Theory of mind tested by implicit false belief: a simple and full‐fledged mental state attribution

Abstract

About 40 years have passed since ‘theory of mind (ToM)’ research started. The false‐belief test is used as a litmus test for ToM ability. The implicit false‐belief test has renewed views of ToM in several disciplines, including psychology, psychiatry, and neuroscience. Many important questions have been considered via the paradigm of implicit false belief. We recently addressed the phylogenetic and physiological aspects of ToM using a version of this paradigm combined with the chemogenetic technique on Old World monkeys. We sought to create animal models for autism that exhibit behavioral phenotypes similar to human symptoms. The simultaneous manipulation of neural circuits and assessments of changes in phenotypes can help identify the causal neural substrate of ToM.

The implicit false‐belief test has renewed views of Theory of mind (ToM) ability. We addressed the phylogenetic and physiological aspects of ToM using the test combined with the chemogenetic technique on old‐world monkeys. It may create animal models for autism that exhibit behavioral phenotypes similar to human symptoms. The simultaneous manipulation of neural circuits and assessments of changes in phenotypes can help identify the causative neural substrate of ToM.

Abbreviations

AL

anticipatory looking

DREADD

designer receptor exclusively activated by designer drugs

FB

false belief

FB

false belief

IFG

inferior frontal gyrus

IPL

inferior parietal lobule

LING

lingual gyrus

MFG

middle frontal gyrus

mPFC

medial prefrontal cortex

MTG

middle temporal gyrus

NB

no belief

PC

precuneus

SMG

supramarginal gyrus

STS

superior temporal sulcus

TB

true belief

TempP

temporal pole

ToM

theory of mind

TPJ

temporo‐parietal junction

VoE

violation of expectation

Introduction

Human beings conduct themselves based on others’ predictions and beliefs as well as those of their own. Theory of mind (ToM), the ability to attribute mental states of self and others to predict their behaviors, is indispensable to humans who adapt and live with both cooperation and competition. ToM has been studied in multiple academic fields, such as comparative psychology, developmental psychology and psychiatry, and cognitive neuroscience. This review focuses on the implicit false‐belief (FB) test, which is a widely used paradigm for testing ToM ability in human adults, infants, and non‐human primates (NHPs) (Fig. 1). Many essential questions that this paradigm has considered are as follows: Is ToM an innate capacity? Is it unique to humans? Does it involve single or multiple processes? What is the neural correlate of ToM? Is a neural circuit causally related to ToM? We start with a brief timeline and discuss the importance of recent advances in the neuroscientific approach to understanding ToM using implicit FB.

Fig. 1.

A timeline of ToM research. This figure shows the development of the concept and experiment on ToM and FB tests (including both explicit and implicit ones) in chronological order. The goal of the timeline is to show the first report of each paradigm rather than providing a comprehensive overview.

Concept of the theory of mind

Theory of mind was first introduced by Premack and Woodruff, American primatologists who studied chimpanzee communication skills. In ‘Does the chimpanzee have a theory of mind?’ [1], they questioned whether chimpanzees that form social groups and live in collective communities might be able to act based on inferences of their companions’ unobservable mental states. They called the ability underlying such behavior ToM, defining it as the ability to attribute intents, beliefs, thoughts, doubts, speculations, and preferences to self or others [1]. They named it a ‘theory’ because the invisible states of others’ minds are inferred like the laws of gravity, electricity, and magnetism rather than directly observed. Once the theory was constructed, it became possible to predict others’ behaviors based on the ToM, just as a scientific theory can be used to predict the occurrence of various phenomena [2, 3].

A critical benchmark for ToM: false‐belief test

Dennett, an American philosopher, suggested that attributing FBs to others is an appropriate diagnostic situation for ToM possession. According to Dennet [4], it is critical to create a situation where others’ beliefs and reality diverge in content. If others’ beliefs are in line with reality [that is if they have true beliefs (TBs)], predicting others’ behaviors is possible based on the actual states of the world and not others’ beliefs [4]. The original FB test was developed to investigate the developmental process of ToM in early childhood [5]. The test was an unexpected object transfer type of FB test, and its specific story was as follows. ‘Maxi went out to the playground after putting his chocolate in the green cabinet. While Maxi was absent, Mom took the chocolate out of the cabinet, used a little of it to make a cake, and put it in the blue cabinet instead of the green. Mom went shopping, and Maxi came home hungry after a while’. After telling the short story to a child, researchers would ask, ‘Where does Maxi think chocolate is?’ This story creates a sharp distinction between others’ beliefs and reality as well as one’s own beliefs. Most 3–4‐year‐olds could not answer correctly, choosing the blue cabinet (reality). If the child answered that Maxi thinks it is in the green cabinet, they could represent Maxi’s FB. The proportion of children who gave the correct answer increased from 4 to 7 years old. Therefore, they concluded that the ToM ability starts at about 4 years old [5, 6]. After that, FB comprehension was extensively tested [7, 8] using FB tests under various scenarios with different presentations and measurements; thus, reproducibility and essential factors, such as the motive of the story, the salience of the FB, the subject’s participation in the task (e.g. helping an actor), and the object's presence/absence, have been elucidated [6].

