The target article makes an important theoretical contribution. Comparing knowledge and belief representation provides a compelling account that will shape future research significantly. Phillips et al. rightly note that research on the automaticity of adult belief and knowledge representation is contentious. We remain sceptical that knowledge versus belief representation will ever neatly classify as automatic or not. We consider four aspects of Phillips et al.'s reasoning to illustrate this point.
1. Visual perspectives are not pure analogies to knowledge and belief representations
Phillips et al. invoke the distinction between level-1 and level-2 perspectives taking to distinguish knowledge versus belief representation. The strongest evidence supporting adults' automatic knowledge representation comes from a level-1 perspective taking task (Samson, Apperly, Braithwaite, Andrews, & Bodley Scott, Reference Samson, Apperly, Braithwaite, Andrews and Bodley Scott2010), whereas level-2 tasks are used to support the non-automaticity of belief reasoning (Surtees, Butterfill, & Apperly, Reference Surtees, Butterfill and Apperly2012; Surtees, Apperly, & Samson, Reference Surtees, Apperly and Samson2016a). Both sides of the analogy between visual perspectives and knowledge versus belief are problematic, however. It is not necessarily the case that someone's current level-1 perspective is consistent with their knowledge state. Although seeing is knowing is a reasonable heuristic (Moll & Tomasello, Reference Moll and Tomasello2007), not seeing is not knowing is unlikely to be. We rarely beep our horn as our neighbours reverse towards their houses without looking, because we know they know it is there. Regarding level-2 perspective taking, representing how someone sees something does not necessarily equate to representing their belief about the object. You can look at a number 6 from one angle, although we see it as a number 9 from another, without us holding differing beliefs about the object. By equating knowledge and belief with level-1 and level-2 perspective taking, respectively, Phillips et al. may be over-simplifying knowledge and over-complicating belief.
2. Mental-state sensitivity may not necessitate mental-state representation
Phillips et al. conclude evidence of automaticity based on studies documenting interference from another's perspective on judgements of one's own perspective (Kovács, Téglás, & Endress, Reference Kovács, Téglás and Endress2010; Samson et al., Reference Samson, Apperly, Braithwaite, Andrews and Bodley Scott2010). We question whether the incidental sensitivity to others' mental states revealed by these altercentric-interference effects necessitates their being represented. Several accounts leave open the possibility that other mechanisms underlie such mental-state sensitivity. As Phillips et al. note, submentalizing accounts propose lower-level, domain-general explanations of altercentric interference (Heyes, Reference Heyes2014). Two-systems accounts, in contrast, posit that altercentric interference reflects domain-specific registration of “belief-like states” (Apperly & Butterfill, Reference Apperly and Butterfill2009). Taking cues from both of these accounts, process-dissociation accounts hold that altercentric interference is not a process-pure index of mental-state registration; rather, such effects can be decomposed into at least two component processes: calculation of the agent's perspective and detection of one's own perspective (Todd, Cameron, & Simpson, Reference Todd, Cameron and Simpson2017, Reference Todd, Cameron and Simpsonin press; Todd, Simpson, & Cameron, Reference Todd, Simpson and Cameron2019), with the latter process likely reflecting something more domain-general (Payne, Reference Payne2005). Phillips et al. reason that knowledge representation is more basic based on evidence from tasks finding altercentric interference, but this could be evidence of arbitrary correlation between “knowledge” and stimulus features, coupling to more basic mental-state-like states, or poorer discrimination from self-knowledge.
3. Automaticity is not categorical
Phillips et al. aim to categorize knowledge and belief representation as automatic or not. Such categorization is likely an over-simplification of how cognitive systems operate. Automaticity is not a single feature, but rather a set of conceptually separable features that often do not co-occur (Melnikoff & Bargh, Reference Melnikoff and Bargh2018; Moors & De Houwer, Reference Moors and De Houwer2006). Thus, specifying in what way(s) belief and knowledge representation are automatic is crucial. The automaticity features receiving most empirical attention are goal-independence and efficiency. We agree with Phillips et al. that current evidence supports level-2 altercentric interference and process-dissociation estimates of agent-perspective calculation as consistently not-automatic. On a level-2 task, when participants only ever considered their own perspective, we showed that altercentric interference (Surtees et al., Reference Surtees, Apperly and Samson2016a) and agent-perspective calculation (Todd et al., Reference Todd, Cameron and Simpsonin press) were absent, suggesting level-2 perspective taking is goal-dependent. Using the same task, Todd et al. (Reference Todd, Simpson and Cameron2019) found that time pressure also impaired agent-perspective calculation, suggesting it is relatively inefficient. Evidence for level-1 automaticity is more equivocal. Some studies suggest level-1 altercentric interference emerges regardless of participant task goals (Conway, Lee, Ojaghi, Catmur, & Bird, Reference Conway, Lee, Ojaghi, Catmur and Bird2017; Surtees et al., Reference Surtees, Apperly and Samson2016a); others do not (Ferguson, Apperly, & Cane, Reference Ferguson, Apperly and Cane2017; Todd et al., Reference Todd, Cameron and Simpsonin press). Some studies suggest level-1 altercentric interference and agent-perspective calculation are efficient, in that they were unimpaired by time pressure (Todd et al., Reference Todd, Cameron and Simpson2017, Reference Todd, Simpson and Cameron2019) or a concurrent resource-consuming task (Qureshi, Apperly, & Samson, Reference Qureshi, Apperly and Samson2010); another found the opposite (Qureshi & Monk, Reference Qureshi and Monk2018). Although Phillips et al. acknowledge empirical uncertainty, our view is that different mental-state representations are not necessarily categorized fully as automatic or non-automatic. The impact of context on automaticity further supports this contention.
