Hoist by its own petard: The ironic and fatal flaws of dual-process theory

Abstract

By stipulating the existence of a system 1 and a system 2, dual-process theories raise questions about how these systems function. De Neys identifies several such questions for which no plausible answers have ever been offered. What makes the nature of systems 1 and 2 so difficult to ascertain? The answer is simple: The systems do not exist.

Dual-process theories of human reasoning have yet to provide plausible answers to basic questions about the nature of system 1 and system 2 processing. Can system 1 reason logically? How do people switch from system 1 to system 2? As De Neys convincingly argues, existing answers to these questions fail under logical and empirical scrutiny.

There is an irony to this. The questions that confound dual-process theories are the very questions that these theories introduce in the first place. By positing that systems 1 and 2 exist in some meaningful sense, dual-process theories saddle themselves with the challenge of explaining how these systems (or types of processes) operate and interact. Having long failed to meet this challenge with even a single coherent hypothesis of something as basic as how system 2 is activated, it is worth asking if dual-process theories are wrong at the most fundamental level: Maybe system 1 and system 2 simply do not exist. Maybe what De Neys presents as important puzzles in need of solving are just red herrings symptomatic of a flawed theoretical foundation. We suspect that this is the case, and therefore recommend that instead of developing dual-process theories further, researchers abandon dual-process theories altogether (Melnikoff & Bargh, 2018).

What would it look like to abandon dual-process theories? Instead of asking questions about system 1 and system 2, researchers would ask questions like: How much time and effort is required to perform well on a given type of reasoning problem? Under what conditions do different components of reasoning occur spontaneously versus intentionally? How do different types of working memory load (e.g., visual vs. verbal) interact with different aspects of the reasoning process? By eschewing commitment to the existence of systems 1 and 2, questions like these avoid the obstacles that thwart dual-process theories.

Consider the central question of the target article: How do people switch between effortless (“intuitive”) and effortful (“deliberative”) reasoning? As De Neys shows, the answers offered by dual-process theories fail to meet even the minimum threshold of logical coherence. For instance, multiple dual-process theories claim that the activation of system 2 depends on system 2 already being activated. Such paradoxes vanish, however, when the concepts of system 1 and system 2 are abandoned. To illustrate, consider the learned value of control (LVOC) model of Lieder, Shenhav, Musslick, and Griffiths (2018). This model says that people use reinforcement learning to estimate the value of exerting different amounts of effort in particular situations. For example, people may learn from prior experience that when presented with certain reasoning problems in particular contexts, they tend to obtain better outcomes when they ignore their initial hunch and invest effort in further deliberation. This learned value can be used to decide how much effort to invest when similar situations are encountered in the future.

There are no paradoxes here. The LVOC model is a perfectly coherent account of how people modulate the effort they invest in reasoning – one of many (e.g., Abrahamse, Braem, Notebaert, & Verguts, 2016; Restle, 1962; Shenhav, Botvinick, & Cohen, 2013) – and the fact that it makes no reference to systems 1 and 2 is no coincidence. Just imagine if it did. Reframed in dual-process terms, the LVOC might posit that system 2 learns the expected value of deliberating, and then uses this information to decide when to ignore system 1 processing in favor of system 2 processing. Now we have a problem: The new theory implies that the activation of system 2 is a precondition for itself, introducing an infinite regress. An alternative reframing of the LVOC might posit that whenever system 1 processing is ignored in favor of system 2 processing, system 1 updates the value of deliberation and uses this updated value to decide when to deliberate in the future. But this is paradoxical. If system 1 processing is ignored, system 1 cannot, by definition, update the value of anything.

The point here is that by embracing the system 1/system 2 framework, nothing is gained but confusion. The conceptual commitments of the framework only make it harder to generate coherent answers to the very questions that dual-process theorists care about.

Of course, none of this would matter if there were evidence showing that systems 1 and 2 do in fact exist. Although fictional theoretical constructs can be discarded as soon as they prove unhelpful, actual features of the mind cannot be ignored simply because they introduce conceptual conundrums. So, are systems 1 and 2 real?

There is no reason to think so. Fundamentally, system 1/system 2 is a proposed dimension along which psychological processes vary: At one end of the spectrum, processes are fast, effortless, and spontaneous, and at the other end, processes are slow, effortful, and intentional. So, the reality of the system 1/system 2 distinction hinges on whether it is true that the features of speed, effort, and intentionality are in some sense reducible to a single, more basic dimension.

Evidence supporting this idea does not exist. There has been no attempt to establish that speed, effort, and intentionality are intercorrelated to any meaningful degree, let alone to the point that they might be reducible to a single underlying dimension (Bargh, 1994; Keren & Schul, 2009; Kruglanski & Gigerenzer, 2011; Melnikoff & Bargh, 2018; Moors & De Houwer, 2006). On the contrary, countless examples of “misalignments” between processing features have been documented, such as fast processes that are intentional (Melnikoff & Bailey, 2018) and inefficient (Gilbert & Hixon, 1991), and spontaneous processes that are effortful (Kim, Kim, & Chun, 2005). Such phenomena suggest that an attempt to validate the system 1/system 2 distinction, were it ever undertaken, is unlikely to succeed.

In short, there is no evidence that systems 1 and 2 exist in any meaningful sense, and head-scratching paradoxes vanish if we assume they do not. Therefore, we should abandon dual-process theory rather than embark on a doomed mission to save it.