Loss of ToM in autism spectrum disorder

The FB task has been applied to both developmental psychology and psychiatry, and it has made outstanding contributions. Autism spectrum disorder (ASD) is characterized by atypical social communication problems and restricted and repetitive behaviors [9]. Baron‐Cohen et al. [10] hypothesized that ToM dysfunction is behind atypical social communication after researching ToM in ASD. They reported that even in high‐functioning autistic children with an average age of 11 years and 11 months (mental age of 9 years and 3 months according to a nonverbal intelligence test and of 5 years and 5 months according to a verbal intelligence test), the proportion of children who gave the correct answer for the FB test was about 20%. However, the rates of typically developing children with an average age of 4 years and 5 months and Down’s syndrome children with an average age of 10 years and 11 months (mental age was 5 years and 11 months according to a nonverbal intelligence test and 2 years and 11 months according to a verbal intelligence) were 85% and 86%, respectively. In the study, the test characters were changed to two girls, and it was named the ‘Sally–Anne task’. From the results, Baron‐Cohen et al. proposed the ‘mind‐blindness’ theory, arguing that the core disorder of ASD is caused by a lack of ToM [11, 12, 13].

Furthermore, the correct answer rate for the FB test was lower in the ASD group than in the Down's syndrome group, which had a lower intellectual level than the ASD group, so ToM dysfunction is not due to general intellectual disability but rather the deletion of a specific function. Therefore, it was expected that this task could play a role as a differential diagnosis tool for ASD. However, it turned out to be problematic, as daily social life difficulties are not always reflected in the test results [14].

‘Implicit’ FB test: Is ToM an innate capacity?

Passing the original explicit FB test requires a high cognitive load of cognitive abilities, such as memory, reaction inhibition, and the verbal ability to understand instructions, in addition to ToM ability [14, 15]. Therefore, someone’s failure to pass the test may not prove that their ToM is impaired. A simpler and entirely nonverbal FB task, now called the ‘implicit’ FB task, was conceived as a cognitively simple, nonverbal variant of the verbal FB task. On the test, the looking behavior during a scenario presentation is measured without verbal instruction [16] (thus, the classical tests with verbal instruction and elicited response are called ‘explicit’ FB tests). In general, implicit FB tests are always nonverbal, but explicit ones include verbal and nonverbal. As we will discuss later, measures in nonhumans are always nonverbal but regarded as explicit depending on the response type. It is typically considered as implicit when one measures eye gaze and as explicit when one measures behavioral choice such as reaching.

Onishi and Baillargeon [16] used the violation‐of‐expectation (VoE) method, which has been used extensively to investigate infants’ understanding of others’ goals [17, 18, 19]. For example, in one experiment [17], infants were familiarized with an actor reaching for and grasping one of two toys (defined as the target toy). Next, the locations of the two toys were reversed, and the actor reached for the target or nontarget toy. The infants looked reliably longer at nontarget reaches. The results suggested that the infants encoded the target toy as the actor’s goal object, expected her to reach for it in its new location, and responded with increased attention when she did not. Onishi and Baillargeon’s experiment [16], involved a woman (agent) sitting behind a tabletop facing a yellow box on one side and a green box on the other. Three familiarization trials were introduced. First, the agent placed a watermelon slice inside the green box. Then, in the second and third trials, she simply reached into the green box and paused. The belief induction phase involved the object moving so that the agent should either have a true or a false belief. In the TB condition, the agent remained visible behind the boxes. As the infants watched, the watermelon moved from the green box to the yellow on its own accord. During the test phase, the woman then reached into the green or yellow boxes, and it was analyzed which scene the infants saw for a longer time. When the woman reached into the empty green box, the infants looked longer than when the woman reached into the yellow box. In the FB condition, the agent’s view of the watermelon’s transfer was occluded. The watermelon then moved just the same as in TB condition from the green box to the yellow. The infants that saw FB condition, this time, looked longer when the agent reached into the yellow box with the watermelon. This was interpreted as that the infants’ expectation that the agent would reach for the place where the agent had last seen the melon had been violated (the agent actually reached for where the melon was). They were ‘surprised’ by the unexpected situation and stared for longer. The study overturned the consensus of that time that children under the age of 3 years could not pass the FB task; the result was reproduced repeatedly with multiple paradigms of the implicit FB test (while there are also a number of recent failures to replicate, see ‘Reproducibility issues of implicit FB paradigms’ section). For example, by measuring the impact (interference) of others’ FB representation to one’s own TB representation, Kovács et al. [20] showed that 7‐month‐old infants reportedly changed the amount of time they spent looking at the screen depending on whether the character’s beliefs and the real scene were dissimilar. Their procedure using the VoE paradigm was used for ToM research in multiple NHPs, and multiple neural correlates research in ToM. The detailed procedure for the task is complicated, but the procedure is briefly shown below. They presented four different groups of infants with a video of a cartoon agent watching the ball rolling behind the occluder along with the table. The four situations included both the agent and infant had TB (AT‐IT), both the agent and the infants had FB (AF‐IF), the agent had TB and the infants had FB (AT‐IF), and the agent had FB and the infants had TB (AF‐IT). First, the infants looked longer at the scene with the AF‐IF condition than with the AT‐IT condition. Second, the infants looked at the AT‐IF condition longer than the AT‐IT condition. Last, in the AF‐IT condition, infants looked at the scene longer than in the AT‐IT condition. This finding suggests that infants responded by looking longer at the test outcome that violated the agent's beliefs, even if the outcomes were in line with infants’ beliefs, showing that infants calculated (potentially automatically) the FBs of agents, even from the age of 7 months. In Onishi and Baillargeon study [16] and Kovács et al. study [20], whether an event is FB congruent or incongruent is based on the assumption that FB attribution predicts that a particular object is in a particular place (in the former study, the object is in a box on the left or right; in the latter study, there is one box and the object is there or not), yet the indicative measure is the length of time the participant watched the incongruent scene longer, which is insufficient for location specificity.

