The Neural Development of Trust and Cooperation
The Neural Development of Trust and Cooperation
This study examined age-related changes in the neural correlates of trust and cooperation during social interactions. Participants played trust games with two hypothetical partners with a cooperative and an unfair decision-making style. Age was associated with higher levels of trust at the onset of social interactions and throughout interactions with a cooperative partner, but also with a steeper decline in the levels of trust throughout interactions with an unfair partner. Associations between age and increased activation in the left TPJ were present during investment decisions towards both game partners. During cooperative interactions, activity in the orbitofrontal cortex and caudate nucleus decreased with age, but increased in the posterior cingulate/precuneus region. The anterior cingulate became increasingly responsive to unfair behaviour as age increased. The findings may reflect a shift from negative to positive expectations of trustworthiness and suggest that differential neural activation patterns in brain areas associated with mentalizing, reward learning and cognitive control may underlie age-related increases in the sensitivity to others' social signals.
In the trust game, mentalizing and social reinforcement learning are important for strategic reasoning about the game partner's intentions and to infer how the game partner perceives one's own behaviour (King-Casas et al., 2005; Sanfey, 2007). As hypothesized, age was not only associated with higher initial trust but also with higher trust during interactions with a cooperative partner and with a steeper decline in trust during interactions with an unfair partner. Still, all participants showed a learning effect regardless of age. They increased their levels of trust in response to cooperation and decreased their levels of trust in response to unfair behaviour. The current imaging findings are in line with hypothesized increases in mentalizing. A better understanding of how the game partner will react and interpret one's own behaviour may support and initially higher investment, a move that signals the intention to cooperate. Age-related increases in activity in the left TPJ were present regardless of the nature of the game partner. The TPJ has been described as part of the 'mentalizing system' (Fletcher, 1995; Ruby, 2004; Gobbini et al., 2007; Van Overwalle, 2009) and trust game research found this area activated when the investor's decision was revealed to the trustee (Van den Bos et al., 2011b). We found increased activation within the TPJ during investment decisions. Possibly, this area plays a role in mentalizing during decisions about how much to trust that are made while predicting the game partner's behaviour. The current data also showed a specific age-related activation during cooperation in the right posterior cingulate/precuneus region which also has been described as part of the mentalizing network (Van Overwalle, 2009; Wolf et al., 2010). Our results show that different areas of the mentalizing network are differentially activated during social interactions. The TPJ is involved regardless of the nature of the social interaction and its activation increases with age. However, the posterior cingulate/precuneus region shows specific activation in response to trustworthiness. In contrast to Van den Bos et al. (2011b), current age-related changes in the mentalizing network were predominantly located in the posterior and not the medial prefrontal brain areas. This discrepancy could be explained by the differential nature of the tasks, the specified contrasts or age differences in the samples (i.e. 12–22 vs 13–49 years).
In order to make sensible decisions to trust, humans need to learn the associations between their behaviour and the feedback that they receive from others in response. For example, if a certain person answers trust with betrayal, we are less likely to trust that person a second time. However, if the person proves to be trustworthy, we are more likely to trust this person in future (King-Casas et al., 2005). Thus, reward learning shapes behaviour towards optimal decision making. In this study, age was associated with higher levels of trust at the onset of anonymous social interactions. This suggests an increased expectation of benevolence of others and offers a suitable explanation for why feedback learning during cooperation becomes less important as individuals get older. Where cooperation is anticipated, experiencing cooperative behavioural feedback from the trustee matches the prediction. Therefore, learning of new behaviour–outcome associations is unnecessary. fMRI research showed that the orbitofrontal cortex is important for feedback learning and subsequent goal-directed behaviour. It is thought to signal reward value during decision making and is sensitive to value changes, i.e. becomes less responsive when stimuli are less novel or expected (Rolls, 2000; O'Doherty et al., 2001; Rolls and Grabenhorst, 2008). The caudate also plays a role in the anticipation and reception of rewards and has been found to decrease signalling when certain predictions match an outcome (Delgado et al., 2005; Schiffer and Schubotz, 2011). Accordingly, we expected that with increasing age, brain areas that are important for reward learning (e.g. ventromedial prefrontal and orbitofrontal cortex and the striatum) would be recruited to a lesser extent. During cooperative interactions, we found age-related decreases in orbitofrontal cortex and caudate nucleus activation. We did not find the reverse pattern during interactions with an unfair partner. This is in line with previous trust game research of Phan et al. (2010) who found that only positive reciprocity by the partner engages the ventral striatum and orbitofrontal cortex. Also, King-Casas et al. (2005) found the caudate specifically active during cooperation.
