A win-win situation: Does familiarity with a social robot modulate feedback monitoring and learning?

Social species rely on the ability to modulate feedback-monitoring in social contexts to adjust one's actions and obtain desired outcomes. When being awarded positive outcomes during a gambling task, feedback-monitoring is attenuated when strangers are rewarded, as less value is assigned to the awarded outcome. This difference in feedback-monitoring can be indexed by an event-related potential (ERP) component known as the Reward Positivity (RewP), whose amplitude is enhanced when receiving positive feedback. While the degree of familiarity influences the RewP, little is known about how the RewP and reinforcement learning are affected when gambling on behalf of familiar versus nonfamiliar agents, such as robots. This question becomes increasingly important given that robots may be used as teachers and/or social companions in the near future, with whom children and adults will interact with for short or long periods of time. In the present study, we examined whether feedback-monitoring when gambling on behalf of oneself compared with a robot is impacted by whether participants have familiarized themselves with the robot before the task. We expected enhanced RewP amplitude for self versus other for those who did not familiarize with the robot and that self-other differences in the RewP would be attenuated for those who familiarized with the robot. Instead, we observed that the RewP was larger when familiarization with the robot occurred, which corresponded to overall worse learning outcomes. We additionally observed an enhanced P3 effect for the high-familiarity condition, which suggests an increased motivation to reward. These findings suggest that familiarization with robots may cause a positive motivational effect, which positively affects RewP amplitudes, but interferes with learning.

Subthreshold error corrections predict adaptive post-error compensations.

Relatively little is known about the relation between subthreshold error corrections and post-error behavioral compensations. The present study utilized lateralized beta power, which has been shown to index response preparation, to examine subthreshold error corrections in a task known to produce response conflict, the Simon task. We found that even when an overt correction is not made, greater activation of the corrective response, indexed by beta suppression ipsilateral to the initial responding hand, predicted post-error speeding, and enhanced post-error accuracy at the single-trial level. This provides support for the notion that response conflict associated with errors can be adaptive, and suggests that subthreshold corrections should be taken into account to fully understand error-monitoring processes. Furthermore, we expand on previous findings that demonstrate that post-error slowing and post-error accuracy can be dissociated, as well as findings that suggest that frontal midline theta oscillations and the error-related negativity (ERN) are dissociable neurocognitive processes.

Contrasting time and frequency domains: ERN and induced theta oscillations differentially predict post-error behavior.

The present study investigated the neural dynamics of error processing in both the time and frequency domains, as well as associated behavioral phenomena, at the single-trial level. We used a technique that enabled us to separately investigate the evoked and induced aspects of the EEG signal (Cohen & Donner, 2013, Journal of Neurophysiology, 110[12], 2752-2763). We found that at the single-trial level, while the (evoked) error-related negativity (ERN) predicted only post-error slowing (PES)-and only when errors occurred on incongruent trials-induced frontal midline theta power served as a robust predictor of both PES and post-error accuracy (PEA) regardless of stimulus congruency. Mediation models of both electrophysiological indices demonstrated that although the relationship between theta and PEA was mediated by PES, there was not a relationship between the ERN and PEA. Our data suggest that although the ERN and frontal midline theta index functionally related underlying cognitive processes, they are not simply the same process manifested in different domains. In addition, our findings are consistent with the adaptive theory of post-error slowing, as PES was positively associated with post-error accuracy at the single-trial level. More generally, our study provides additional support for the inclusion of a time-frequency approach to better understand the role of medial frontal cortex in action monitoring.

Learning From the Slips of Others: Neural Correlates of Trust in Automated Agents.

With the rise of increasingly complex artificial intelligence (AI), there is a need to design new methods to monitor AI in a transparent, human-aware manner. Decades of research have demonstrated that people, who are not aware of the exact performance levels of automated algorithms, often experience a mismatch in expectations. Consequently, they will often provide either too little or too much trust in an algorithm. Detecting such a mismatch in expectations, or trust calibration, remains a fundamental challenge in research investigating the use of automation. Due to the context-dependent nature of trust, universal measures of trust have not been established. Trust is a difficult construct to investigate because even the act of reflecting on how much a person trusts a certain agent can change the perception of that agent. We hypothesized that electroencephalograms (EEGs) would be able to provide such a universal index of trust without the need of self-report. In this work, EEGs were recorded for 21 participants (mean age = 22.1; 13 females) while they observed a series of algorithms perform a modified version of a flanker task. Each algorithm's degree of credibility and reliability were manipulated. We hypothesized that neural markers of action monitoring, such as the observational error-related negativity (oERN) and observational error positivity (oPe), are potential candidates for monitoring computer algorithm performance. Our findings demonstrate that (1) it is possible to reliably elicit both the oERN and oPe while participants monitored these computer algorithms, (2) the oPe, as opposed to the oERN, significantly distinguished between high and low reliability algorithms, and (3) the oPe significantly correlated with subjective measures of trust. This work provides the first evidence for the utility of neural correlates of error monitoring for examining trust in computer algorithms.

