Natalie C. Ebner

Successful social interactions depend on regulating approach-avoidance tendencies as well as accurately understanding those motivations in others. Direct gaze signals approach intentions, whereas averted gaze signals avoidance intentions. Older relative to younger adults display a stronger direct gaze bias (i.e., are more likely to perceive eye gaze as direct vs. averted). This bias in motivational cue perception likely impacts social trait attributions. This study, for the first time, leveraged a lifespan sample across a wide age range (18 – 88 years; N = 307) to investigate adult lifespan differences in gaze perception using a novel set of gaze-modified face stimuli with varying expressions and tested whether perception of gaze aversion affected attributions of facial trustworthiness. Participants were recruited and the study conducted online via Prolific. Faces with more averted gaze were perceived as more direct-looking with age. This bias was relatively weaker for fearful faces compared to all other facial expressions. Further, greater perception of direct gaze was associated with greater facial trustworthiness attributions, and especially with age. Combined, these findings confirm a direct gaze bias in aging and support its impact on social trait attributions. These adult lifespan differences in approach-avoidance motivational cue perception may reflect shifts in sensitivity, capacity, and/or motivation and are discussed as a possible mechanism in greater trust and susceptibility to deception with age. This work also adds substantially to the widely used FACES Lifespan Database corpus by making the newly gaze-modified faces, and their norm ratings, publicly available for research on motivational and cognitive aging.

Conversational generative AI has rapidly reached majority adoption among U.S. adolescents. These systems now provide advice, emotional support, academic assistance, simulated social interaction, and opportunities for identity exploration at a critical phase in development when mental health needs are rising and capacities for self-regulation, autonomy, and social competence are still developing. Scholars and clinicians have raised urgent questions about what these systems do to adolescents, and early findings point to both benefits and risks. In this article, we ask what adolescents do with AI and introduce the concept of adolescent agency over conversational AI use: the degree to which adolescents purposefully direct, critically evaluate, and set limits on their engagement with conversational AI. Existing constructs such as AI literacy and AI self-efficacy describe what young people know or believe about AI and the risks associated with its use, but they do not capture how adolescents exercise agency across areas of their lives. We identify nine processes that arise at the intersection of conversational AI’s features and key developmental tasks during the period of adolescence, which we label AI-ERA: autonomy, identity, emotion regulation, relationships, and achievement. We distinguish this agency from existing constructs, including agentic engagement with AI and AI overreliance, and then propose how it is shaped by adolescents’ developing capacities, adult scaffolding, and peer influence. We conclude with implications for families, schools, clinicians, and researchers, as well as considerations regarding the design, regulation, and ethics of AI systems.

Initial visual impressions guide judgment and decision-making in social contexts, and social cues, such as facial trustworthiness, may bias choices by influencing individuals to prioritize appearance over actual behavioral outcomes. Key challenges for research in this field include clarifying the extent to which decision-making and learning under uncertainty are affected (1) by facial cues of trustworthiness vs. untrustworthiness, (2) by the congruence vs. incongruence of face-based visual cues and behavioral outcomes, and (3) in a social relative to a non-social decision context. This study used the Social Iowa Gambling Task (S-IGT), an adaptation of the Iowa Gambling Task (IGT). Extending traditional decision-making paradigms by systematically integrating several novel aspects, the S-IGT features: 1) the experimental manipulation of facial cues perceived as trustworthy vs. untrustworthy and 2) the incorporation of decision options paired with congruent (CS-IGT) vs. incongruent (IS-IGT) social cues; and was used here in combination with 3) a non-social/neutral control condition (the standard IGT). Across 558 participants, we found that initial performance in the CS-IGT was higher than in the other two task conditions, yet participants’ learning progress diminished over time in the CS-IGT. These findings suggest that while social cues of facial trustworthiness may support early decision-making, they also may encourage reliance on initial impressions over feedback-driven exploration, thus crucially influencing choice and learning trajectories under uncertainty in social contexts. The results further highlight the complex role of visual facial and behavior-based information congruence vs. incongruence in dynamic social decision environments.

Introduction: Initial visual impressions guide judgment and decision-making in social contexts, and social cues, such as facial trustworthiness, may bias choices by influencing individuals to prioritize appearance over actual behavioral outcomes. Key challenges for research in this field include clarifying the extent to which decision-making and learning under uncertainty are affected (1) by facial cues of trustworthiness vs. untrustworthiness, (2) by the congruence vs. incongruence of face-based visual cues and behavioral outcomes, and (3) in a social relative to a non-social decision context. Methods: This study used the Social Iowa Gambling Task (S-IGT), an adaptation of the Iowa Gambling Task (IGT). Extending traditional decision-making paradigms by systematically integrating several novel aspects, the S-IGT features: (1) the experimental manipulation of facial cues perceived as trustworthy vs. untrustworthy and (2) the incorporation of decision options paired with congruent (CS-IGT) vs. incongruent (IS-IGT) social cues; and was used here in combination with (3) a non-social/neutral control condition (the standard IGT). Results: Across 558 participants, we found that initial performance in the CS-IGT was higher than in the other two task conditions, yet participants’ learning progress diminished over time in the CS-IGT. Discussion: These findings suggest that while social cues of facial trustworthiness may support early decision-making, they also may encourage reliance on initial impressions over feedback-driven exploration, thus crucially influencing choice and learning trajectories under uncertainty in social contexts. The results further highlight the complex role of visual facial and behavior-based information congruence vs. incongruence in dynamic social decision environments.

