Repeated Trust Game Research Articles

Trust building with artificial intelligence: comparing with human in investment behaviour, emotional arousal and neuro activities

This study investigated on people’s two-stage iterative trust building process (the initial trust stage and the iteration from feedback stage) with AI partners, and made comparisons with human partners in the aspects of investment behavior, emotional arousal and neuro activities. The experiment was designed using repeated trust games and the investment amount was transformed into the initial trust, the feedback from the partner and the final trust. Comparisons of behavior were conducted by regression analysis. The results showed that the initial trust for AI had stronger influence on the final trust than human partners, while the feedback from human during interaction had stronger influence than the feedback from AI. By emotion scales, it was found that the arousal of participants’ positive emotions was more significant when interacting with human than with AI, and the emotional arousal (both positive and negative ones) had stronger correlation with the feedback from human than from AI.. The findings indicated a stronger influence of feedback from human partners on decision making and emotional arousal than from AI partners. The fNIRS results also showed that people’s neuro activities in the PFC were stronger when making investment decisions towards human than AI partners.

The Immunization Effect of Internal Control System Transparency following a Supplier’s Violation of Buyer Trust

ABSTRACT In this study, we show that a supplier’s internal controls (ICs) that lead to either falling short of or to exceeding buyer expectations play an important role in the trust a buyer has in its supplier. In a 12-round repeated trust game, we examine the impact of supplier ICs and the transparency of those ICs to the buyer on buyer trusting behavior across three phases of the buyer-supplier relationship: (1) trust formation, (2) trust violation, and (3) trust repair. We find that, although a supplier’s trust violation reduces buyer trusting behavior, the least amount of damage to trusting behavior occurs for suppliers whose ICs led to supplier actions that fell marginally short of buyer expectations before the violation and when the IC was transparent to the buyer. We refer to this as the IC transparency “immunization effect.” We show that suppliers can benefit from making IC choices known to partners. Data Availability: Contact the authors. JEL Classifications: C91; D91; M41.

Exemplary Self-Discipline, Leniency towards Others: Competitive Contexts Amplify the "Black Sheep Effect" in Restoring Ingroup Trust.

Intergroup interaction, a pivotal aspect of social interaction, encompasses both cooperation and competition. Group identity significantly impacts individual behaviors and decision-making processes. This influence manifests in two contrasting ways when addressing rule-breaking by interaction partners: in-group favoritism, where individuals are more lenient towards infractions committed by in-group members, and the black sheep effect, where in-group members are penalized for their rule-breaking. Although trust is crucial in intergroup interactions, the precise impact of group identity on trust restoration and the potential moderating role of intergroup interaction types remain to be elucidated. This study presents two experiments designed to explore these dynamics. In Study 1, the manipulation of group identity through a point estimation task was utilized to evaluate its impact on intergroup trust restoration via a series of repeated trust games. Study 2 aimed to explore the moderating role of intergroup interaction on intergroup trust restoration by contrasting cooperation and competition situations. The results uncovered a "black sheep effect", where participants demonstrated a greater propensity for trust restoration with out-group members than with in-group members. This effect, however, was only evident in competitive contexts. Conversely, in cooperative contexts, the individual's trust in the in-group and out-group members is effectively repaired. These findings contribute to a deeper comprehension of trust dynamics in intergroup interactions, promoting trust establishment and repair between diverse groups, thereby boosting team collaboration efficiency and mitigating conflicts.

Blocking D2/D3 dopamine receptors in male participants increases volatility of beliefs when learning to trust others

The ability to learn about other people is crucial for human social functioning. Dopamine has been proposed to regulate the precision of beliefs, but direct behavioural evidence of this is lacking. In this study, we investigate how a high dose of the D2/D3 dopamine receptor antagonist sulpiride impacts learning about other people’s prosocial attitudes in a repeated Trust game. Using a Bayesian model of belief updating, we show that in a sample of 76 male participants sulpiride increases the volatility of beliefs, which leads to higher precision weights on prediction errors. This effect is driven by participants with genetically conferred higher dopamine availability (Taq1a polymorphism) and remains even after controlling for working memory performance. Higher precision weights are reflected in higher reciprocal behaviour in the repeated Trust game but not in single-round Trust games. Our data provide evidence that the D2 receptors are pivotal in regulating prediction error-driven belief updating in a social context.