As indicated by increased levels of trust, the anticipation of benevolent behaviour by others increases with age and responding to deviations of trust requires a stronger correction in response to an unfair interaction partner. Hence, the interaction is more demanding for older individuals, as it requires them to adapt their beliefs and behaviour. Younger people need to make fewer adjustments to their investments because of their more distrusting mindset. In line with this reasoning, we found age to be associated with a stronger decrease in trust during the interactions with an unfair partner. During these interactions, expectations of cooperation will have resulted in cognitive conflict in the face of the actual returns. The need to reduce the levels of trust was not reflected in increases in activation in brain areas associated with reward learning, but activation in the dorsal anterior cingulate was modulated by age and the nature of the game partner. This area of the anterior cingulate is known to play a role in conflict monitoring (Botvinick et al., 1999, 2004; Bush et al., 2000; Greene et al., 2004; Ridderinkhof et al., 2004) and there is evidence that it is important for behavioural adjustments (Kerns, 2004; Magno et al., 2006). With age, participants increasingly recruited the anterior cingulate during unfair when compared with cooperative interactions. This may reflect the conflict between expectations of cooperation and experienced social feedback and shows a possible explanation of how feedback shapes future decisions towards more optimal levels of trust (Chang and Sanfey, 2013).
Finally, we found age-related changes in brain areas that were not part of our hypotheses, but that previously have been associated with social cognitive processes. During cooperative interactions, age was also associated with activation in the precentral and bilateral middle frontal gyri. These areas have been described as parts of the 'mirror system' (Van Overwalle and Baetens, 2009), a network which is thought to enable humans to understand the goals of observed (physical) actions of others in an intuitive way by internal simulation. Some evidence indicates that this brain network also engages in more abstract forms of social cognition, such as mentalizing (Vogeley, 2001; Wolf et al., 2010) and empathy (de Greck et al., 2012). Previous trust game research found this area to be activated during decisions to trust (Delgado et al., 2005) and age-related increases of activation have been found during social perception (Beadle et al., 2012). During unfair interactions, the middle cingulate gyrus was increasingly active with age. This area has been suggested to function as a relay node between negative emotions and motor action and has been reported to be strongly active during decisions to trust in earlier trust game research (King-Casas et al., 2005). With age, activation in the dorsomedial prefrontal cortex decreased during cooperative interactions. Earlier research showed this area to be involved in the management of uncertainty in decision making, whereby more uncertainty was associated with increased activation (Volz et al., 2005).
The current results have to be interpreted in the light of some limitations. First, it might be possible that the differences in trust that we see with increasing age could be due to age differences in risk aversion.While a higher risk aversion may influence the degree of trust, there is extensive literature, which shows that age is typically positively associated with increased risk aversion (Steinberg and Sheffield Morris, 2001; Steinberg, 2004; Deakin et al., 2004; Burnett et al., 2010; Paulsen et al., 2011, 2012). This implies that teenagers should invest more money if risk aversion would be an important motive in the trust game. Furthermore, the concern over whether behaviour in the trust game actually measures trust or risk attitudes has frequently been raised. Eckel and Wilson (2004) provide a comprehensive analysis of the way behaviour in two-person sequential trust games correlates with a variety of behavioural and survey-based risk measures (Eckel and Wilson, 2004). They did not find evidence that any of their risk measures predicts the decision to trust. Also, Houser et al. (2006) found that in a risk game with a computer as counterpart, the probability of investing more was significantly higher for risk seeking subjects than for subjects in the risk averse group but risk aversion did not predict investments towards a human counterpart (Houser et al., 2006). Second, our manipulation check showed that six individuals of the adolescent (n = 25) and one of the adult group (n = 20) did have doubts that they were playing with a real human. The fact that fewer adolescents believed that they were playing a human may have reduced their mentalizing operations. Yet, the current behavioural results show increases in trust with age that are in line with those of other research and do not support such concerns (Sutter and Kocher, 2007). Third, it is important to note that age-related effects may also be caused by differences in neurovasculature (Harris et al., 2011). Yet, research by Kang et al. (2003) showed that in voxelwise group comparison of images of visual and motor cortex regions, only minimal differences were found between children of 7 and 8 years and adults. The small differences in time courses and locations of activation foci between child and adult brains do support the feasibility of direct statistical comparison of these groups within a common space. Fourth, it should be noted that other cognitive factors that may offer an alternative explanation for age-related changes in trust game performance to increases in mentalizing, such as executive functioning or IQ, have not been included as confounders in this study. While evidence from other research of Van den Bos et al. (2011b) supports that non-social cognitive factors such as intelligence have a modest influence on social decision making, it would be valuable if future studies would include additional measures of theory of mind and other cognitive functions to validate the interpretation in terms