From Human Reasoning to Belief: An Empirical Account

Joshua Mugg’s book, From Human Reasoning to Belief: An Empirical Account draws on a vast empirical literature to develop and defend a single-process theory (SPT) of human reasoning against popular dual-process theories (DPT). In the second half of the book, it leverages its SPT to help explain various empirical results from belief and belief formation.

DPTs have cropped up in virtually every part of psychological and neural science (Evans & Stanovich, 2013; De Neys, 2021). They distinguish between Type-1 processing, which is often described as “fast”, “intuitive”, “automatic”, etc., and Type-2 processing, which is often described as “slow”, “reflective”, “controlled”, etc. The labels ‘Type-1’ and ‘Type-2’ are fashionable today because they maintain neutrality on different characterisations of the distinction. The common commitment, as Mugg points out, is that the distinction, whatever it is, “is supposed to cut cognition itself in two” (19).

However, Mugg notes that DPT has been engaged in a steady retreat for decades. At first, DPT was associated with massively conjunctive claims, like that cognition itself was cut into two, where one side consisted of processes that are fast and intuitive and automatic, etc. and the other side consisted of processes that are slow and reflective and controlled, etc. Mugg notes that these are most charitably interpreted as statistical, clustering claims (a laSamuels, 2009). This is often known as “Standard Menu” DPT.

Decades of functional dissociation studies have called this view into question: it’s proven easy to find processes that are fast but reflective, slow but intuitive, etc. So, dual-process theorists have retreated to what I’ll call “Essentialised” formulations of DPT: cognition itself is cut into two, where one part has one property, and the other part lacks that property. For example, Evans suggests that Type-1 processing doesn’t involve working memory, while Type-2 does (Evans & Stanovich, 2013). Since Essentialised DPT makes no clustering claims, it is quite resistant to functional dissociation studies.

The retreat from Standard Menu DPT to Essentialised DPT is cause for concern, though. It is reminiscent of what Lakatos (1970) calls a “degenerative” research program. It started off making many specific predictions, which were then falsified. In response, DPT has protected its core theoretical commitment (that cognition itself is split in two) by reducing the empirical content of its auxiliary hypotheses (rejecting conjunctive claims about Type-1 and Type-2 processing), resulting in fewer, less specific predictions (as in Essentialised DPT). (To be clear, I’m not saying that DPT is degenerative—just that it’s reasonable to worry it is.)

The apparent degeneracy of Mugg’s target, DPT, puts him at a significant advantage since he aims to defend a particular instance of single process theory (SPT), which he calls the “Soundboard Account of Reasoning”. Unfortunately, I worry that his account squanders some of its advantage: the first half of the book presses objections against DPT that seem to be mutually inconsistent, and then halfway through the book, Mugg shifts his focus from human reasoning to belief. The beleaguered DPT will live to see another day, I suspect.

Chapter 1 lays out some of Mugg’s starting commitments. He endorses Craver- and Khalidi-style weakenings of Boyd’s account of (natural) kinds (Boyd, 1991; Craver, 2009; Khalidi, 2023). He endorses the view that belief is a kind in this sense—i.e., it is individuated (at least partly) by a cluster of functional properties. It’s a bit difficult to tell from this stage why all this matters, but I think this chapter was meant to provide connective tissue between Chapters 2 and 3 (which focus on cognitive architecture) and Chapters 4–6 (which focus on belief).

Chapter 2 is more focused. Most of this chapter is dedicated to reviewing experimental literature on DPT. His review does well to highlight DPT’s (apparent) degeneracy—although the Lakatosian framing is mine, not his. Consistent with my Lakatosian framing of DPT’s problem, though, Mugg opts to present its problem in the form of a dilemma.

The first horn is Standard Menu DPT. It is rich in empirical content, so it makes bold, testable claims. The problem is just that these claims are widely disproven. Standard Menu DPT is interesting but empirically inadequate, then. The second horn is Essentialised DPT. It rejects the inadequate empirical content of Standard Menu DPT, but at the cost of explanatory power. Essentialised DPT is empirically adequate, but uninteresting (non-explanatory).

Even if we think Essentialised DPT is a retreat from Standard Menu DPT, it isn’t obvious that this is untoward in any way. Rejecting clustering claims is probably a good idea. Why not think, as Evans & Stanovich (2013) clearly seem to, that Essentialised DPT is a disciplining of Standard Menu DPT’s excesses? It’s not enough to say, as Mugg does, that this retreat “reduces the explanatory power of DPT” (31). Giving up some explanatory power for a lot more plausibility and empirical adequacy might be a perfectly acceptable bargain for DPT.