Ethnic discrimination unlearned: experience in the repeated Trust Game reduces trust bias.

Discrimination toward ethnic minorities is a persistent societal problem. One reason behind this is a bias in trust: people tend to trust their ingroup and comparatively distrust outgroups. In this study, we investigated whether and how people change their explicit trust bias with respect to ethnicity based on behavioral interactions with in- and outgroup members in a modified Trust Game. Subjects' initial explicit trust bias disappeared after the game. The change was largest for ingroup members who behaved unfairly, and the reduction of trust bias generalized to a small sample of new in- and outgroup members. Reinforcement learning models showed subjects' learning was best explained by a model with only one learning rate, indicating that subjects learned from trial outcomes and partner types equally during investment. We conclude that subjects can reduce bias through simple learning, in particular by learning that ingroup members can behave unfairly.

Impaired social learning in patients with major depressive disorder revealed by a reinforcement learning model

Background/objectivePatients with major depressive disorder (MDD) have altered learning rates for rewards and losses in non-social learning paradigms. However, it is not well understood whether the ability to learn from social interactions is altered in MDD patients. Using reinforcement learning during the repeated Trust Game (rTG), we investigated how MDD patients learn to trust newly-met partners in MDD patients. MethodSixty-eight MDD patients and fifty-four controls each played as ‘investor’ and interacted with ten different partners. We manipulated both the level of trustworthiness by varying the chance of reciprocity (10, 30, 50, 70 and 90%) and reputation disclosure, where partners’ reputation was either pre-disclosed or hidden. ResultsOur reinforcement learning model revealed that MDD patients had significantly higher learning rates for losses than the controls in both the reputation disclosure and non-disclosure condition. The difference was larger when reputation was not disclosed than disclosed. We observed no difference in learning rates for gains in either condition. ConclusionsOur findings highlight that abnormal learning for losses underlies the social learning process in MDD patients. This abnormality is higher when situational unpredictability is high versus low. Our findings provide novel insights into social rehabilitation of MDD.

Superior bias in trust-related decisions

Trust is conditional. Many studies have revealed its relative conditions in different situations, but when social status is characteristic of the person who receives trust (the “trustee”), our knowledge of how social status affects trust still remains limited. In this study, we used the trust game in order to: (1) characterize the effect of trustees’ social hierarchy on trust-related decisions in different trustworthiness situations and (2) explore the underlying computational process regarding the impact that social status has on trust-related decisions by using the computational modeling approach to integrate social status into trust-related situations. In Experiment 1, using a one-shot trust game with no feedback of information about reciprocity, we found that compared with inferiors, superiors gained more trust-related behaviors (investments) in spite of the fact that they were not rated as having higher trustworthiness. Then, in Experiment 2, when we controlled the trustworthiness of different social status partners by providing the same neutral reciprocity rate (50%) in a repeated trust game, the high-status partner gained more trust than the low-status partner. This superior bias extended to Experiment 3a and 3b, in which we set different levels of trustworthiness to match the different social statuses of partners. With respect to modeling results, we found that higher status holds an additional social value independent of trust profit, resulting in superior bias. Ultimately, this study has shed light on the superior bias that commonly leads people to grant high-status individuals goodwill in social interactions.

Self-reported & revealed trust: Experimental evidence

I study the relationship between self-declared trust attitudes – using a well-recognised and established personality questionnaire — and trust choices in an induced infinitely repeated trust game. I find that self-reported trust measures are significantly related with trust choices as long as trust is part of equilibrium strategies. I find that questions regarding others’ intentions is a missing component in previous work that studies self-reports of trust. An important aspect of the design is that first movers are not privy to the choices made by their partners. This design feature, coupled with an uncertainty element introduced in determining the first mover’s final payoff, allows me to analyse how first movers react to bad outcomes. Trusting individuals are more likely to give the benefit of doubt to others. Analysis of the incentivised subjective beliefs reveals that the effect of personality traits on trust choices is not through the formation of beliefs.