of mentalizing. Fifth, subjects knew that they were taking part in a study with participants between 13 and 18 and 19 and 49, respectively. It is therefore likely that the current effects pertain to trusting behaviour regarding individuals within these age ranges. Trusting behaviour towards other age groups may differ from the current results. It would therefore be interesting for future research to systematically examine age effects with respect to the age of the trustee. Sixth, in the current paradigm, we did not directly control for aspects of monetary reward in the trust game. However, as a means to control that age-related differences in brain activation are not merely due to differences in the investments that were made, we did analyse the association between the invested amounts and brain activation (Supplementary Table S2). Age-related increases and decreases in brain activation in certain areas did not overlap with the foci that were associated with higher and lower investments. The associations between brain activation and age do not seem to be caused by differences in investments. Finally, fMRI allows for the investigation of the role of certain brain regions in certain cognitive functions and caution is required when cognitive processes are inferred from activation in specific brain regions (Poldrack, 2006). Within the brain, it is unlikely that a particular region is activated solely by one cognitive process. The current interpretations are in line with the growing literature about the social brain systems and should be regarded as a guide for future inquiries rather than direct explanations of certain findings.
Our findings render preliminary support to our working hypothesis of improved mentalizing as an underlying mechanism of age-related behavioural preferences of trust and cooperation. Initial trust increased with age. During interactions with a cooperative partner, older people continued to make higher investments. However, when playing with an unfair game partner, older people quickly reduced their levels of trust to those of younger ones. The neuroimaging data showed age to be associated with increased recruitment of brain areas that seem to be important for mentalizing. In line with a stronger preference for cooperation, age was associated with decreased activation in areas involved in reward learning in response to cooperative behaviour by the partner and increased activation in areas associated with cognitive control in response to an unfair game partner.
Discussion
This study examined age-related changes in the neural correlates of trust and cooperation during social interactions. Participants played trust games with two hypothetical partners with a cooperative and an unfair decision-making style. Age was associated with higher levels of trust at the onset of social interactions and throughout interactions with a cooperative partner, but also with a steeper decline in the levels of trust throughout interactions with an unfair partner. Associations between age and increased activation in the left TPJ were present during investment decisions towards both game partners. During cooperative interactions, activity in the orbitofrontal cortex and caudate nucleus decreased with age, but increased in the posterior cingulate/precuneus region. The anterior cingulate became increasingly responsive to unfair behaviour as age increased. The findings may reflect a shift from negative to positive expectations of trustworthiness and suggest that differential neural activation patterns in brain areas associated with mentalizing, reward learning and cognitive control may underlie age-related increases in the sensitivity to others' social signals.
In the trust game, mentalizing and social reinforcement learning are important for strategic reasoning about the game partner's intentions and to infer how the game partner perceives one's own behaviour (King-Casas et al., 2005; Sanfey, 2007). As hypothesized, age was not only associated with higher initial trust but also with higher trust during interactions with a cooperative partner and with a steeper decline in trust during interactions with an unfair partner. Still, all participants showed a learning effect regardless of age. They increased their levels of trust in response to cooperation and decreased their levels of trust in response to unfair behaviour. The current imaging findings are in line with hypothesized increases in mentalizing. A better understanding of how the game partner will react and interpret one's own behaviour may support and initially higher investment, a move that signals the intention to cooperate. Age-related increases in activity in the left TPJ were present regardless of the nature of the game partner. The TPJ has been described as part of the 'mentalizing system' (Fletcher, 1995; Ruby, 2004; Gobbini et al., 2007; Van Overwalle, 2009) and trust game research found this area activated when the investor's decision was revealed to the trustee (Van den Bos et al., 2011b). We found increased activation within the TPJ during investment decisions. Possibly, this area plays a role in mentalizing during decisions about how much to trust that are made while predicting the game partner's behaviour. The current data also showed a specific age-related activation during cooperation in the right posterior cingulate/precuneus region which also has been described as part of the mentalizing network (Van Overwalle, 2009; Wolf et al., 2010). Our results show that different areas of the mentalizing network are differentially activated during social interactions. The TPJ is involved regardless of the nature of the social interaction and its activation increases with age. However, the posterior cingulate/precuneus region shows specific activation in response to trustworthiness. In contrast to Van den Bos et al. (2011b), current age-related changes in the mentalizing network were predominantly located in the posterior and not the medial prefrontal brain areas. This discrepancy could be explained by the differential nature of the tasks, the specified contrasts or age differences in the samples (i.e. 12–22 vs 13–49 years).
