Unifying theories of reasoning and decision making

Abstract

De Neys offers a welcome departure from the dual-process accounts that have dominated theorizing about reasoning. However, we see little justification for retaining the distinction between intuition and deliberation. Instead, reasoning can be treated as a case of multiple-cue decision making. Reasoning phenomena can then be explained by decision-making models that supply the processing details missing from De Neys's framework.

This provocative target article questions several key assumptions of popular dual-process models of reasoning and outlines a novel cognitive architecture for explaining the relationship between intuition and deliberation. A central argument is that a reasoner can have multiple, competing “intuitions” about the correct solution to a problem, with the activation strength of each intuition varying over time. We see this as a potentially valuable step in theory building in the field of human reasoning, that brings the field closer to other productive areas of research in human judgment and decision making.

We question, however, whether there is any need to retain the distinction between intuitive and deliberative processing. If we allow for multiple intuitions, some of which align with normative principles of probability or logic, there seems little need for an additional deliberative system. This is evident in the target article, where De Neys struggles to define a unique role for “deliberation.” One suggestion is that deliberation involves the application of an algorithm or execution of a set of rules when solving a problem. But given that such rules can become automated with experience (Logan & Klapp, 1991), this seems like a weak definition. Another suggestion by De Neys removes deliberation from the decision-making process altogether – relegating it the role of rationalizing or justifying decisions that have already been made.

As an alternative approach, we suggest that the notion of multiple “intuitions” should be re-framed in more general terms as attention to multiple cues that define alternative decision options. In this approach, reasoning in tasks like ratio bias, moral judgment, or verbal syllogisms, can be captured by the same general cognitive architecture used to explain other decisions involving multiple cues or features. To illustrate the basic idea, consider planning to purchase a new car. This is likely to involve consideration of multiple cues (e.g., electric vs. petrol power source, price, manufacturer's reputation). As in the most interesting reasoning problems, these cues will often be in conflict (e.g., electric cars are more environmentally sustainable but are often more expensive). A theoretical model of decision making in such cases needs to explain how the various cues are weighted when comparing options and how trade-offs between cues take place. In this framework, decision-making cues may vary in complexity, salience, and familiarity. However, there is no need to assume discrete types of processing (e.g., intuition vs. deliberation, system 1 vs. system 2) for dealing with different cues.

A key implication is that models of multiple-cue decision making (e.g., Busemeyer & Townsend, 1993) can be applied to understand reasoning phenomena. We believe that this has many advantages. For one thing, the processing assumptions of these decision-making models have been laid out in far more detail, and been subject to more extensive empirical testing, than the architecture sketched by De Neys. For example, following the structure of popular “evidence accumulation” models of decision making (Busemeyer & Townsend, 1993; Ratcliff, Smith, Brown, & McKoon, 2016), reasoning could be thought of as a process of the dynamic accumulation of evidence relevant to each decision option (e.g., options based on absolute number vs. ratio in ratio bias problems; utilitarian vs. deontological responses in moral judgments; judging whether a verbal argument is valid or invalid). A decision is made when the evidence for a given option reaches a threshold. Unlike De Neys's approach, such models provide a principled account of how and why the “activation strength” associated with each cue changes over time (cf. Ratcliff et al., 2016). They also explain how the accumulation of evidence for each cue interacts with other components of the decision-making process such as how one sets a decision threshold and how one encodes the relevant cues.

Together these model components have the prospect of explaining many key reasoning phenomena. For example, the fact that arguments with believable conclusions are more likely to be judged as valid regardless of logical structure (Dube, Rotello, & Heit, 2010), may be explained by assuming that believable arguments have a higher “start-point” for evidence accumulation than unbelievable arguments. Hence, they require less evidence to reach threshold for a “valid” response. Higher rates of endorsement of arguments based on their believability rather than validity under time pressure (Evans & Curtis-Holmes, 2005; Hayes, Stephens, Ngo, & Dunn, 2018) can be explained by adjustment of the relevant decision thresholds. Evidence accumulation models are also well-equipped to explain the inconsistency we often see in individual reasoning patterns, such as shifts between utilitarian and deontological options across different moral judgments (e.g., Cushman, Young, & Hauser, 2006) or shifts between a focus on the visual appearance of text as opposed to logical structure or argument plausibility in verbal reasoning (Hayes et al., 2022). Such shifts can be explained as context-driven changes in the rate of evidence accumulation for rival decision options.

To date, evidence accumulation models have most often been applied to simple perceptual decisions. However, there is good evidence that they “scale-up” to capturing the processes involved in complex decisions that more closely resemble those involved in reasoning tasks (e.g., Hawkins, Hayes, & Heit, 2016; Krajbich, Bartling, Hare, & Fehr, 2015; Palada et al., 2016).

In sum, the approach suggested by De Neys is a welcome departure from the dual-process accounts that have dominated recent theorizing about human reasoning. Retaining a hard distinction between intuitive and deliberative processes (regardless of whether this distinction is viewed as “qualitative” or “quantitative”), however, does little to advance our understanding. Instead, we suggest that reasoning in classic conflict tasks be treated as a special case of multiple-cue decision making. Doing so will allow us to apply powerful theoretical models that supply much of the processing detail missing from the architecture proposed by De Neys.