Integration of social status and trust through interpersonal brain synchronization

Trust can be a dynamic social process, during which the social identity of the interacting agents (e.g., an investor and a trustee) can bias trust outcomes. Here, we investigated how social status modulates trust and the neural mechanisms underlying this process. An investor and a trustee performed a 10-round repeated trust game while their brain activity was being simultaneously recorded using functional near-infrared spectroscopy. The social status (either high or low) of both investors and trustees was manipulated via a math competition task. The behavioral results showed that in the initial round, individuals invested more in low-status partners. However, the investment ratio increased faster as the number of rounds increased during trust interaction when individuals were paired with a high-status partner. This increasing trend was particularly prominent in the low (investor)-high (trustee) status group. Moreover, the low-high group showed increased investor-trustee brain synchronization in the right temporoparietal junction as the number of rounds increased, while brain activation in the right dorsolateral prefrontal cortex of the investor decreased as the number of rounds increased. Both interpersonal brain synchronization and brain activation predicted investment performance at the early stage; furthermore, two-brain data provided earlier predictions than did single-brain data. These effects were detectable in the investment phase in the low-high group only; no comparable effects were observed in the repayment phase or other groups. Overall, this study demonstrated a multi-brain mechanism for the integration of social status and trust.

Basolateral and central amygdala orchestrate how we learn whom to trust

Cooperation and mutual trust are essential in our society, yet not everybody is trustworthy. In this fMRI study, 62 healthy volunteers performed a repeated trust game, placing trust in a trustworthy or an untrustworthy player. We found that the central amygdala was active during trust behavior planning while the basolateral amygdala was active during outcome evaluation. When planning the trust behavior, central and basolateral amygdala activation was stronger for the untrustworthy player compared to the trustworthy player but only in participants who actually learned to differentiate the trustworthiness of the players. Independent of learning success, nucleus accumbens encoded whether trust was reciprocated. This suggests that learning whom to trust is not related to reward processing in the nucleus accumbens, but rather to engagement of the amygdala. Our study overcomes major empirical gaps between animal models and human neuroimaging and shows how different subnuclei of the amygdala and connected areas orchestrate learning to form different subjective trustworthiness beliefs about others and guide trust choice behavior.

Gender differences in social interactions

We study how the random assignment of new students to introductory-week groups shapes subsequent friendship networks. Both women and men report being much more likely to be friends with same-gender students with whom they were (randomly) assigned in a group during their first week on campus, and the effect is much stronger for women. When students from the same cohort play a repeated trust game in the experimental laboratory, their behavior helps explain what we observed in the field. Women display more stability and less flexibility than men in their interactions with individuals with whom they had previously played. This difference is enough to generate homophily in the observational data even though subjects show no intrinsic preference for same-gender interaction.

Interpersonal distance adjustments after interactions with a generous and selfish trustee during a repeated trust game

How physically close to an individual we choose to stand—our interpersonal distance preference—is thought to closely reflect the trust placed in the individual. However, the exact relationship between interpersonal distance and trust is not well understood and the causal effects of trust on interpersonal distance have not yet been demonstrated and quantified experimentally. Here, we induced trust (vs distrust) in an interaction partner through generous (vs selfish) monetary transactions in a repeated trust game. Subjects' investments in, experienced trust in, and expected back-transfers from the two interaction partners indicated that they learned to distrust more the selfish (vs generous) interaction partner. Further, across two experiments, using a pre-post design in which interpersonal distance preferences were measured before and after the trust game, preferred reflective and reflexive interpersonal distance from the selfish interaction partner increased, whereas interactions with the generous partner did not change these preferences. Importantly, the post-game distance was associated with subjects' experienced trust in the selfish interaction partner, and not their monetary decisions made during the game. These findings are a starting point for our understanding of how interactions with untrusted others can be successful. We speculate that an increased interpersonal distance might be accompanied by an increased psychological distance (e.g. through impersonal communication), which might help prevent a break-down of an interaction with a distrusted partner.

