Limits...
Effects of direct social experience on trust decisions and neural reward circuitry.

Fareri DS, Chang LJ, Delgado MR - Front Neurosci (2012)

Bottom Line: Participants' trust decisions were influenced by their prior experience in the ball-tossing game, investing less often with the bad partner compared to the good and neutral.Reinforcement learning models revealed that participants were more sensitive to updating their beliefs about good and bad partners when experiencing outcomes consistent with initial experience.These results suggest that initial impressions formed from direct social experience can be continually shaped by consistent information through reward learning mechanisms.

View Article: PubMed Central - PubMed

Affiliation: Department of Psychology, Rutgers University Newark, NJ, USA.

ABSTRACT
The human striatum is integral for reward-processing and supports learning by linking experienced outcomes with prior expectations. Recent endeavors implicate the striatum in processing outcomes of social interactions, such as social approval/rejection, as well as in learning reputations of others. Interestingly, social impressions often influence our behavior with others during interactions. Information about an interaction partner's moral character acquired from biographical information hinders updating of expectations after interactions via top down modulation of reward circuitry. An outstanding question is whether initial impressions formed through experience similarly modulate the ability to update social impressions at the behavioral and neural level. We investigated the role of experienced social information on trust behavior and reward-related BOLD activity. Participants played a computerized ball-tossing game with three fictional partners manipulated to be perceived as good, bad, or neutral. Participants then played an iterated trust game as investors with these same partners while undergoing fMRI. Unbeknownst to participants, partner behavior in the trust game was random and unrelated to their ball-tossing behavior. Participants' trust decisions were influenced by their prior experience in the ball-tossing game, investing less often with the bad partner compared to the good and neutral. Reinforcement learning models revealed that participants were more sensitive to updating their beliefs about good and bad partners when experiencing outcomes consistent with initial experience. Increased striatal and anterior cingulate BOLD activity for positive versus negative trust game outcomes emerged, which further correlated with model-derived prediction error learning signals. These results suggest that initial impressions formed from direct social experience can be continually shaped by consistent information through reward learning mechanisms.

No MeSH data available.


Modulation of prediction error BOLD responses by subject specific learning rates. Subject specific learning rates for losses (αloss) significantly modulated BOLD responses at the time of experienced prediction errors during the trust game in regions including (A) cingulate cortex, (B) inferior frontal/precentral gyrus, insula, and superior temporal gyrus. Scatter plots for illustrative purposes depict the relationship between the mean parameter estimates of BOLD activation that parametrically varies with model estimated prediction error values and model estimated learning rates for losses in the (C) cingulate and (D) insula.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC3472892&req=5

Figure 5: Modulation of prediction error BOLD responses by subject specific learning rates. Subject specific learning rates for losses (αloss) significantly modulated BOLD responses at the time of experienced prediction errors during the trust game in regions including (A) cingulate cortex, (B) inferior frontal/precentral gyrus, insula, and superior temporal gyrus. Scatter plots for illustrative purposes depict the relationship between the mean parameter estimates of BOLD activation that parametrically varies with model estimated prediction error values and model estimated learning rates for losses in the (C) cingulate and (D) insula.

Mentions: To further characterize how participants updated their beliefs, we conducted an exploratory analysis in which we searched for areas of the brain responsible for adapting beliefs based on PE in the context of gains or losses. Specifically, we looked for neural regions which were positively associated with the parametric PE regressor and were further moderated by individual variability in participants’ learning rates as estimated by the model fitting procedure. Larger learning rate values indicate a greater weighting of PEs (i.e., when outcomes differ from expectations) when participants are updating their beliefs. In other words, PEs will lead to larger changes in trial-to-trial beliefs when learning rates are high and will result in lower changes in learning when learning rates are low. Thus, any regions that emerged in this analysis would likely be computationally involved in dynamically adapting beliefs based on positive or negative outcomes. Overall, participants demonstrated a behavioral bias toward updating behavior more readily from gains than losses. We conducted whole brain correlations between participant specific learning rates for gains and losses and BOLD activation corresponding to the PE regressor. We note that this regressor contained both positive and negative PE values. Significant positive correlations were observed between PE related BOLD responses and the αloss learning parameter in regions denoted in Table 6 and illustrated in Figure 5, including two portions of inferior frontal gyrus (BA47: x, y, z = −37, 34, 0; BA45: x, y, z = −49, 19, 9), as well the insula (x, y, z = −40, −2, 12) superior temporal gyrus (BA22/42: x, y, z = −46, −26, 9) and cingulate cortex (BA24: x, y, z = 8, −8, 42).