Speeded response errors and the error-related negativity modulate early sensory processing.

Empirical research demonstrates that when the time following error commission is constrained, subsequent sensory processing can be impaired (Buzzell et al., 2017). This reduction in sensory processing is presumably due to a bottleneck for cognitive resources produced by an overlap between error processing and subsequent stimulus processing. This finding suggests that the system dedicated to improving task performance can actually sometimes be the source of performance failures. Although this finding established that data-limited errors lead to a reduction in sensory processing at short response stimulus intervals (RSIs), it remains unclear if the relationship between error processing and subsequent sensory processing can be modulated by speeded-response errors. In the present study, event-related potentials and behavioral measures were recorded while participants performed a modified version of a Simon task, in which RSI duration was varied. We found that sensory processing, indexed by the P1 component, was reduced following errors at short (200-533 ms), but not long (866-1200 ms), RSIs. Moreover, the magnitude of error processing differentially influenced subsequent sensory processing as a function of RSI. However, whereas prior work demonstrated that the error positivity (Pe) modulated sensory processing on the subsequent trial, only the error-related negativity (ERN) did so within the Simon task. This suggests that although both data-limited errors and speeded-response errors can impact subsequent sensory processing, different stages of error processing appear to mediate this phenomenon.

Seeing minds in others: Mind perception modulates low-level social-cognitive performance and relates to ventromedial prefrontal structures.

In social interactions, we rely on nonverbal cues like gaze direction to understand the behavior of others. How we react to these cues is affected by whether they are believed to originate from an entity with a mind, capable of having internal states (i.e., mind perception). While prior work has established a set of neural regions linked to social-cognitive processes like mind perception, the degree to which activation within this network relates to performance in subsequent social-cognitive tasks remains unclear. In the current study, participants performed a mind perception task (i.e., judging the likelihood that faces, varying in physical human-likeness, have internal states) while event-related fMRI was collected. Afterwards, participants performed a social attention task outside the scanner, during which they were cued by the gaze of the same faces that they previously judged within the mind perception task. Parametric analyses of the fMRI data revealed that activity within ventromedial prefrontal cortex (vmPFC) was related to both mind ratings inside the scanner and gaze-cueing performance outside the scanner. In addition, other social brain regions were related to gaze-cueing performance, including frontal areas like the left insula, dorsolateral prefrontal cortex, and inferior frontal gyrus, as well as temporal areas like the left temporo-parietal junction and bilateral temporal gyri. The findings suggest that functions subserved by the vmPFC are relevant to both mind perception and social attention, implicating a role of vmPFC in the top-down modulation of low-level social-cognitive processes.

Error-Induced Blindness: Error Detection Leads to Impaired Sensory Processing and Lower Accuracy at Short Response-Stimulus Intervals.

Empirical evidence indicates that detecting one's own mistakes can serve as a signal to improve task performance. However, little work has focused on how task constraints, such as the response-stimulus interval (RSI), influence post-error adjustments. In the present study, event-related potential (ERP) and behavioral measures were used to investigate the time course of error-related processing while humans performed a difficult visual discrimination task. We found that error commission resulted in a marked reduction in both task performance and sensory processing on the following trial when RSIs were short, but that such impairments were not detectable at longer RSIs. Critically, diminished sensory processing at short RSIs, indexed by the stimulus-evoked P1 component, was predicted by an ERP measure of error processing, the Pe component. A control analysis ruled out a general lapse in attention or mind wandering as being predictive of subsequent reductions in sensory processing; instead, the data suggest that error detection causes an attentional bottleneck, which can diminish sensory processing on subsequent trials that occur in short succession. The findings demonstrate that the neural system dedicated to monitoring and improving behavior can, paradoxically, at times be the source of performance failures.SIGNIFICANCE STATEMENT The performance-monitoring system is a network of brain regions dedicated to monitoring behavior to adjust task performance when necessary. Previous research has demonstrated that activation of the performance monitoring system following incorrect decisions serves to improve future task performance. However, the present study provides evidence that, when perceptual decisions must be made rapidly (within approximately half a second of each other), activation of the performance-monitoring system is predictive of impaired task-related attention on the subsequent trial. The data illustrate that the cognitive demands imposed by error processing can interfere with, rather than enhance, task-related attention when subsequent decisions need to be made quickly.