Fairness norm violations in anti-social psychopathic offenders in a repeated trust game

Psychopathic offenders have a high propensity to violate social norms, as indicated for instance by their widespread lying and cheating behaviour. The reasons for their norm violations are not well understood, though, as they are able to recognise norms in a given situation and also punish norm violators. In this study, we investigated whether psychopathic offenders would violate fairness norms during a repeated trust game because of increased profit-maximising concerns. We measured back-transfer decisions in the repeated trust game, and affective arousal by means of skin conductance responses (SCR) in violent offenders with varying degrees of psychopathy, and non-offenders with low-trait psychopathy. Psychopathy in offenders was measured with the Psychopathy Checklist–Revised (PCL-R). In the task, a participant and an interaction partner entrusted each other money for multiple rounds with the goal to earn as much money as possible. Fairness norm violations were positively associated with Factor 2 scores (the lifestyle/anti-social psychopathy subscale) of the PCL-R, but this was not accompanied by clear profit-maximising behaviour. In addition, anticipatory arousal to self-advantageous decisions was higher in all offenders, independent of their degree of psychopathy, compared with non-offenders. The results of our study widen our understanding of social decision-making in psychopathy. They also suggest treatment possibilities in offenders scoring high on Factor 2, targeting empathic concern and related prosocial intentions to overcome norm-violating behaviour.

The Human Basolateral Amygdala Is Indispensable for Social Experiential Learning.

Trust and betrayal are central to our social world, and adaptive responses to generous and selfish behavior are crucial to our economic and social well-being [1]. We learn about others' trustworthiness through trial and error during repeated interactions [2]. By reinforcing and suppressing behavior during positive and negative interactions with conspecifics, rodent research has established a crucial role for the basolateral amygdala (BLA) in social experiential learning [3, 4]. The human BLA has undergone a reorganization with massive expansion relative to other amygdala nuclei [5], and there is no translational research on its role in experiential learning. The human amygdala is traditionally researched as a single structure [6], neglecting the sub-nuclei's structural und functional differences [7], which might explain inconsistent findings in research on social interactions [8, 9]. Here, we study whether the human BLA is necessary for social andnon-social experiential learning by testing a group of five humans with selective bilateral damage to the BLA. We compared their learning behavior in a repeated trust game, and a non-social control task, to healthy, matched controls. Crucially, BLA-damaged subjects, unlike control subjects, completely failed to adapt their investments when interacting with a trustworthy and an untrustworthy partner. In the non-social task, BLA-damaged subjects learned from positive outcomes but differed from the controls by not learning from negative outcomes. Our data extend findings in rodent research by showing that the human BLA is essential for social experiential learning and provide confirmatory evidence of divergent mechanisms for differentially valenced outcomes in non-social learning.

On Trust and Disgust: Evidence From Face Reading and Virtual Reality

We report three studies exploring the relationship between disgust and trust. Study 1a measured emotions using face-reading technology while participants played a repeated trust game. We observed a negative correlation between trust and disgust. Study 1b employed self-reports along with the face reader. The self-report procedure adversely affected participants’ emotional state and eliminated the correlation between trust and other emotions. Study 2 induced incidental disgust or sadness using virtual reality and manipulated participants’ awareness of the source of their emotions. Disgusted participants judged others as less trustworthy and sent less in a trust game than sad or control participants. An interaction indicated that awareness of the source of emotions eliminated the effect. Our data are consistent with the association between disgust and harsher moral judgments and suggest that disgust is antithetical to the building of trust. However, the association disappears if individuals are aware that their disgust is unrelated to the setting.

Belief-dependent preferences and reputation: Experimental analysis of a repeated trust game

We study in a theoretical and experimental setting the interaction between belief-dependent preferences and reputation building in a finitely repeated trust game. We focus mainly on the effect of guilt aversion. In a simple two-type model, we analyze the effect of reputation building in the presence of guilt-averse trustees and derive behavioral predictions. We test these predictions in a laboratory experiment where we elicit information on trustees’ belief-dependent preferences and disclose it to the paired trustor before the repeated game.

