards instead of one, they can posit more simultaneous sequences of compositional operations and changes in modes of operation than the single-soundboard account can.
To wrap up, I think the framing of Mugg’s book has concealed its main contribution: that experimental psychologists have misrepresented the architecture of human reasoning by interpreting their simplistic cognitive models too literally. A more accurate literal model of human reasoning will have to attribute to it a significantly more complex architecture. Strictly speaking, then, Mugg hasn’t raised any serious objection for DPT per se, since it can easily to co-opt the soundboard account and posit multiple soundboards. Still, I think this move defuses a lot of the motivation for DPT. After all, it’s unclear why we should think an architecture as flexible as the soundboard model suggests is duplicated. Reading between the lines, we might find a better argument for his single-soundboard account: we have no special reason to think there is more than one soundboard. Again, then, the onus is on DPT.
Aliya Rumana
Department of Philosophy
University of Texas at El Paso
Reasoning and belief are different matters. Juxtaposition is a non-starter…