This commentary addresses several major omissions in De Neys's “working model.” We predominantly focus on gaps in the conceptualization of the “switch feature” and stopping deliberative processes (S2).
Metacognitive research deals with the monitoring and control of thinking processes (Nelson & Narens, Reference Nelson, Narens and Bower1990). More than 30 years of research have dealt with the processes that inform subjective assessments of success (e.g., confidence) and the subsequent decisions (e.g., to rethink, see Fiedler, Ackerman, & Scarampi, Reference Fiedler, Ackerman, Scarampi, Sternberg and Funke2019). Of particular relevance is the meta-reasoning framework (Ackerman & Thompson, Reference Ackerman and Thompson2017), which is mentioned briefly in section 4.4. By using well-established metacognitive concepts, this framework opens the “black box” of mental effort regulation. It details monitoring and control processes that take place in the early intuitive reasoning stages (S1) separately from the deliberative stages (S2), including processes discussed in the target article and more.
First, the processes covered by the “switch feature” are discussed in length in the literature initiated by Thompson, Prowse Turner, and Pennycook (Reference Thompson, Prowse Turner and Pennycook2011) using the two-response paradigm with feeling of rightness judgment (FOR, mentioned in sect. 4.4; Ackerman & Thompson, Reference Ackerman and Thompson2017). FOR is the metacognitive judgment that accompanies the initial response that comes to mind. It has been considered to trigger the switch between S1 and S2 and found to predict S2 engagement (e.g., Thompson et al., Reference Thompson, Prowse Turner, Pennycook, Ball, Brack, Ophir and Ackerman2013).
A further issue is that the proposed model is incomplete in that the alleged “switch mechanism” is considered to depend entirely on the relative activation levels of competing intuitions and the mysterious “deliberation threshold.” In fact, a variety of situational and personal factors have been found to affect metacognitive control decisions, such as reasoning time and response choice. Specifically, task design, such as instructions to reason logically (e.g., Ferreira, Garcia-Marques, Sherman, & Sherman, Reference Ferreira, Garcia-Marques, Sherman and Sherman2006; Morsanyi, Primi, Chiesi, & Handley, Reference Morsanyi, Primi, Chiesi and Handley2009), cognitive load (De Neys, Reference De Neys2006; Morsanyi, Busdraghi, & Primi, Reference Morsanyi, Busdraghi and Primi2014), and time pressure (Sidi, Shpigelman, Zalmanov, & Ackerman, Reference Sidi, Shpigelman, Zalmanov and Ackerman2017), as well as individual characteristics, such as thinking dispositions (Cacioppo, Petty, Feinstein, & Jarvis, Reference Cacioppo, Petty, Feinstein and Jarvis1996), cognitive ability (e.g., Stanovich & West, Reference Stanovich and West2000), task-relevant knowledge (e.g., Chiesi, Primi, & Morsanyi, Reference Chiesi, Primi and Morsanyi2011; Stanovich & West, Reference Stanovich and West2008), and anxiety levels (e.g., Beilock & DeCaro, Reference Beilock and DeCaro2007; Primi, Donati, Chiesi, & Morsanyi, Reference Primi, Donati, Chiesi and Morsanyi2018) affect reasoning time and response choice. Thus, any model explaining the “switch feature” should incorporate and account for the contextual and individual factors that influence the reasoning process.
Second, the target article discusses stopping deliberative processes (S2) and reverting to S1. An overlooked issue, though, is when to stop S2 and provide a response. Within the metacognitive literature, several models address stopping effortful thinking: The discrepancy reduction models (Nelson & Narens, Reference Nelson, Narens and Bower1990), the region of proximal learning (Metcalfe & Kornell, Reference Metcalfe and Kornell2005), and the diminishing criterion model (DCM, Ackerman, Reference Ackerman2014; see Ackerman, Yom-Tov, & Torgovitsky, Reference Ackerman, Yom-Tov and Torgovitsky2020, for a review). According to the most recent model, the DCM, stopping thinking efforts is guided by a combination of two stopping criteria: (a) Confidence in each considered answer is compared to a desired confidence level. Importantly, this stopping criterion dynamically drops as people deliberate longer, reflecting compromising on expected success. (b) A time limit for thinking about each task item, beyond which people are reluctant to think any further (see also Hawkins & Heathcote, Reference Hawkins and Heathcote2021).
