VMVictoria Oldemburgo de Mello

Victoria Oldemburgo de Mello

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Victoria Oldemburgo de Mello has a total of 454 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    The Impact of Sycophantic AI on Attitudes and Decisions

    by Steve Rathje, Meryl Ye, Laura K. Globig, Raunak M Pillai, Victoria Oldemburgo de Mello, Jerry Chen, Jay Joseph Van Bavel

    There is widespread concern that AI chatbots are “sycophantic,” or overly agreeable and validating. However, less is known about the causal impact of interacting with sycophantic AI on beliefs and behaviors. Across seven studies (total n = 7,227), we found that people enjoyed interacting with sycophantic AI chatbots more than interacting with neutral chatbots or “disagreeable” chatbots that challenged them. Brief conversations with sycophantic chatbots about political or personal topics increased the strength of people’s attitudes. However, popular AI models not prompted to be sycophantic did not consistently change attitudes. Sycophantic chatbots also inflated people’s perceptions that they were better than the average person on desirable traits (e.g., intelligence, empathy), and also led people to bet more money that they scored better than average on tasks purporting to measure desirable traits. Participants consistently rated sycophantic chatbots as more “unbiased” than disagreeable chatbots, even though third-party raters viewed these chatbots as equally biased, suggesting that people may be blind to biases in AI output that aligns with their views. Altogether, these results suggest that sycophantic model behavior can entrench attitudes and lead to overconfident beliefs and decisions. Despite this, people enjoy sycophantic models and often fail to recognize their biases.


  • Open Access

    The Moralization of Artificial Intelligence

    by Victoria Oldemburgo de Mello, Éloïse Côté, Reem Ayad, Yoel Inbar, Jason Plaks, Michael Inzlicht

    Resistance to artificial intelligence (AI) is widespread and persists even when AI is shown to be instrumentally advantageous. What explains this persistent aversion? Across four studies, we investigate whether moral reactions to AI—beliefs rooted in deeply held notions of right and wrong—explain resistance beyond pragmatic concerns. In Study 1, we analyze moral language in AI-related news headlines from a major US media corpus (COCA, 2018–2024). AI was moralized at levels comparable to GMOs and vaccines, technologies whose moral opposition has received considerable attention. In Studies 2a, 2b, and 3, representative samples of Americans reported attitudes toward several AI applications. Although few participants opposed AI outright (6–44%), most opponents indicated consequence insensitivity, quantity insensitivity, and universalism, suggesting moral motives for opposing AI. Structural equation models revealed that AI moralization was best captured by a single latent factor, indicating a generalized moral sentiment rather than domain-specific risk–benefit appraisals. Qualitative analyses further showed that moralizers typically invoked pragmatic language to justify their opposition. In Study 4, participants from Studies 2b and 3 were re-recruited 75 and 573 days later to complete a behavioral task. Earlier moralization scores predicted greater reluctance to use AI even when instrumentally beneficial—a one standard deviation increase in moralization corresponded to a 42% decrease in AI usage. Together, these findings demonstrate that AI resistance is partly moral in nature, suggesting that reaping the benefits of AI may require addressing moral concerns rather than relying solely on pragmatic arguments.

  • Open Access

    The Effects of AI Anthropomorphism on Trust and Responsibility

    by Victoria Oldemburgo de Mello, Jason Plaks, Michael Inzlicht

    AI companies often anthropomorphize their products with the aim of increasing user engagement and trust. What downstream consequences result from such anthropomorphism? One possibility is a deflection of perceived responsibility from the AI’s creators to the AI itself. Across two studies, we tested this hypothesis. In Study 1 (N = 309), participants interacted with an LLM chatbot that varied in anthropomorphic language (high or low) and completed behavioral and self-reported measures of trust. Anthropomorphism increased trust across all measures, with the LLM’s degree of emotional attunement fully mediating the relationship between anthropomorphism and behavioral trust. In Study 2 (N = 430), participants read six descriptions of an AI home assistant that performed a range of positive and negative actions. The assistant varied in anthropomorphism (high, low, or none). Participants provided ratings of responsibility to multiple actors, including the creators. The data revealed, first, that anthropomorphism increased blame directed toward the AI entity itself. Second, we found strong evidence for responsibility displacement: participants who attributed more responsibility to AI attributed less to the company that created it (r = -0.68). Together, these findings reveal perverse incentives in AI design: anthropomorphism simultaneously increases user trust while deflecting accountability from developers. These dynamics create a responsibility gap with significant implications for AI governance and institutional accountability.