Objectives: Email phishing is a major source of fraud, costing individuals and organizations billions of dollars each year. Older adults are disproportionately vulnerable to phishing and are frequently targeted by scammers. As phishing increases in sophistication, there is a growing need to understand the decision-making processes behind phishing susceptibility: What makes someone susceptible to phishing? Can their susceptibility be predicted, and does phishing susceptibility increase with age? Methods: To address these important research questions, here we take a data-driven approach, fitting a series of neural network models to data from an adult lifespan sample on human decision-making behavior in a phishing email detection task with demonstrated ecological validity. By varying the size of hidden layers in the neural networks, we systematically investigated the dimensionality of phishing susceptibility, asking how many characterizing features of a participant and an email are needed to predict successful detection of phishing emails. We further examined how these dimensions correlated with signal detection metrics and age to elucidate their psychological significance. Results: As few as six dimensions captured almost half the variability in the data. Two dimensions were differentially associated with sensitivity and criterion. Three dimensions showed significant but distinct correlations with age. Discussion: Phishing susceptibility is low-dimensional and moderately predictable. Multiple dissociable components contribute to phishing detection, with age exerting distinct, component-specific influences. This work illustrates the utility of neural network modeling for uncovering latent structure in phishing susceptibility and advances understanding of trust-related decision making in adulthood and aging.

Deception detection represents a complex form of social decision making under uncertainty with real-world consequences. Yet reliable tools to identify vulnerability to fraud and exploitation, marked by large individual variability, particularly among older adults, remain limited. Using a short, scalable laboratory task, the Social Iowa Gambling Task (sIGT), we tested whether adaptive social decision making predicts interpersonal and technologically mediated deception detection. Younger and older adults from a single cohort (Lie Task: N = 153; Deepfake Task: N = 149) completed the sIGT, assessing decision-making in the presence of congruent or incongruent facial trustworthiness cues. Participants judged the veracity of emotional pleas (Lie Task) and face images (Deepfake Task). Deception detection accuracy was analyzed using generalized linear mixed-effects models. Higher sIGT performance predicted greater lie detection accuracy for deceptive, but not truthful, videos, indicating selective sensitivity to misleading cues. Performance under incongruent facial-cue conditions, requiring resistance to trustworthiness bias, showed a stronger relationship with Lie Task accuracy. For the Deepfake Task, higher sIGT scores were associated with greater discernment of deepfakes from real faces, without moderation by facial-cue condition. For both tasks, predictive effects of sIGT performance were age-invariant. Study findings indicate that adaptive social decision-making can predict abilities to detect deception perpetrated through lying and AI-based deepfakes. Results suggest that the ability to adapt under conflicting social cues appears central to resisting lies. As a brief, standardized, and scalable laboratory measure, the sIGT shows promise for identifying individual vulnerability to deception and guiding tailored intervention strategies.

Deepfakes are synthetic media created by deep-generative methods to fake a person’s audio-visual representation. Growing sophistication of deepfake technology poses significant challenges for both machine learning (ML) algorithms and humans. Here we used real and deepfake static face images (Study 1) and dynamic videos (Study 2) (i) to investigate sources of misclassification errors in machines, (ii) to identify psychological mechanisms underlying detection performance in humans, and (iii) to compare humans and machines in their classification decision accuracy and confidence. Study 1 found that machines achieved excellent performance in classifying real and deepfake images, with good accuracy in feature classification. Humans, in contrast, experienced challenges in distinguishing between real and deepfake images. Their classification accuracy was at chance level, and this underperformance relative to machines was accompanied by a truth bias and low confidence for the detection of deepfake images. Using video stimuli, Study 2 found that performance of machines was near chance level, with poor feature classification. Further, the machines showed greater truth bias and low reduced decision confidence relative to humans who outperformed machines in the detection of video deepfakes. Finally, the study revealed that higher analytical thinking, lower positive affect, and greater internet skills were associated with better video deepfake detection in humans. Combined, findings across these two studies advance understanding of factors contributing to deepfake detection in both machines and humans and could inform intervention toward tackling the growing threat from deepfakes by identifying areas of particular benefit from human-AI collaboration to optimize deepfake detection.

Oxytocin, an evolutionarily conserved neuropeptide, plays a crucial role in various physiological and behavioural processes, offering potential therapeutic benefits for several psychiatric and neurodevelopmental conditions. Despite its promise, oxytocin research has been marked by inconsistent results concerning its therapeutic applications and underlying mechanisms. Performing a systematic review and meta-analysis is a popular approach to shed light on mixed findings in a body of literature; however, they can become quickly outdated as new evidence becomes available. Given these challenges, research on oxytocin and its biobehavioural outcomes is ideally positioned for the adoption of ‘living’ meta-analyses, which allow for the continuous integration of new data and updated conclusions. Here we introduce the Active Monitoring of Oxytocin Research Evidence (AMORE) platform (https://amore-project.org), which is a hub that aggregates articles and materials associated with living meta-analyses for biobehavioural oxytocin research. Developed through consensus among 24 expert researchers, a standardized framework was established that either requires or recommends practices ensuring transparency and rigor in living meta-analyses featured on the AMORE platform. Overall, AMORE has been designed to advance oxytocin biobehavioural research by the timely integration of emerging evidence through transparent living meta-analyses