4. Mental-state representation may differ in observational and interactive contexts
Processes for interaction are different when directly engaging with another person, as opposed to passively observing them (Schilbach et al., Reference Schilbach, Timmermans, Reddy, Costall, Bente, Schlicht and Vogeley2013). Phillips et al. largely focus on observational, “third-person” approaches to adults' mental-state representation. It is not self-evident that the more “basic” form of mental-state reasoning in interactive and observational scenarios will be equivalent. Here, level-2 perspective taking provides an example of where they are not. Above we highlighted that level-2 perspective taking in an observational setting seems to be goal-dependent (Surtees et al., Reference Surtees, Apperly and Samson2016a; Todd et al., Reference Todd, Cameron and Simpsonin press). In a closely matched interactive context, however, participants do suffer interference from a partner's perspective, even if they are never asked to report it, suggesting it is goal-independent (Elekes, Varga, & Király, Reference Elekes, Varga and Király2016; Surtees, Samson, & Apperly, Reference Surtees, Samson and Apperly2016b). One possibility is that representations of different mental states are differentially cued by different aspects of social interaction. In this case, perhaps low-level stimulus features cue level-1 perspective taking, whereas “real” interaction or a specific goal may be required to cue level-2 perspective taking. One way to interpret this is that level-1 perspective taking is more basic, but another is to see level-2 perspective taking as linked to social interaction in a more fundamental way. Classifying knowledge or belief representation as more basic may obfuscate subtle variation with context.
The target article makes an important theoretical contribution. Comparing knowledge and belief representation provides a compelling account that will shape future research significantly. Phillips et al. rightly note that research on the automaticity of adult belief and knowledge representation is contentious. We remain sceptical that knowledge versus belief representation will ever neatly classify as automatic or not. We consider four aspects of Phillips et al.'s reasoning to illustrate this point.
1. Visual perspectives are not pure analogies to knowledge and belief representations
Phillips et al. invoke the distinction between level-1 and level-2 perspectives taking to distinguish knowledge versus belief representation. The strongest evidence supporting adults' automatic knowledge representation comes from a level-1 perspective taking task (Samson, Apperly, Braithwaite, Andrews, & Bodley Scott, Reference Samson, Apperly, Braithwaite, Andrews and Bodley Scott2010), whereas level-2 tasks are used to support the non-automaticity of belief reasoning (Surtees, Butterfill, & Apperly, Reference Surtees, Butterfill and Apperly2012; Surtees, Apperly, & Samson, Reference Surtees, Apperly and Samson2016a). Both sides of the analogy between visual perspectives and knowledge versus belief are problematic, however. It is not necessarily the case that someone's current level-1 perspective is consistent with their knowledge state. Although seeing is knowing is a reasonable heuristic (Moll & Tomasello, Reference Moll and Tomasello2007), not seeing is not knowing is unlikely to be. We rarely beep our horn as our neighbours reverse towards their houses without looking, because we know they know it is there. Regarding level-2 perspective taking, representing how someone sees something does not necessarily equate to representing their belief about the object. You can look at a number 6 from one angle, although we see it as a number 9 from another, without us holding differing beliefs about the object. By equating knowledge and belief with level-1 and level-2 perspective taking, respectively, Phillips et al. may be over-simplifying knowledge and over-complicating belief.