Third, based on the suggested model, “System 2 deliberation will extend for as long as the uncertainty remains above the threshold” (target article, sect. 3.4, para. 3). Thus, under substantial uncertainty people are doomed to think forever. Nevertheless, a totally overlooked aspect is when people opt out (e.g., “I don't know”) or turn to external help (see Ackerman, Reference Ackerman2014; Undorf, Livneh, & Ackerman, Reference Undorf, Livneh and Ackerman2021). In particular, considering children and novices brings to the fore that people looking at unfamiliar problems may not have any available heuristics to activate. Developmentally, there is a blurry line between deliberative and intuitive processes (Osman & Stavy, Reference Osman and Stavy2006) in that responses that can be given quasi automatically by adults may require cognitive effort for children (Morsanyi & Handley, Reference Morsanyi and Handley2008) and may become established by learning (Fischbein, Reference Fischbein1987; Gauvrit & Morsanyi, Reference Gauvrit and Morsanyi2014). De Neys briefly considers lack of S1 response (sect. 2.1.5). Another possibility is that people may activate a series of distantly related heuristics, but none of these would be sufficiently strong to offer an answer. In contrast, according to the DCM, when people get to a pre-set time limit, they may prefer opting out over providing a low confidence response. This topic was discussed in metacognitive research already in the 1990s (Koriat & Goldsmith, Reference Koriat and Goldsmith1996) and was further developed since then (see Undorf et al., Reference Undorf, Livneh and Ackerman2021). Thus, there are processes that prevent people from thinking forever.
Fourth, De Neys asks in the Introduction “how do we know that we can rely on an intuitively cued problem solution” (target article, para. 4 in the Introduction) and mentions that “the internal switch decision is itself intuitive in nature” (target article, para. 4 in the Introduction). In metacognitive terms, these intuitions are based on heuristic cues that underlie all metacognitive judgments (Koriat, Reference Koriat1997). Metacognitive judgments combine an extensive amount of features (Undorf & Bröder, Reference Undorf and Bröder2021), including individual self-perceptions and beliefs (“beyond my expertise”), task characteristics (time pressure), and item characteristics (conclusion believability) that may influence, and sometimes mislead, metacognitive judgments (see Ackerman, Reference Ackerman2019). Given the wide-spread biases in judgments like FOR and confidence (Thompson et al., Reference Thompson, Prowse Turner, Pennycook, Ball, Brack, Ophir and Ackerman2013), considering potential misleading factors must be incorporated in any model of switch and stopping mechanisms.
Finally, from a developmental perspective, adults have a larger repertoire of heuristics and better ability to integrate them into their cognitive and metacognitive processes than children (Koriat, Ackerman, Adiv, Lockl, & Schneider, Reference Koriat, Ackerman, Adiv, Lockl and Schneider2014). However, in the proposed model, the more heuristics are considered, the longer the thinking process that deals with potential conflicts among them. This contrasts with the traditional role assigned to reasoning heuristics – that they offer immediately available (and highly compelling) responses immediately (e.g., Evans, Reference Evans2006), which is why they are considered to be adaptive and essential parts of the cognitive architecture.
In sum, the proposed model ignores well-established bodies of literature that address the central issues it was meant to cover. Particularly, metacognitive research offers switch and stopping rules, heuristic processes, individual characteristics, and developmental trajectories required for describing the complex processes underlying reasoning.
This commentary addresses several major omissions in De Neys's “working model.” We predominantly focus on gaps in the conceptualization of the “switch feature” and stopping deliberative processes (S2).