Investments in and returns on network embeddedness: An experiment with trust games

Trust problems are ubiquitous in social and economic exchange. They are known to be mitigated if exchange partners are embedded in social structures that disseminate information on past behavior. If such “network embeddedness” makes exchanges possible that would not be possible otherwise, it is also expected that actors are willing to exert effort to establish embeddedness. Theory suggests that the degree to which network embeddedness facilitates trust depends on the size of the trust problem, and there are reasons to expect that embeddedness facilitates trust more strongly if it is established endogenously rather than imposed exogenously. We tested these predictions in a laboratory experiment in which 342 participants played repeated trust games with exogenous or endogenous embeddedness under varying sizes of the trust problem. The results confirm that embeddedness promotes trustfulness and trustworthiness. The results also show that endogenously chosen embeddedness promotes trustfulness more strongly than exogenously imposed embeddedness. However, we find no systematic variation in investments in embeddedness or effects of embeddedness in the size of the trust problem.

Availability of better data on communication networks can undermine community enforcement

I consider a repeated trust game played between a group of insiders, arranged in a network representing their lines of communication, and a single outsider. Insiders follow a local punishment rule, shunning the outsider if it has cheated them or a neighbor. The object of interest is that the outsider may know either summary statistics about the nature of the network, or know its precise non-anonymous structure. For the outsider to have knowledge of the precise structure may, depending on the shape of the network, increase or decrease the volume of honest interaction that can be sustained. In extreme cases, a small `local' vulnerability to outside exploitation can result in a total breakdown of the chance for mutually beneficial trade across the whole network. Strategic ignorance and obfuscation of network structure may therefore be valuable to both sides, complicating the problem for a network operator seeking to monetize data on the network graph. I discuss recent decisions by Twitter and Facebook in this framework.

Trust in Cohesive Communities

This paper studies which social networks maximize trust and welfare when agreements are implicitly enforced. We study a repeated trust game in which trading opportunities arise exogenously and a social network determines the information each player has. We show that cohesive communities, modeled as social networks of complete components, emerge as the optimal community design. Cohesive communities generate some degree of common knowledge of transpired play that allows players to coordinate their punishments and, as a result, yield relatively high equilibrium payoffs. We also show that when news swiftly travel through the network, Pareto efficient networks are minimally connected: the removal of any link isolates some community members. Our results then clarify a sociological debate on the merits of different social structures to foster cooperation and trust.

Trusting Behavior Following a Violation of Trust: The Immunization Effect of Control System Transparency

Customer-supplier relationships depend in a crucial way on the trust that the customer has in the supplier. We propose that two supplier internal control (IC) design choices – type of IC and transparency of the IC to the customer – interact to affect a supplier’s ability to build and maintain customer trust. In a 12-round 2x2 laboratory experiment employing a repeated trust game, we examine the impact of supplier IC type and IC transparency on customer trust and trusting behavior across three phases of the customer-supplier relationship: (1) a trust formation phase, (2) a trust violation phase, and (3) a post-violation trust repair phase. We manipulate supplier IC type based on whether the IC results in supplier service that exceeds customer expectations (e.g., customer satisfaction performance incentives) or whether the IC results in service that falls short of customer expectations (e.g., budgetary controls). We manipulate IC transparency as high or low. We find that supplier ICs resulting in service that exceeds customer expectations during the trust formation phase of the relationship engender greater customer trusting behavior than ICs that result in service that falls short of customer expectations. However, we also find that the difference in customer trusting behavior between these two types of supplier ICs is smaller when IC transparency is high as compared to when transparency is low. Further, while a significant trust violation damages customer trusting behavior regardless of the supplier IC employed, we find the least amount of damage to trusting behavior from a trust violation when the customer is exposed to prior repeated IC-induced service disappointments and the presence of the IC was transparent. We refer to this as the IC transparency “immunization effect.” This difference in trusting behavior between IC types persists through the post-violation trust repair phase. This study contributes to prior research by examining the effects of differences in supplier ICs on customer trusting behavior throughout the trust environment phases of the supplier-customer relationship.