In order to make sensible decisions to trust, humans need to learn the associations between their behaviour and the feedback that they receive from others in response. For example, if a certain person answers trust with betrayal, we are less likely to trust that person a second time. However, if the person proves to be trustworthy, we are more likely to trust this person in future (King-Casas et al., 2005). Thus, reward learning shapes behaviour towards optimal decision making. In this study, age was associated with higher levels of trust at the onset of anonymous social interactions. This suggests an increased expectation of benevolence of others and offers a suitable explanation for why feedback learning during cooperation becomes less important as individuals get older. Where cooperation is anticipated, experiencing cooperative behavioural feedback from the trustee matches the prediction. Therefore, learning of new behaviour–outcome associations is unnecessary. fMRI research showed that the orbitofrontal cortex is important for feedback learning and subsequent goal-directed behaviour. It is thought to signal reward value during decision making and is sensitive to value changes, i.e. becomes less responsive when stimuli are less novel or expected (Rolls, 2000; O'Doherty et al., 2001; Rolls and Grabenhorst, 2008). The caudate also plays a role in the anticipation and reception of rewards and has been found to decrease signalling when certain predictions match an outcome (Delgado et al., 2005; Schiffer and Schubotz, 2011). Accordingly, we expected that with increasing age, brain areas that are important for reward learning (e.g. ventromedial prefrontal and orbitofrontal cortex and the striatum) would be recruited to a lesser extent. During cooperative interactions, we found age-related decreases in orbitofrontal cortex and caudate nucleus activation. We did not find the reverse pattern during interactions with an unfair partner. This is in line with previous trust game research of Phan et al. (2010) who found that only positive reciprocity by the partner engages the ventral striatum and orbitofrontal cortex. Also, King-Casas et al. (2005) found the caudate specifically active during cooperation.
As indicated by increased levels of trust, the anticipation of benevolent behaviour by others increases with age and responding to deviations of trust requires a stronger correction in response to an unfair interaction partner. Hence, the interaction is more demanding for older individuals, as it requires them to adapt their beliefs and behaviour. Younger people need to make fewer adjustments to their investments because of their more distrusting mindset. In line with this reasoning, we found age to be associated with a stronger decrease in trust during the interactions with an unfair partner. During these interactions, expectations of cooperation will have resulted in cognitive conflict in the face of the actual returns. The need to reduce the levels of trust was not reflected in increases in activation in brain areas associated with reward learning, but activation in the dorsal anterior cingulate was modulated by age and the nature of the game partner. This area of the anterior cingulate is known to play a role in conflict monitoring (Botvinick et al., 1999, 2004; Bush et al., 2000; Greene et al., 2004; Ridderinkhof et al., 2004) and there is evidence that it is important for behavioural adjustments (Kerns, 2004; Magno et al., 2006). With age, participants increasingly recruited the anterior cingulate during unfair when compared with cooperative interactions. This may reflect the conflict between expectations of cooperation and experienced social feedback and shows a possible explanation of how feedback shapes future decisions towards more optimal levels of trust (Chang and Sanfey, 2013).
Finally, we found age-related changes in brain areas that were not part of our hypotheses, but that previously have been associated with social cognitive processes. During cooperative interactions, age was also associated with activation in the precentral and bilateral middle frontal gyri. These areas have been described as parts of the 'mirror system' (Van Overwalle and Baetens, 2009), a network which is thought to enable humans to understand the goals of observed (physical) actions of others in an intuitive way by internal simulation. Some evidence indicates that this brain network also engages in more abstract forms of social cognition, such as mentalizing (Vogeley, 2001; Wolf et al., 2010) and empathy (de Greck et al., 2012). Previous trust game research found this area to be activated during decisions to trust (Delgado et al., 2005) and age-related increases of activation have been found during social perception (Beadle et al., 2012). During unfair interactions, the middle cingulate gyrus was increasingly active with age. This area has been suggested to function as a relay node between negative emotions and motor action and has been reported to be strongly active during decisions to trust in earlier trust game research (King-Casas et al., 2005). With age, activation in the dorsomedial prefrontal cortex decreased during cooperative interactions. Earlier research showed this area to be involved in the management of uncertainty in decision making, whereby more uncertainty was associated with increased activation (Volz et al., 2005).