Anticipatory looking (AL) paradigm in which the child's specific expectation of where the actor will search is measured could help address this issue. Southgate et al. [21] used an AL method to test a capacity closer to conscious experience than that available with VoE. They experimented on whether infants could predict an agent's movement after showing a scene in which a toy moved while the agent was not watching. They showed in an AL task that a 25‐month‐old child could correctly predict where an agent with an FB would search for an object. A bear doll hid a toy in one of the two boxes in the familiarization trials, while a female agent stared at the boxes. Her head was visible above a panel with two small doors, and there was one above each box. Two doors turned on after the bear hid the toy. The agent then opened the correct door and collected the toys. In a test trial, the agent saw the bear hide the toy in one box. While she was facing away, the bear retrieved the toy. She turned toward the boxes and the doors lit up. Most infants correctly anticipated the agent's behavior and looked at the door above the box where she falsely believed the toy to be hidden. The study had two test scenarios, FB1 and FB2, that included elements that controlled each other. In the FB1 scenario, the FB‐congruent box was the box where the infants and the agent saw the object last placed. In the FB2 scenario, the FB‐congruent box was not the same as the box in which the infants saw the object last‐placed because the object was moved to another box after the agent turned away. First, this was intended to control the tendency to see where the ball was last found. Second, in the FB1 scenario, the box that the agent last attended was not the box where the agent last saw the object to control for any potential tendency to focus on the box last fixated upon by the agent. If there was no evidence to suggest that infants had either type of bias, both FB1 and FB2 were equally possible, so both FB tests contained important control elements. That is, they included two FB scenarios instead of the TB scenario control. One reason for applying these two FB conditions is that these two conditions lead to contradictory predictions. The other is that TB controls are themselves challenging to interpret and are originally included in verbal FB simply because TB and FB conditions make opposing predictions, not because they offer any unique insight into belief reasoning. If the looking patterns of infants differed between FB1 and FB2, the success of these conditions as a whole could not be explained by the use of a particular nonmentalization strategy. Furthermore, removing the toy at the end was adjusted so that the reaction that drew attention to where the candy was located did not occur. The procedure was intended not to require response inhibition. If the toy remained on the scene, the participants might have made incorrect responses resulting from a ‘reality’ bias. The authors preferred to call the task ‘spontaneous’ rather than ‘implicit’ because the former does not presuppose the absence of conscious awareness [14]. However, compared to the explicit FB test, the variability of the result was large between studies (see also the next section) [22, 23]. These findings with preverbal infants suggest that humans may be equipped with an early emerging system for representing others' beliefs and are not necessarily inconsistent with the possibility that ToM is automatic and innate. Senju et al. [24] applied the AL paradigm to adult patients with high‐functioning ASD and reported that even if they passed the explicit FB task, they could not pass the implicit one. The trend was found to be similar in comparison with the pediatric age group [25, 26]. Taken together, it shows that the implicit FB task can be a biomarker of childhood or adulthood ASD. However, some studies have questioned the reproducibility of this task in recent years.

Reproducibility issues of implicit FB paradigms

In recent years, the reproducibility of the AL and VoE paradigms, which are implicit FB tasks that use looking behavior as a measurement, has been questioned [27, 28]. For example, for the AL paradigm, Kulke et al. [29, 30] performed the AL paradigm with a relatively large sample size of subjects of various ages from 2 years old to adults. The results have a relatively high nonreplication rate. A close replication study of the original AL task [21] in which original authors were involved obtained a negative result [31]. The key to reproducing it is not yet clear, but the high rejection rate of the familiarization phase in the replication is noteworthy, which may be due to changes in the environment of human society over the past 15 years. However, the reproduction rate for adults is relatively high [29]. It may be thus that the AL paradigm itself is valid. But the AL paradigms are difficult to control because AL may be closer to an explicit reaction or close to an implicit response. How close to one of these depends strongly on the details of the experimental environment. In fact, from a physiological perspective, there are two types of eye movements: voluntary and involuntary, and it is not easy to separate them. Also, the fact that there are so few examples of the AL paradigm being used other than false belief tests may cause distrust.

Is ToM unique to humans?