However, it seems like Mugg wants to say that Essentialised DPT has no explanatory power at all. At one point, for example, he complains how “it becomes unclear how to test these nuanced versions of DPT” (33). At another point, he even says that “the Logical Intuition Model renders itself compatible with the results at the cost of making the theory explanatorily and predictively inert” (35–36). If true, this is a serious problem. No amount of plausibility and empirical adequacy would be worth that much sacrifice in explanatory power.

Suppose Mugg is right. Essentialised DPT has no empirical content. All that’s left to it would be two bare claims about type-individuation: (a) some processes have a property-like access to working memory, while others don’t and (b) that property difference is sufficient for them to be different types of processes. Mugg appears to disagree with the latter claim, since he goes in for clustering accounts of types, such that we can change the type a cognitive process belongs to only by making sufficiently large changes to sufficiently many of its properties.

If that’s right, then Mugg’s disagreement with Essentialised DPT is metaphysical, not empirical. But then it’s strange that Mugg attempts to give an empirical argument against DPT at all (via inference to the best explanation). We can expect an empirical argument to work against Standard Menu DPT (since it has empirical content), but it shouldn’t work against Essentialised DPT. If Mugg does want to disagree with DPT on empirical grounds (as seems to), then he’d need to clearly attribute empirical content to DPT.

Chapter 3 introduces Mugg’s Soundboard Account of Reasoning. His account is centred on a metaphor: that reasoning has a soundboard architecture that is modulated with knobs and switches. Put literally, reasoning is such that it can be adequately characterised by dynamical functions over independent parameters, continuous (“knobs”) and/or discrete (“switches”). It’s the independence of the parameters that distinguishes this view from Standard Menu DPT (which maintains clustering and hence, dependence between the parameters).

As Mugg notes in Chapter 2, though, most dual-process theorists today endorse Essentialised DPT. By itself, though, the Soundboard Account is fully consistent with Essentialised DPT, which agrees that the parameters are independent. Essentialised DPT just insists on an additional commitment—that one parameter is privileged, such that changing the value of that parameter (but not any other parameter) makes a type-difference to the whole reasoning process (this works best if the parameter is discrete).

Mugg anticipates this worry:

One might think that I am…retaining diachronic Type 1 and Type 2 processing, which really makes the Soundboard Account a specific version of default-interventionist DPT. This is not so. Initial responses (as a category) and subsequent responses (as a category) will be too heterogeneous on the Soundboard Account to constitute kinds or serve as helpful constructs because the properties on the Standard Menu crosscut one another on my account, even for initial settings. (43)

However, this response begs the question against Essentialised DPT, which rejects that initial responses must have a different cluster of properties than subsequent responses in order to belong to different types of processes.

I think it would have been better if Mugg had just granted the metaphysical claim that default vs. intervening processing would be different types if the former were diachronically separated from the latter (perhaps by a discrete event like metacognitive control), and then he could insist that the empirical evidence suggests there is no discrete distinguishing event like this, and hence, there is no type-difference between default vs. intervening processing. For example, Grayot et al. (forthcoming) press an empirical objection like this against my proposed formulation of Essentialised DPT (Rumana, 2023).

Chapters 4–6 take a significant turn. Mugg stops critiquing DPT and starts critiquing various accounts of belief formation: Chapter 4 criticises Spinozan accounts of belief formation, Chapter 5 criticises conventional explanations of implicit bias results, and Chapter 6 criticises doxastic involuntarism.

In personal correspondence, Mugg has clarified that he meant to show that we “can indirectly test competing accounts of belief by tying them to cognitive architectures”. As such, his Soundboard Account is meant to provide a standard by which to arbitrate different accounts of belief. Since many accounts of belief draw support from outdated formulations of DPT (like Tamar Gendler’s account of aliefs), I think he’s quite right that support for these accounts can be undercut by reiterating criticisms against these forms of DPT. Therefore, I would encourage readers (especially readers interested in belief and belief formation) to read Chapters 4–6 in light of this.

Nevertheless, the book itself provides a different (and less plausible) justification for these chapters: “my aim in [Chapter 3] was not to argue for the Soundboard Account, but merely to build the theory. My argument for the theory will come as an inference to the best explanation [in Chapters 4–6]” (61). My review will take the book at its word that the Soundboard Account draws its justification from the content of Chapters 4–6 (especially since no other justification is provided for it, at least relative to Essentialised DPT).

I doubt the content of Chapters 4–6 can justify the Soundboard Account, though. First, Mugg himself admits that (at least some of) the accounts he criticises could also be couched in terms of his Soundboard Account (65). That seems true for all the accounts. Prima facie, that might seem like a problem for Mugg: if the Soundboard Account is equally compatible with all accounts of belief formation, how could confirming any account of belief formation confirm the Soundboard Account? One would start to worry that this is a real-life case of Clark Glymour’s (1980) tacking problem.