2. Mental-state sensitivity may not necessitate mental-state representation
Phillips et al. conclude evidence of automaticity based on studies documenting interference from another's perspective on judgements of one's own perspective (Kovács, Téglás, & Endress, Reference Kovács, Téglás and Endress2010; Samson et al., Reference Samson, Apperly, Braithwaite, Andrews and Bodley Scott2010). We question whether the incidental sensitivity to others' mental states revealed by these altercentric-interference effects necessitates their being represented. Several accounts leave open the possibility that other mechanisms underlie such mental-state sensitivity. As Phillips et al. note, submentalizing accounts propose lower-level, domain-general explanations of altercentric interference (Heyes, Reference Heyes2014). Two-systems accounts, in contrast, posit that altercentric interference reflects domain-specific registration of “belief-like states” (Apperly & Butterfill, Reference Apperly and Butterfill2009). Taking cues from both of these accounts, process-dissociation accounts hold that altercentric interference is not a process-pure index of mental-state registration; rather, such effects can be decomposed into at least two component processes: calculation of the agent's perspective and detection of one's own perspective (Todd, Cameron, & Simpson, Reference Todd, Cameron and Simpson2017, Reference Todd, Cameron and Simpsonin press; Todd, Simpson, & Cameron, Reference Todd, Simpson and Cameron2019), with the latter process likely reflecting something more domain-general (Payne, Reference Payne2005). Phillips et al. reason that knowledge representation is more basic based on evidence from tasks finding altercentric interference, but this could be evidence of arbitrary correlation between “knowledge” and stimulus features, coupling to more basic mental-state-like states, or poorer discrimination from self-knowledge.
3. Automaticity is not categorical
Phillips et al. aim to categorize knowledge and belief representation as automatic or not. Such categorization is likely an over-simplification of how cognitive systems operate. Automaticity is not a single feature, but rather a set of conceptually separable features that often do not co-occur (Melnikoff & Bargh, Reference Melnikoff and Bargh2018; Moors & De Houwer, Reference Moors and De Houwer2006). Thus, specifying in what way(s) belief and knowledge representation are automatic is crucial. The automaticity features receiving most empirical attention are goal-independence and efficiency. We agree with Phillips et al. that current evidence supports level-2 altercentric interference and process-dissociation estimates of agent-perspective calculation as consistently not-automatic. On a level-2 task, when participants only ever considered their own perspective, we showed that altercentric interference (Surtees et al., Reference Surtees, Apperly and Samson2016a) and agent-perspective calculation (Todd et al., Reference Todd, Cameron and Simpsonin press) were absent, suggesting level-2 perspective taking is goal-dependent. Using the same task, Todd et al. (Reference Todd, Simpson and Cameron2019) found that time pressure also impaired agent-perspective calculation, suggesting it is relatively inefficient. Evidence for level-1 automaticity is more equivocal. Some studies suggest level-1 altercentric interference emerges regardless of participant task goals (Conway, Lee, Ojaghi, Catmur, & Bird, Reference Conway, Lee, Ojaghi, Catmur and Bird2017; Surtees et al., Reference Surtees, Apperly and Samson2016a); others do not (Ferguson, Apperly, & Cane, Reference Ferguson, Apperly and Cane2017; Todd et al., Reference Todd, Cameron and Simpsonin press). Some studies suggest level-1 altercentric interference and agent-perspective calculation are efficient, in that they were unimpaired by time pressure (Todd et al., Reference Todd, Cameron and Simpson2017, Reference Todd, Simpson and Cameron2019) or a concurrent resource-consuming task (Qureshi, Apperly, & Samson, Reference Qureshi, Apperly and Samson2010); another found the opposite (Qureshi & Monk, Reference Qureshi and Monk2018). Although Phillips et al. acknowledge empirical uncertainty, our view is that different mental-state representations are not necessarily categorized fully as automatic or non-automatic. The impact of context on automaticity further supports this contention.
4. Mental-state representation may differ in observational and interactive contexts
Processes for interaction are different when directly engaging with another person, as opposed to passively observing them (Schilbach et al., Reference Schilbach, Timmermans, Reddy, Costall, Bente, Schlicht and Vogeley2013). Phillips et al. largely focus on observational, “third-person” approaches to adults' mental-state representation. It is not self-evident that the more “basic” form of mental-state reasoning in interactive and observational scenarios will be equivalent. Here, level-2 perspective taking provides an example of where they are not. Above we highlighted that level-2 perspective taking in an observational setting seems to be goal-dependent (Surtees et al., Reference Surtees, Apperly and Samson2016a; Todd et al., Reference Todd, Cameron and Simpsonin press). In a closely matched interactive context, however, participants do suffer interference from a partner's perspective, even if they are never asked to report it, suggesting it is goal-independent (Elekes, Varga, & Király, Reference Elekes, Varga and Király2016; Surtees, Samson, & Apperly, Reference Surtees, Samson and Apperly2016b). One possibility is that representations of different mental states are differentially cued by different aspects of social interaction. In this case, perhaps low-level stimulus features cue level-1 perspective taking, whereas “real” interaction or a specific goal may be required to cue level-2 perspective taking. One way to interpret this is that level-1 perspective taking is more basic, but another is to see level-2 perspective taking as linked to social interaction in a more fundamental way. Classifying knowledge or belief representation as more basic may obfuscate subtle variation with context.
Financial support
This research received no specific grant from any funding agency, commercial, or not-for-profit sectors.
Conflict of interest
None.