Metacognitive research deals with the monitoring and control of thinking processes (Nelson & Narens, Reference Nelson, Narens and Bower1990). More than 30 years of research have dealt with the processes that inform subjective assessments of success (e.g., confidence) and the subsequent decisions (e.g., to rethink, see Fiedler, Ackerman, & Scarampi, Reference Fiedler, Ackerman, Scarampi, Sternberg and Funke2019). Of particular relevance is the meta-reasoning framework (Ackerman & Thompson, Reference Ackerman and Thompson2017), which is mentioned briefly in section 4.4. By using well-established metacognitive concepts, this framework opens the “black box” of mental effort regulation. It details monitoring and control processes that take place in the early intuitive reasoning stages (S1) separately from the deliberative stages (S2), including processes discussed in the target article and more.
First, the processes covered by the “switch feature” are discussed in length in the literature initiated by Thompson, Prowse Turner, and Pennycook (Reference Thompson, Prowse Turner and Pennycook2011) using the two-response paradigm with feeling of rightness judgment (FOR, mentioned in sect. 4.4; Ackerman & Thompson, Reference Ackerman and Thompson2017). FOR is the metacognitive judgment that accompanies the initial response that comes to mind. It has been considered to trigger the switch between S1 and S2 and found to predict S2 engagement (e.g., Thompson et al., Reference Thompson, Prowse Turner, Pennycook, Ball, Brack, Ophir and Ackerman2013).
A further issue is that the proposed model is incomplete in that the alleged “switch mechanism” is considered to depend entirely on the relative activation levels of competing intuitions and the mysterious “deliberation threshold.” In fact, a variety of situational and personal factors have been found to affect metacognitive control decisions, such as reasoning time and response choice. Specifically, task design, such as instructions to reason logically (e.g., Ferreira, Garcia-Marques, Sherman, & Sherman, Reference Ferreira, Garcia-Marques, Sherman and Sherman2006; Morsanyi, Primi, Chiesi, & Handley, Reference Morsanyi, Primi, Chiesi and Handley2009), cognitive load (De Neys, Reference De Neys2006; Morsanyi, Busdraghi, & Primi, Reference Morsanyi, Busdraghi and Primi2014), and time pressure (Sidi, Shpigelman, Zalmanov, & Ackerman, Reference Sidi, Shpigelman, Zalmanov and Ackerman2017), as well as individual characteristics, such as thinking dispositions (Cacioppo, Petty, Feinstein, & Jarvis, Reference Cacioppo, Petty, Feinstein and Jarvis1996), cognitive ability (e.g., Stanovich & West, Reference Stanovich and West2000), task-relevant knowledge (e.g., Chiesi, Primi, & Morsanyi, Reference Chiesi, Primi and Morsanyi2011; Stanovich & West, Reference Stanovich and West2008), and anxiety levels (e.g., Beilock & DeCaro, Reference Beilock and DeCaro2007; Primi, Donati, Chiesi, & Morsanyi, Reference Primi, Donati, Chiesi and Morsanyi2018) affect reasoning time and response choice. Thus, any model explaining the “switch feature” should incorporate and account for the contextual and individual factors that influence the reasoning process.
Second, the target article discusses stopping deliberative processes (S2) and reverting to S1. An overlooked issue, though, is when to stop S2 and provide a response. Within the metacognitive literature, several models address stopping effortful thinking: The discrepancy reduction models (Nelson & Narens, Reference Nelson, Narens and Bower1990), the region of proximal learning (Metcalfe & Kornell, Reference Metcalfe and Kornell2005), and the diminishing criterion model (DCM, Ackerman, Reference Ackerman2014; see Ackerman, Yom-Tov, & Torgovitsky, Reference Ackerman, Yom-Tov and Torgovitsky2020, for a review). According to the most recent model, the DCM, stopping thinking efforts is guided by a combination of two stopping criteria: (a) Confidence in each considered answer is compared to a desired confidence level. Importantly, this stopping criterion dynamically drops as people deliberate longer, reflecting compromising on expected success. (b) A time limit for thinking about each task item, beyond which people are reluctant to think any further (see also Hawkins & Heathcote, Reference Hawkins and Heathcote2021).