The implicit FB test suggested a lower ToM acquisition age and that this ability is independent of language. While ToM studies in humans have been advanced, studies have focused for decades on the original question: ‘Does the chimpanzee have a theory of mind?’ Studies have suggested that NHPs can infer and track the goals, perceptions, and knowledge that motivate others' actions [32, 33]. As a result of many in‐depth studies, the evidence for the existence of ToM in a broad sense, such as the ability to infer and track others' goals, perceptions, and knowledge, has been piling up [33, 34, 35, 36, 37, 38, 39, 40, 41]. However, among these studies at that point in time, there has been no evidence that NHPs can understand others' FB, In 2008, 30 years after Premack and Woodruff's paper, Call and Tomasello concluded in a review article [33] that ‘Yes in the broad sense’ and ‘No in the narrow sense’. ‘Yes in the broad sense’ means that chimpanzees can understand the perceptions, knowledge, goals, and intentions of others, including what they see, what they are trying to do, and what they want to do under certain conditions. ‘No in the narrow sense’ means that there is no evidence that NHPs, including chimpanzees, understand the FBs of others. The FB test is critical in verifying the presence or absence of ToM in the narrow sense. If ToM in the narrow sense is core ToM, then the FB test is also critical in core ToM (Table 1). FB tests in NHPs have been repeatedly attempted but failed in numerous tests directly aimed at this question [33, 42, 43]. In recent years, Krupenye et al. adapted a competitive scenario version of the implicit FB test by AL [21] for great apes, suggesting that they can at least implicitly understand others' FBs [44]. The authors measured eye‐movement responses of bonobos, chimpanzees, and orangutans watching two types of videos in the experiment. Compared to Southgate et al. [21], these videos depicted antagonistic scenarios that were more likely to attract the participants' attention. More than half of the participating apes looked at the correct region, which represented the screen region of the box expected to be approached based on the Agent's FB in both scenarios that they expected to be approached based on the FB agent. The result suggests that they may predict others' behaviors based on their FB. There was no significant difference in the results among the three species of apes. All three species showed first looks that predicted the agent's following action based on their understanding of the agent's FB in the video. Additionally, using an AL method, Kano et al. [45] conducted a version of the ‘goggles’ test, a more rigorously controlled experiment, to show that great apes' AL is done by mentalizing rather than responding to behavioral cues. The study prepared two barriers that looked the same from a distance, but one was transparent, and the other was not. The researchers had the participants experience in real life the differences between the transparent and opaque barriers in one population. The researchers then showed the belief formation scenario movie. The agent observed the ball moving through the barriers in the test movie, and it was shown to both transparent and opaque experienced individuals. At this time, the agents in the videos behaved exactly the same, so the expectations resulting from behavioral observations were controlled. As a result, only individuals in the learned group showed predictive eye movements based on FBs. Therefore, great apes appeared to predict agent behavior in FB tests, especially based on their own past perceptual experience of visual access. Other apes also passed the FB test in the interactive and nonverbal task, modeled after the infant paradigm [46]. These findings have increased the attention to the new phylogenetic boundary of ToM. Furthermore, great apes passed the interactive helping task, which is a nonverbal task but is classified as an explicit task [46]. In the AL, VoE, and interactive helping tasks, elements other than mentalizing are simplified compared to the explicit tasks that NHPs (apes) have repeatedly failed [47, 48, 49]. A phylogenetic sister group of Old World monkeys (the group that includes macaques) has not demonstrated implicit FB representational competence. Specifically, results from VoE FB tasks measured [42, 43, 50] that monkeys generally appear to understand others' knowledge but not beliefs. Positive findings have also indicated FB attribution in apes; this, coupled with negative findings in macaques, raises the possibility that a basic understanding of FBs evolved uniquely in the hominoid lineage. Hayashi et al. [51] challenged this view by testing Japanese monkeys (Macaca fuscata) using implicit FB with an AL measure. We applied three similar scenarios to previous studies, two competitive scenarios [44], and one noncompetitive conventional change location [21], to investigate whether these Japanese monkeys also understand others' FB. We found spontaneous gaze bias that indicated implicit FB attribution in Japanese monkeys [51]. We also found two lines of evidence for implicit FB attribution in macaques. First, the monkeys' first look was biased toward the FB target over the nontarget. Second, the overall time spent looking at the target area during the test phase was longer than the time spent looking at the nontarget area. The differential looking time scores, the difference between the target viewing time and the nontarget viewing time divided by their sum, were significantly biased to the FB target. Furthermore, there was no significant difference in the results between competitive and noncompetitive presentation stimuli in our experiment, although Krupenye et al. [44] emphasized the importance of the presentation stimulus being competitive.

Table 1.

Implicit FB studies of nonhuman primates. AL, anticipatory looking; FB, false belief; MPFC, medial prefrontal; VoE, violation of expectation.

References	Paradigm	Design similarities	Sample species and size	Experimental environment	Excluded samples	Device	Main results
Marticorena et al. [43]	VoE	Onishi and Baillargeon [16]	62 Rhesus macaques	Free‐ranging	58	Human's Judgment	Macaques looked significantly longer when a human experimenter who knew the place of hidden food reached the incorrect place rather than the correct place in TB condition, but the difference was not observed in FB condition
Martin and Santos [42]	VoE	Kovács et al. [20]	121 Rhesus macaques	Free‐ranging	214	Human's judgment	Macaques looked significantly longer at events that violated their own beliefs, but not those of a human experimenter, about where object was hidden
Krupenye et al. [44]	AL	Competitive ver. of Southgate et al. [21]*	41 great apes (19 chimpanzees, 15 bonobos, and 7 orangutans)	Captive	−	Eye tracker	In the first Look indicator, with one movie scenario apes did not passed the test but did passed with another scenario as well as when combining data across the two experiments
Kano et al. [45]	AL	Krupenye et al. [44]	47 great apes (29 chimpanzees, 14 bonobos, and 4 orangutans)	Captive	−	Eye tracker	In the first Look and Looking time indicators, Great apes predicted the behavior of other's based on his FB (only FB2 performed)
Hayashi et al. [51]	AL	Southgate et al. [21] and Krupenye et al. [44]	8 Japanese macaques	Captive (head fixation)	−	Eye tracker	In the first Look and Looking time indicators, Macaques predicted the behavior of other's based on his FB; significant differences between MPFC active/inactive with first look but not looking time