Limitations
The current results have to be interpreted in the light of some limitations. First, it might be possible that the differences in trust that we see with increasing age could be due to age differences in risk aversion.While a higher risk aversion may influence the degree of trust, there is extensive literature, which shows that age is typically positively associated with increased risk aversion (Steinberg and Sheffield Morris, 2001; Steinberg, 2004; Deakin et al., 2004; Burnett et al., 2010; Paulsen et al., 2011, 2012). This implies that teenagers should invest more money if risk aversion would be an important motive in the trust game. Furthermore, the concern over whether behaviour in the trust game actually measures trust or risk attitudes has frequently been raised. Eckel and Wilson (2004) provide a comprehensive analysis of the way behaviour in two-person sequential trust games correlates with a variety of behavioural and survey-based risk measures (Eckel and Wilson, 2004). They did not find evidence that any of their risk measures predicts the decision to trust. Also, Houser et al. (2006) found that in a risk game with a computer as counterpart, the probability of investing more was significantly higher for risk seeking subjects than for subjects in the risk averse group but risk aversion did not predict investments towards a human counterpart (Houser et al., 2006). Second, our manipulation check showed that six individuals of the adolescent (n = 25) and one of the adult group (n = 20) did have doubts that they were playing with a real human. The fact that fewer adolescents believed that they were playing a human may have reduced their mentalizing operations. Yet, the current behavioural results show increases in trust with age that are in line with those of other research and do not support such concerns (Sutter and Kocher, 2007). Third, it is important to note that age-related effects may also be caused by differences in neurovasculature (Harris et al., 2011). Yet, research by Kang et al. (2003) showed that in voxelwise group comparison of images of visual and motor cortex regions, only minimal differences were found between children of 7 and 8 years and adults. The small differences in time courses and locations of activation foci between child and adult brains do support the feasibility of direct statistical comparison of these groups within a common space. Fourth, it should be noted that other cognitive factors that may offer an alternative explanation for age-related changes in trust game performance to increases in mentalizing, such as executive functioning or IQ, have not been included as confounders in this study. While evidence from other research of Van den Bos et al. (2011b) supports that non-social cognitive factors such as intelligence have a modest influence on social decision making, it would be valuable if future studies would include additional measures of theory of mind and other cognitive functions to validate the interpretation in terms of mentalizing. Fifth, subjects knew that they were taking part in a study with participants between 13 and 18 and 19 and 49, respectively. It is therefore likely that the current effects pertain to trusting behaviour regarding individuals within these age ranges. Trusting behaviour towards other age groups may differ from the current results. It would therefore be interesting for future research to systematically examine age effects with respect to the age of the trustee. Sixth, in the current paradigm, we did not directly control for aspects of monetary reward in the trust game. However, as a means to control that age-related differences in brain activation are not merely due to differences in the investments that were made, we did analyse the association between the invested amounts and brain activation (Supplementary Table S2). Age-related increases and decreases in brain activation in certain areas did not overlap with the foci that were associated with higher and lower investments. The associations between brain activation and age do not seem to be caused by differences in investments. Finally, fMRI allows for the investigation of the role of certain brain regions in certain cognitive functions and caution is required when cognitive processes are inferred from activation in specific brain regions (Poldrack, 2006). Within the brain, it is unlikely that a particular region is activated solely by one cognitive process. The current interpretations are in line with the growing literature about the social brain systems and should be regarded as a guide for future inquiries rather than direct explanations of certain findings.
Conclusion
Our findings render preliminary support to our working hypothesis of improved mentalizing as an underlying mechanism of age-related behavioural preferences of trust and cooperation. Initial trust increased with age. During interactions with a cooperative partner, older people continued to make higher investments. However, when playing with an unfair game partner, older people quickly reduced their levels of trust to those of younger ones. The neuroimaging data showed age to be associated with increased recruitment of brain areas that seem to be important for mentalizing. In line with a stronger preference for cooperation, age was associated with decreased activation in areas involved in reward learning in response to cooperative behaviour by the partner and increased activation in areas associated with cognitive control in response to an unfair game partner.