Third, based on the suggested model, “System 2 deliberation will extend for as long as the uncertainty remains above the threshold” (target article, sect. 3.4, para. 3). Thus, under substantial uncertainty people are doomed to think forever. Nevertheless, a totally overlooked aspect is when people opt out (e.g., “I don't know”) or turn to external help (see Ackerman, Reference Ackerman2014; Undorf, Livneh, & Ackerman, Reference Undorf, Livneh and Ackerman2021). In particular, considering children and novices brings to the fore that people looking at unfamiliar problems may not have any available heuristics to activate. Developmentally, there is a blurry line between deliberative and intuitive processes (Osman & Stavy, Reference Osman and Stavy2006) in that responses that can be given quasi automatically by adults may require cognitive effort for children (Morsanyi & Handley, Reference Morsanyi and Handley2008) and may become established by learning (Fischbein, Reference Fischbein1987; Gauvrit & Morsanyi, Reference Gauvrit and Morsanyi2014). De Neys briefly considers lack of S1 response (sect. 2.1.5). Another possibility is that people may activate a series of distantly related heuristics, but none of these would be sufficiently strong to offer an answer. In contrast, according to the DCM, when people get to a pre-set time limit, they may prefer opting out over providing a low confidence response. This topic was discussed in metacognitive research already in the 1990s (Koriat & Goldsmith, Reference Koriat and Goldsmith1996) and was further developed since then (see Undorf et al., Reference Undorf, Livneh and Ackerman2021). Thus, there are processes that prevent people from thinking forever.
Fourth, De Neys asks in the Introduction “how do we know that we can rely on an intuitively cued problem solution” (target article, para. 4 in the Introduction) and mentions that “the internal switch decision is itself intuitive in nature” (target article, para. 4 in the Introduction). In metacognitive terms, these intuitions are based on heuristic cues that underlie all metacognitive judgments (Koriat, Reference Koriat1997). Metacognitive judgments combine an extensive amount of features (Undorf & Bröder, Reference Undorf and Bröder2021), including individual self-perceptions and beliefs (“beyond my expertise”), task characteristics (time pressure), and item characteristics (conclusion believability) that may influence, and sometimes mislead, metacognitive judgments (see Ackerman, Reference Ackerman2019). Given the wide-spread biases in judgments like FOR and confidence (Thompson et al., Reference Thompson, Prowse Turner, Pennycook, Ball, Brack, Ophir and Ackerman2013), considering potential misleading factors must be incorporated in any model of switch and stopping mechanisms.
Finally, from a developmental perspective, adults have a larger repertoire of heuristics and better ability to integrate them into their cognitive and metacognitive processes than children (Koriat, Ackerman, Adiv, Lockl, & Schneider, Reference Koriat, Ackerman, Adiv, Lockl and Schneider2014). However, in the proposed model, the more heuristics are considered, the longer the thinking process that deals with potential conflicts among them. This contrasts with the traditional role assigned to reasoning heuristics – that they offer immediately available (and highly compelling) responses immediately (e.g., Evans, Reference Evans2006), which is why they are considered to be adaptive and essential parts of the cognitive architecture.
In sum, the proposed model ignores well-established bodies of literature that address the central issues it was meant to cover. Particularly, metacognitive research offers switch and stopping rules, heuristic processes, individual characteristics, and developmental trajectories required for describing the complex processes underlying reasoning.
Financial support
This work was funded by the Israel Science Foundation (grant reference: 234/18) to R.A. and by the Economic and Social Research Council, UK (grant reference: ES/W002914/1) to K.M.
Competing interest
None.