Sensitivity of implicit FB tests in nonhuman primates

In contrast to the situations in human studies where the experimental environments, subjects, and ages are thoroughly controlled and investigated, in NHP, the experimental environment differs. There is no test with apes in the VoE paradigm, and there is no test with monkeys in the VoE paradigm in a controlled laboratory situation. Different studies use different analytical methods and indexes, which may also affect the results. For example, one study used looking duration in the analysis of variance (ANOVA) models [44] and Southgate et al. [21], and another study used DLS in t‐tests [29, 30]. A richer comparison with a larger number of controlled experiments would advance our understanding of implicit FB.

Although the VoE paradigm has not been examined in Apes, the positive results for the AL paradigm in studies using great apes and Japanese monkeys [44, 45, 51] contrast the negative results for the VoE paradigm with rhesus monkeys reported [42, 43, 50]. Two explanations may account for this discrepancy. First, unlike human infants, monkeys tested with the VoE paradigm may exhibit equivalent looking time by being impressed by both the agent's unexpected action toward the nontarget (incongruent to the agent's FB) and the agent's deceived movement toward the FB congruent target. If monkeys can represent the agent's FB, they may show longer looking at the agent's FB incongruent scene by the preference for the agent’s deceived situation. Since NHP is more prone to attention to competitive situations, we speculate that NHP may be particularly focused on deceived situations from a survival perspective, such as food capture [39]. Thus, even if monkeys have representation of agent's FB, looking duration may reflect the strife between VoE and preference for deception. Second, the experimental environments are critical to allowing FB to manifest in VoE paradigms. The VoE paradigm, which had a negative result, was applied to macaques in a free‐ranging population, monkeys were free to walk away from the experimenters at any time during familiarization or the test trial, and only about 35–50% of all involved monkeys completed the investigation. Although all monkeys in the VoE studies that completed the test trial were included in the analyses [42, 43, 50]. Meanwhile, in our study, macaque monkeys were trained to gaze within the monitor range and had their heads fixed during the test. They completed 100% of the test, and about 80% of the data is analyzable [51], although it is difficult to make a direct comparison between voluntary participation and physically restrained situations. Besides, the more subtle differences including experimental setups (presence or absence of head fixation and observer position, familiarization procedure, video content, and combinations of these) affected the results. Therefore, it will be important to verify the VoE paradigm with our measurement method.

The failure/difficulty of reproducing the AL experiment in human infants [31] is noteworthy for research at NPH. In AL studies in NHPs, there are aspects of the task that are being optimized independently from infants, and these may lead to positive results. Since the results of AL studies in NHP have shown some reproducibility, it is hoped that they will be used as a clue for further understanding of implicit FB and ToM. In addition, since VoE, which is considered to be relatively reproducible in humans, has not shown positive results in monkeys, paying attention to species differences is also considered important for future understanding of ToM.

Is ToM a single or a dual process?

Originally, the development of the implicit FB test was an attempt to verify core ToM by reducing a load of cognitive abilities, such as memory, reaction inhibition, and verbal abilities, and ToM was assumed to be a single process [20]. Butterfill and Apperly have produced a new hypothesis worthy of assessment [52, 53]. ToM cognition generally involves metarepresentation [54]. Minimal ToM only requires representing goals to which actions are directed, encounterings, and registrations, none of which are representations. Thus, no metarepresentation is involved. The hypothesis is that humans have two ToM, ‘minimal ToM’ and ‘ToM proper’ and infants use ‘minimal’ ToM when solving implicit FB tests. Infants can solve implicit FB, but cannot solve explicit FB. It has been argued that adult humans also use the same minimal ToM as infants when predicting another person's behavior instantly [53]. Furthermore, unlike ToM proper, minimal ToM performs a simpler process [52, 53]. From the perspective of double dissociation, the combination of the deficit in implicit FB and intact explicit FB can be observed in the adult Asperger's syndrome population, and deficit in explicit FB and intact implicit FB can be observed in normal infants, but it is probably more a matter of memory, impulse control, and other abilities. Experimental testing of the hypothesis supported either the dual [55] or single hypothesis [56, 57]. This discussion has not yet been settled. This issue is also addressed in the neural correlates study, which will be discussed in the next section, and may help resolve it.

What is the neural correlate of ToM?

Pursuing the neural mechanisms of implicit FB provides critical insights into the understanding of the FB process. How is implicit FB related to explicit FB from a neuronal perspective? In human imaging studies, whole‐brain analysis has explored FB's neuronal network with explicit paradigms [58, 59, 60, 61]. A neuronal network has been revealed in studies that included the temporoparietal junction (TPJ), medial prefrontal cortex (mPFC), superior temporal sulcus (STS), and precuneus (PC). Among them, the TPJ and mPFC have been found to be consistently active in recent meta‐analyses [62, 63, 64, 65].