Mugg endorses Elliott Sober’s (1999) contrastive account of confirmation (fn. 4, p. 15), so perhaps he could forestall this objection by taking his explanations of belief formation to confirm his Soundboard Account against DPT, but not to confirm the Soundboard Account tout court. If that’s right, then confirming his account of belief formation can confirm the Soundboard Account against DPT only if his account of belief formation is better accommodated by the Soundboard Account than by DPT. Mugg suggests that it is, but he only considers the Standard Menu DPT (whose empirical inadequacy is already well-known).

Again, though, Essentialised DPT seems equally capable of accommodating accounts of belief formation, since empirical explanations of belief are neutral to type individuation. Thus, it’s hard to see how various explanations of belief formation can help arbitrate the disagreement between SPT and DPT. As a result, Chapters 4–6 may read like a long digression if you, like most dual-process theorists today, are more interested in the viability of Essentialised DPT (and if you take the content of these chapters to be justification for the Soundboard Account against DPT).

Chapter 7 tries to wrap everything up. It’s clear from this conclusion that the focus of the book has shifted from human reasoning to (various observations about) belief (and its formation). I’m left thinking that the title “From human reasoning to belief” should be taken as a literal description of the book’s focus. Perhaps, it’s unfair that my review hasn’t been more willing to follow this shift. I do expect this book can inspire a reader with interests in both belief and human reasoning to see the explanatory relationships between them. Either way, this book does not provide the sustained critique of DPT and empirical argument for the Soundboard Account that it promises.

REFERENCES

Boyd, R. (1991). Realism, anti-foundationalism and the enthusiasm for natural kinds. Philosophical Studies, 61(1), 127–148.

Craver, C. F. (2009). Mechanisms and natural kinds. Philosophical Psychology, 22(5), 575–594. https://doi.org/10.1080/09515080903238930

De Neys, W. (2021). On dual- and single-process models of thinking. Perspectives on Psychological Science, 16(6), 1412–1427. https://doi.org/10.1177/1745691620964172

Evans, J. St. B. T., & Stanovich, K. E. (2013). Dual-process theories of higher cognition: Advancing the debate. Perspectives on Psychological Science, 8(3), 223–241. https://doi.org/10.1177/1745691612460685

Glymour, Clark N. (1980). Theory and evidence. Princeton University Press.

Grayot, J. D., Beck, L., & Heijmeskamp, T. (forthcoming). Dual process theory and the challenges of functional individuation. Phenomenology and the Cognitive Sciences: 1–23.

Khalidi, M. A. (2023). Cognitive ontology: Taxonomic practices in the mind-brain sciences. Cambridge University Press.

Lakatos, I. (1970). Falsification and the methodology of scientific research programmes. In I. Lakatos & A. Musgrave (Eds.), Criticism and the growth of knowledge (pp. 91–196). Cambridge University Press.

Mugg, J. (2026). Cognitive architectures, kinds, and belief. In J. Lewis-Jong & E. Schwitzgebel (Eds.), The nature of belief (pp. 215–237). Oxford University Press. https://doi.org/10.1093/9780197744208.003.0010

Rumana, A. (2023). Metacognitive control in single- vs. Dual-process theory. Thinking & Reasoning, 29(2), 177–212. https://doi.org/10.1080/13546783.2022.2047106

Samuels, R. (2009). The magical number two, plus or minus: Dual-process theory as a theory of cognitive kinds. In J. S. B. Evans & K. Frankish (Eds.), In two minds: Dual processes and beyond (pp. 129–146). Oxford University Press.

Sober, E. (1999). Testability. Proceedings and Addresses of the American Philosophical Association, 73(2), 47–76.

Stephens, R. G., Dunn, J. C., & Hayes, B. K. (2018). Are there two processes in reasoning? The dimensionality of inductive and deductive inferences. Psychological Review, 125(2), 218–244. (2017-57168-001). https://doi.org/10.1037/rev0000088

Stephens, R. G., Dunn, J. C., & Hayes, B. K. (2019). Belief bias is response bias: Evidence from a two-step signal detection model. Journal of Experimental Psychology: Learning, Memory, and Cognition, 45(2), 320–332. (2018-35215-001). https://doi.org/10.1037/xlm0000587.

ACKNOWLEDGMENTS

I thank Josh Mugg for multiple rounds of patient comments on this review.

[1] This is consistent with Mugg’s previous work, where he says, e.g., “Since the properties of belief are drawn from a cognitive architecture that is testable, the metaphysics of belief I offer is indirectly empirically testable” (2026, 215).