Not feeling right about uncertainty monitoring

Abstract

De Neys proposed a “switch” model to address what he argued to be lacuna in dual-process theory, in which he theorized about the processes that initiate and terminate analytic thinking. We will argue that the author neglected to acknowledge the abundant literature on metacognitive functions, specifically, the meta-reasoning framework developed by Ackerman and Thompson (2017), that addresses just those questions.

The meta-reasoning framework (Ackerman & Thompson, 2017) differentiates between object-level processes that are responsible for thinking, reasoning, and deciding, and the meta-level process that monitors and controls them. Part of this theory addresses the processes that initiate and terminate analytic thinking, the so-called switch function proposed by De Neys. In the meta-reasoning model, the feeling of rightness (Thompson, 2009; Thompson, Prowse Turner, & Pennycook, 2011, 2013) has been proposed as one way that intuitive responses can be monitored and analytic processes initiated.

Thompson argued that system 1 processing is accompanied by a metacognitive experience, a feeling of rightness or certainty that the response(s) generated by system 1 are correct (Thompson, 2009). The relative strength of one's feeling of rightness determines the probability one will switch to and engage system 2; when the feeling is strong, the response is often accepted with little system 2 analysis. Conversely, when the feeling of rightness is weak, system 2 is often engaged. In other words, the feeling of rightness is a cue to either accept the outcome of system 1 processing or “switch” to system 2 processing. Thompson also proposed potential determinants of the strength of a feeling of rightness, such as fluency of processing (speed and ease of response generation) and, importantly, the presence of conflicting responses (Thompson & Johnson, 2014; Thompson et al., 2011).

Although De Neys briefly acknowledged that his working model could be integrated with the feeling of rightness (sect. 4.4), he neglected a broad and well-established model to propose an explanation for a narrower range of phenomena. The meta-reasoning framework (Ackerman & Thompson, 2017) is a multifaceted framework that encompasses a variety of monitoring and control processes, in addition to a “switch” mechanism. Thus, De Neys's proposal fleshed out the specific case of monitoring for conflict in the broader meta-reasoning framework. We also note the similarity between monitoring feelings of rightness and “uncertainty monitoring,” where one can think about “uncertainty” as the inverse of confidence or feeling of rightness.

De Neys also argued that leading switch accounts presuppose exclusivity, where certain types of responses (e.g., a normatively correct response) are exclusively generated by one system or the other. Thompson and Newman (2020) noted that the exclusivity assumption has been long abandoned in dual-process theorizing (Evans, 2019; Evans & Stanovich, 2013; Stanovich, 2018; Thompson, 2011) and have obtained evidence in support of this in our own lab (Newman, Gibb, & Thompson, 2017). Most importantly, the feeling of rightness does not depend on the normative qualities of the response generated and, therefore, does not rely on an exclusivity assumption to be instantiated. Essentially, the claim that leading switch accounts presuppose exclusivity is not consistent with current metacognitive models.

We agree with De Neys that many of the processes that monitor our cognitions are likely “system 1” in character: They should not demand working memory resources and their origins are likely not subject to introspection (Koriat, 2007). However, we think there are several important differences between the uncertainty monitoring mechanisms proposed by De Neys and the types of monitoring mechanisms that are common in the metacognitive literature. First, although their origins may not be subject to introspection, their outputs are thought to enter awareness. We have hypothesized that feelings of rightness are felt subjectively (Thompson et al., 2011), meaning one is consciously aware, at least to some degree, that they have these feelings. Alternatively, it is not clear that uncertainty monitoring has the same properties.

This difference is subtle but crucial. From a metacognitive perspective, one can be aware of when they feel their response is completely right, completely wrong, or somewhere in between. From an uncertainty monitoring perspective, it seems that one is only aware when one feels a sufficient degree of uncertainty – either from a weakly generated response or two (or more) conflicting responses relatively similar in strength. Therefore, there is no mechanism by which an individual could be aware of when they feel extremely right or wrong, nor for how such feelings could cue the engagement of system 2. Finally, we note that although most monitoring is likely to be implicit, there is also a role for explicit beliefs about oneself, the task, the types of behaviours that constitute good reasoning, and so on (Ackerman, 2019; Jost, Kruglanski, & Nelson, 1998; Koriat, 2007). Thus, not all monitoring processes can be lumped under the “system 1” rubric.

Of most relevance, De Neys's proposed model cannot explain the phenomenon of being highly certain that something is wrong (i.e., feeling of wrongness), such as when a firefighter has a sudden but strong feeling that something is amiss and they must get out of the building immediately (Klein, Calderwood, & Clinton-Cirocco, 2010). Similarly, I could ask you: Is Edinburgh the capital of Botswana? You may not know the actual capital of Botswana (it is Gaborone), but you likely would have a remarkably strong feeling that it is definitely not Edinburgh. Similarly, one may have a strong feeling of error about a response one has given (Ackerman & Thompson, 2017; Fernandez Cruz, Arango-Muñoz, & Volz, 2016). Both dimensions of the spectrum are important because a strong feeling about a response (whether it be rightness or wrongness or error; Ackerman & Thompson, 2017) is a cue that action is necessary (Thompson, 2017).

De Neys provided a detailed analysis of how reasoners monitor for conflict. However, conflict monitoring is just one part of the role that metacognitive monitoring plays. Moreover, the proposal failed to acknowledge the more comprehensive framework developed by Ackerman and Thompson. It is ultimately an oversimplification of an existing framework that substitutes uncertainty for feeling of rightness, uncertainty monitoring for metacognitive monitoring, and an uncertainty criterion for the diminishing criterion model of confidence (Ackerman, 2014). We agree with much of the criticism and questions raised by De Neys here (both directly and indirectly) and acknowledge that they are important, but note that frameworks, such as the meta-reasoning framework of Ackerman and Thompson (2017), already exist to answer them.