Most studies with implicit FB addressed comparing the neural substrates of implicit FB and explicit FB (Table 2) using functional magnetic resonance imaging (fMRI). The question was closely related to a single or dual‐process problem. The standard approach was that first, the specific regions of interest (ROI) identified by explicit FB task and the ROI activities were measured in implicit FB. The activities of the ROI were evaluated by the signal contrast between FB and TB conditions. Schneider et al. [66] measured brain activity during an AL task in adults. The study replicated eye movement bias across FB and TB conditions, although the procedure did not deal with reality bias. The results showed that only the left anterior STS and PC revealed more significant blood‐oxygen‐level‐dependent (BOLD) activity in the implicit FB process during the belief test phase. The results emphasized limited neural share between implicit and explicit FB. Naughtin et al. [67] examined using the same AL task in adults. The study introduced the nonbelief (NB) condition as a critical comparison. In the NB condition, the agent did not hold any pre‐existing beliefs about the object's location. Since the TB condition is considered to require some degree of mentalizing and is not suitable for comparison with FB (see discussion in explicit FB studies [58]), the authors compared FB conditions with NB ones, which have a lower level of metalizing. They revealed that the right TPJ, right STS, PC, and left‐middle prefrontal gyrus during the belief test phase showed significantly more increased activities in FB conditions than in no belief (NB) ones. The study concluded rather shared neural substrates of implicit and explicit FB. Nonreplication of behavioral results (no looking behavior bias based on FB recognition) made the interpretation of the imaging results difficult. Yet, studies using other behavioral paradigms or other measurements also supported the ‘sheared’ view; Kovács et al. [68] measured brain activity using fMRI in adults during a task using the VoE paradigm. They showed that the right TPJ and the mPFC are recruited by spontaneous belief tracking. Hyde et al. [69] used functional near‐infrared spectroscopy (fNIRS) to compare brain activity when responding explicitly to the AL task and when merely watching the video in adults. They found a significant increase in oxygenated hemoglobin concentrations in the right TPJ during the belief‐formation phase. In addition, they conducted a similar experiment in infants. They observed that activity in the TPJ spontaneously tracked the other person's beliefs, responding more during scenarios when the other person's belief regarding the object's location was false compared to scenarios when their belief was true [70].

Table 2.

Neuroimaging studies of the implicit FB test in human. a, anterior; d, dorsal; FB, false belief; i, inferior; IFG, inferior frontal gyrus; IPL, inferior parietal lobule; l, left; LING, lingual gyrus; MFG, middle frontal gyrus; MTG, middle temporal gyrus; NB, no belief; PC, precuneus; r, right; SMG, supra marginal gyrus; STS, superior temporal sulcus; TB, true belief; TempP, temporal pole; TPJ, temporo‐parietal junction; v, ventral.

References	Age	Sample size	Compared condition	Behavioral analysis	Device	Design	Activated regions during FB condition
Anticipatory looking paradigm
Schneider et al. [66]	22.70 ± 4.44 years	20 (9 males)	FB vs. TB	Significantly more likely to make their first fixation to the box that did not contain the ball (No ball location) when the actor had a FB that the ball was at that location as opposed to a TB that the ball was not at this location (P = 0.045)	fMRI	ROI (TPJ, PC, STS, l TempP, l MFG)	l anterior STS and PC during the belief test phase (focusing the belief test phase only)
Hyde et al. [69]	19.5 ± 1.53 years	25 (12 males)	FB vs. TB	Significantly more time looking to the actual location of the hidden object relative to the other location (P < 0.00005) in the TB condition while an equal amount of time looking at both locations in the FB condition (P = 0.32)	fNIRS	ROI (TPJ, MPFC)	r TPJ during the belief formation phase (focusing the belief formation phase only)
Naughtin et al. [67]	24.4 ± 4.0 years	33 (13 males)	FB vs. NB	No behavioral difference in eye movement between FB and NB conditions	fMRI	ROI (MFG, PC, TPJ, STS, TempP, dorsal MPFC, ventral MPFC)	r TPJ, r STS, PC, and left middle prefrontal gyrus during the belief test phase
Hyde et al. [46]	7.37 ± 0.80 months	20 (5 males)	FB vs. TB	Not shown	fNIRS	ROI (TPJ, MPFC)	r TPJ during the belief formation phase (focusing the belief formation phase only)
Grosse Wiesmann et al. [73]	4.07 (3.07–4.58) years	38 (17 males)	Correlation of implicit ToM Score	Marginally significantly more to the correct than the incorrect location (P = 0.064)	MRI	Whole (structure)	Surface area: r TPJ (r SMG)Cortical thickness: l PC
Violation of expectation paradigm
Kovács et al. [68]	21.6 (18–27) years	15 (6 males)	FB vs. TB	No main effect of agent's beliefs (Ball present vs. Ball absent)	fMRI	ROI (TPJ, MPFC)	r TPJ (no phase effect)
Bardi et al. [71]	22 (19–25) years	23 (5 males)	FB vs. TB	Significant main effect of agent’s beliefs (P < 0.05)	fMRI	Whole/ROI (TPJ, MPFC)	Whole: r TPJ (Angular gyrus) and Fusiform gyrus/collateral sulcus during the belief formation phase. a MPFC, r Sensorimotor, Both Occipital cortex, l a Insula, Supplementary motor area, Thalamus, LING during the belief test phase ROI: r TPJ during the belief formation phase a MPFC during the belief test phase
Boccadoro et al. [72]	31.13 ± 10.49 years	68 (17 males) including Bardi et al. [71]	FB vs. TB	Not reported main effect of agent's beliefs	fMRI	Whole/ROI (TPJ, MPFC)	Whole: r TPJ (r SMG, r MTG), r LING, r IFG, r MFG, I IPL, both thalamus during the belief formation phase ROI: TPJ (r > l)