Effects of direct social experience on trust decisions and neural reward circuitry.

Fareri DS, Chang LJ, Delgado MR - Front Neurosci (2012)

Modulation of prediction error BOLD responses by subject specific learning rates. Subject specific learning rates for losses (αloss) significantly modulated BOLD responses at the time of experienced prediction errors during the trust game in regions including (A) cingulate cortex, (B) inferior frontal/precentral gyrus, insula, and superior temporal gyrus. Scatter plots for illustrative purposes depict the relationship between the mean parameter estimates of BOLD activation that parametrically varies with model estimated prediction error values and model estimated learning rates for losses in the (C) cingulate and (D) insula.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC3472892&req=5

Figure 5: Modulation of prediction error BOLD responses by subject specific learning rates. Subject specific learning rates for losses (αloss) significantly modulated BOLD responses at the time of experienced prediction errors during the trust game in regions including (A) cingulate cortex, (B) inferior frontal/precentral gyrus, insula, and superior temporal gyrus. Scatter plots for illustrative purposes depict the relationship between the mean parameter estimates of BOLD activation that parametrically varies with model estimated prediction error values and model estimated learning rates for losses in the (C) cingulate and (D) insula.
Mentions: To further characterize how participants updated their beliefs, we conducted an exploratory analysis in which we searched for areas of the brain responsible for adapting beliefs based on PE in the context of gains or losses. Specifically, we looked for neural regions which were positively associated with the parametric PE regressor and were further moderated by individual variability in participants’ learning rates as estimated by the model fitting procedure. Larger learning rate values indicate a greater weighting of PEs (i.e., when outcomes differ from expectations) when participants are updating their beliefs. In other words, PEs will lead to larger changes in trial-to-trial beliefs when learning rates are high and will result in lower changes in learning when learning rates are low. Thus, any regions that emerged in this analysis would likely be computationally involved in dynamically adapting beliefs based on positive or negative outcomes. Overall, participants demonstrated a behavioral bias toward updating behavior more readily from gains than losses. We conducted whole brain correlations between participant specific learning rates for gains and losses and BOLD activation corresponding to the PE regressor. We note that this regressor contained both positive and negative PE values. Significant positive correlations were observed between PE related BOLD responses and the αloss learning parameter in regions denoted in Table 6 and illustrated in Figure 5, including two portions of inferior frontal gyrus (BA47: x, y, z = −37, 34, 0; BA45: x, y, z = −49, 19, 9), as well the insula (x, y, z = −40, −2, 12) superior temporal gyrus (BA22/42: x, y, z = −46, −26, 9) and cingulate cortex (BA24: x, y, z = 8, −8, 42).

Bottom Line: Participants' trust decisions were influenced by their prior experience in the ball-tossing game, investing less often with the bad partner compared to the good and neutral.Reinforcement learning models revealed that participants were more sensitive to updating their beliefs about good and bad partners when experiencing outcomes consistent with initial experience.These results suggest that initial impressions formed from direct social experience can be continually shaped by consistent information through reward learning mechanisms.

View Article: PubMed Central - PubMed

Affiliation: Department of Psychology, Rutgers University Newark, NJ, USA.

ABSTRACT
The human striatum is integral for reward-processing and supports learning by linking experienced outcomes with prior expectations. Recent endeavors implicate the striatum in processing outcomes of social interactions, such as social approval/rejection, as well as in learning reputations of others. Interestingly, social impressions often influence our behavior with others during interactions. Information about an interaction partner's moral character acquired from biographical information hinders updating of expectations after interactions via top down modulation of reward circuitry. An outstanding question is whether initial impressions formed through experience similarly modulate the ability to update social impressions at the behavioral and neural level. We investigated the role of experienced social information on trust behavior and reward-related BOLD activity. Participants played a computerized ball-tossing game with three fictional partners manipulated to be perceived as good, bad, or neutral. Participants then played an iterated trust game as investors with these same partners while undergoing fMRI. Unbeknownst to participants, partner behavior in the trust game was random and unrelated to their ball-tossing behavior. Participants' trust decisions were influenced by their prior experience in the ball-tossing game, investing less often with the bad partner compared to the good and neutral. Reinforcement learning models revealed that participants were more sensitive to updating their beliefs about good and bad partners when experiencing outcomes consistent with initial experience. Increased striatal and anterior cingulate BOLD activity for positive versus negative trust game outcomes emerged, which further correlated with model-derived prediction error learning signals. These results suggest that initial impressions formed from direct social experience can be continually shaped by consistent information through reward learning mechanisms.

No MeSH data available.