There have also been a limited number of studies that use whole‐brain analyses; there have only been two fMRI studies with the VoE paradigm and one structural MRI study with the AL paradigm. Bardi et al. [71] compared brain activity during the implicit task (VoE paradigm) and the explicit task by fMRI in the same subject. In both versions, participants watched videos of a scene that included an agent who acquired a true or false belief about an object's location (belief formation phase). At the end of the movie (belief test phase), participants reacted to the object's presence. During the belief formation phase, greater activity was found for false versus true belief trials in the right TPJ (angular gyrus) and Fusiform gyrus/collateral sulcus and anterior mPFC, right sensorimotor, both occipital cortices, left anterior insula, supplementary motor area, thalamus, and LING during the belief test phase. The ROI analysis of the right TPJ also confirmed this observation. Moreover, the anterior mPFC was found to be active during the belief formation phase, meaning it is sensitive to the violation of both the participant's and the agent's expectations about the object's location. The implicit/explicit task did not modulate activity in the TPJ or anterior mPFC. Overall, these data showed that the neural mechanisms for implicit/explicit ToM overlap. Interestingly, dissociation was found between TPJ and anterior mPFC for belief tracking and outcome evaluation, respectively. Boccadoro et al. [72] pointed out the importance of large sample size for small to medium effect sizes, such as mentalizing studies and pooled data from three fMRI studies, including that by Bardi et al. [71], to better define the neuronal correlates to processing the implicit FB test. The study had 68 participants and resulted in a significantly stronger signal captured by the right TPJ (rSMG and rMTG), lTPJ (rLING, rIFG, rMFG, and iIPL), and both thalami during the belief formation phase. The TPJs were more active on both sides in the FB condition than in the TB condition. Furthermore, a stronger tendency was found on the right than on the left side. Grosse Wiesmann et al. [73] computed the linear relation of 3‐ and 4‐year‐old children's implicit ToM performance with cortical surface area and thickness on the whole cortical surface. They yielded a significant positive correlation with surface area in the right SMG and with cortical thickness in the left PC (within the regions of adult ToM meta‐analysis [63]). Although a brief description of whole‐brain analysis was found [66, 67], the detail was not mentioned. One‐way pass analysis, applying ROI of explicit FB paradigm to implicit FB, examined only necessity conditions. The other direction or independent measure for implicit FB will be required to compare the neural substrate for implicit FB and one for explicit FB.

In summary, TPJ and mPFC were listed as strong candidates in the explicit FB test, while many candidates were raised in the implicit FB test; among them, TPJ currently seems to be the most influential (Fig. 2). The expression pattern of mPFC is probably the most noticeable. mPFC activity, which is consistently observed in explicit FB, is not necessarily consistent in implicit FB. While TPJ activity is consistently observed in the implicit FB test, the contrast between FB and TB does not reveal reliable differences in mPFC activity. Since TB may reflect some degree of belief attribution/tracking just like TB in explicit FB paradigms [58], mPFC involvement can be underestimated. Negative findings may result from inappropriate condition contrast or due to the spatial resolution limitations of fMRI measurement. It is noteworthy that a recent single‐cell recording from human mPFC in explicit FB paradigm revealed that single cells distinguished false and true belief representations, but the response amplitude in each cell can be larger or smaller in false representation. Thus, no net difference in activities for false and true belief representations was observed [74]. mPFC still can underlie the core role of belief attribution but does not involve the components of belief formation specific to FB.

Fig. 2.

Neural correlates of implicit FB in human. This figure shows the neural correlates of the implicit FB in human neuroimaging studies listed in Table 2. The brain regions observed with reproducibility in more than one study are surrounded by a solid line.

There are also some contrasts between explicit and implicit FB‐related areas; for example, SMG and rLING seem to be implicit‐specific regions [72, 73] that seem to be components of implicit FB that accompany sensory‐related processing. Activity seen in the subcortical thalamus regions is also remarkable [72].

Furthermore, Nijhof et al. [75, 76] measured fMRI during the VoE task in ASD patients (mean age = 32.8 years, SD = 8.6 years) and healthy controls (mean age = 31.1 years, SD = 8.4 years). Their results showed reduced activation in the anterior middle temporal pole in ASD for false‐ versus true‐belief trials. Furthermore, the ROI analysis of rTPJ showed more activity when an agent formed FBs than when forming TBs, especially if the agent's beliefs had positive content. The effect was not seen in individuals with ASD. The results suggest a neural difference between ASD and affiliated developing individuals in suggestive mentalization, indicating that rTPJ is decisively involved in ASD. These studies suggest the importance of implicit FB as a core ToM test for investigating neural mechanisms. However, since there are inadequate number of neuro‐functional studies on implicit FB, some of the behavioral procedures in each study were not aligned with those in standard behavioral studies, further evidence must be acquired.

Is a neural circuit causally related to ToM?

However, the parallel behavior between humans and apes and monkeys is not necessarily accomplished by common neural mechanisms conserved between species.

Thus, it is possible that the sprouting of an ability to implicitly comprehend others' mental states by neural circuits centered on the mPFC might go back to the ancestry of a primate species common to humans and macaques.

Thus, a critical question to be asked is whether FB‐attribution‐like behaviors in NHPs originate from functions of homologous neuronal circuits to those of humans. Hayashi et al. [51] elucidated that the ability to attribute FBs to others implicitly is supported by homologous neural circuits between humans and monkeys. The results indicated that the mPFC plays an indispensable role. We focused on the mPFC because neuronal activity in this region can encode others' actions in specific contexts [77, 78] and other mentalizing functions [79]. TPJ can be a target candidate, but its functional homologs in the macaque are still under debate [80]. We injected a lentiviral vector [81] incorporating hM4Di, an inhibitory designer receptor exclusively activated by designer drugs (DREADD) [82], into the mPFC around the medial part of area 9 (9 m) [51]. After the gene expression period, we chemogenetically inactivated the mPFC via the intramuscular injection of clozapine N‐oxide (CNO), a specific ligand to hM4Di. Within 60–80 min after CNO injection, the activation of mPFC was significantly reduced. Then, we tested for implicit FB ability; the gaze bias to the FB target disappeared both in terms of the first‐look ratio and differential looking score (DLS), while other reference abilities, such as oculomotor object tracking and memory, remained intact. The results provide evidence for a causal link between neuronal activity in the mPFC and the manifestation of FB‐attribution‐like behaviors in primates. Thus, it is possible that the sprouting of an ability to implicitly comprehend others' mental states by neural circuits centered on the mPFC might go back to the ancestry of a primate species common to humans and macaques.

Noninvasive studies in human subjects or observations of patients have suggested that the mPFC activation in implicit FB has not been evident clearly (as discussed in the previous section). It is particularly important to note that the positive behavioral results were accompanied by the employed AL paradigm in this study. The causal approach, in principle, avoided the inappropriate contrast problem that appeared in the correlational studies. However, the specificity of the effects should be carefully considered as the study examined low‐level controls such as attention and memory. Furthermore, how these or other brain regions function as networks cannot be determined solely from fMRI or neuropsychological studies. Spatial and temporal resolutions are inadequate for identifying which functional connections between regions are responsible for this complex process. In addition, elucidating the causal relationship between network behavior and higher functions requires manipulating neural activity and examining its behavioral effects; therefore, studies with laboratory animals are necessary. The implicit FB paradigm also significantly contributes to exploring neural mechanisms because the lack of language and the low task demands are consistent with the conditions applied to animal models.

Future outlook

Although molecular biological models for autism have mainly used rodents [83], their indexes for behavioral assessments have been limited. It will be essential to study function while simultaneously investigating both the homology of both neural circuits and behavioral phenotypes in monkeys whose anatomical circuitry in relation to humans has been well studied. In recent years, specific neural circuits in monkeys have been manipulated by chemogenetic methods [82], or even genetically modified monkeys [84, 85] have been created. Furthermore, many behavioral experiments have been conducted using the index modified from those in rodents. It will be essential to validate the knowledge with implicit FB or other comparable human social cognitive tasks in the future.

Recent converging evidence from anatomical, physiological, and theoretical research suggests the fast‐track modulator model. The subcortical face detection pathway mediates eye contact processing, referred to as the retinotectal pathway or extrageniculate pathway, including the pulvinar, superior colliculus, and amygdala (Fig. 3) [86, 87]. The pulvinar is especially associated with directed attention, executive function, and working memory, which are impaired in neurodevelopmental disorders, including ASD, schizophrenia, and attention deficit hyperactivity disorder (ADHD) [88]. In parallel, the monkey medial portion of the pulvinar is reciprocally connected to area 9 of mPFC and anterior STS, which is one of the human TPJ homologs in the macaque [27, 78]. Thus, the brain‐wide network containing the pulvinar, mPFC, and TPJ (anterior STS in monkeys) may be responsible for implicit FB and, therefore, full‐fledged ToM. This demonstration will be possible by extending our DREADD‐based preparation. From a neuroscience perspective, it is essential to emphasize the understanding of how ToM functions, which will also contribute to an essential understanding of ASD.

Fig. 3.

A hypothetical brain circuit for implicit FB. This figure shows a hypothesis on neural network for implicit FB attribution. Blue and red arrowhead line depicts the pathway for the geniculate and extrageniculate visual system respectively. The purple dashed two‐sided arrows that connect the mPFC, rTPJ and pulvinar represent our hypothesis that the geniculate and the extrageniculate visual systems are incorporated to the implicit FB attribution. Purple two‐sided arrows indicate another known interaction of the geniculate and the extrageniculate visual system through amygdala and inferior temporal cortex.

Conflict of interest

The authors declare no conflict of interest.

Author contributions

JE and KK wrote and revised the manuscript. TH, RA, and KK analyzed the data. TS and IH supervised the manuscript.

Data availability statement

The data that support the claim of this review are available from the corresponding authors upon request.