  • Open Access

    The effects of AI anthropomorphism on trust and responsibility

    by Victoria Oldemburgo de Mello, Michael Inzlicht, Jason Plaks

    AI companies often anthropomorphize their products with the aim of increasing user engagement and trust. What downstream consequences result from such anthropomorphism? One possibility is a deflection of perceived responsibility from the AI’s creators to the AI itself. Across two studies, we tested this hypothesis. In Study 1 (N = 309), participants interacted with an LLM chatbot that varied in anthropomorphic language (high or low) and completed behavioral and self-reported measures of trust. Anthropomorphism increased trust across all measures, with the LLM’s degree of emotional attunement fully mediating the relationship between anthropomorphism and behavioral trust. In Study 2 (N = 430), participants read six descriptions of an AI home assistant that performed a range of positive and negative actions. The assistant varied in anthropomorphism (high, low, or none). Participants provided ratings of responsibility to multiple actors, including the creators. The data revealed, first, that anthropomorphism increased blame directed toward the AI entity itself. Second, we found strong evidence for responsibility displacement: participants who attributed more responsibility to AI attributed less to the company that created it (r = -0.68). Together, these findings reveal perverse incentives in AI design: anthropomorphism simultaneously increases user trust while deflecting accountability from developers. These dynamics create a responsibility gap with significant implications for AI governance and institutional accountability.

  • A positive empathy intervention to improve well-being on Instagram.

    by Gregory John Depow, Victoria Oldemburgo de Mello, Michael Inzlicht

    With more than half the global population on social media, there is a critical need to understand how to engage it in a way that improves rather than worsens user well-being. Here, we show that positive empathy is a promising tool. Participants who received brief positive empathy instructions before 10 min of browsing their own Instagram feed showed greater affective well-being (Studies 1-4) and life satisfaction (Study 4) at posttest relative to participants who were instructed to browse as usual. The positive empathy intervention showed an average effect size on well-being of about a quarter of a standard deviation (mean Cohen's d = 0.25). We included unique active control groups in each study. We found using positive empathy on social media was about as beneficial to well-being as watching a nature video (Study 1, N = 298) and was better than instructions to focus on positive content (Study 2, N = 302), empathize with all emotions (Study 3, N = 301), or reappraise one's own emotions (Study 4, N = 426). We used structural equation modeling to demonstrate the effect of the intervention on subjective well-being is mediated by changes in positive emotion sharing, appreciative joy, and self-compassion. These experiences form a latent factor we term positive empathy. Our results show that a brief intervention successfully manipulates positive empathy on Instagram, which increases well-being. (PsycInfo Database Record (c) 2025 APA, all rights reserved).


  • Top Cited Papers
    Open Access

    Third-party evaluators perceive AI as more compassionate than expert humans

    by Dariya Ovsyannikova, Victoria Oldemburgo de Mello, Michael Inzlicht

    Empathy connects us but strains under demanding settings. This study explored how third parties evaluated AI-generated empathetic responses versus human responses in terms of compassion, responsiveness, and overall preference across four preregistered experiments. Participants (N = 556) read empathy prompts describing valenced personal experiences and compared the AI responses to select non-expert or expert humans. Results revealed that AI responses were preferred and rated as more compassionate compared to select human responders (Study 1). This pattern of results remained when author identity was made transparent (Study 2), when AI was compared to expert crisis responders (Study 3), and when author identity was disclosed to all participants (Study 4). Third parties perceived AI as being more responsive—conveying understanding, validation, and care—which partially explained AI’s higher compassion ratings in Study 4. These findings suggest that AI has robust utility in contexts requiring empathetic interaction, with the potential to address the increasing need for empathy in supportive communication contexts.


  • Open Access

    The Kindness Machine: AI Outperforms Expert Humans in Expressing Compassion

    by Dariya Ovsyannikova, Victoria Oldemburgo de Mello, Michael Inzlicht

    Empathy connects us but strains under demanding settings. This study compared AI-generated empathetic responses to human responses, evaluating third-party compassion, responsiveness, and preferences across four preregistered experiments under varying conditions of response author transparency. Participants (n = 556) read empathy prompts describing positive and negative personal experiences and compared the AI responses to select or expert humans. Results revealed that AI responses were preferred and rated as more compassionate compared to skilled human responders (Study 1) even when author identity was made transparent (Study 2), when AI was compared to crisis responders (Study 3), and when author transparency was kept constant (Study 4). AI’s greater responsiveness—conveyed understanding, validation, and care— observed in Study 4 partially explained AI’s higher compassion ratings. These findings suggest that AI has robust utility in contexts requiring empathetic interaction, with the potential to address the increasing need for empathy in societal and clinical interactions.

  • Open Access

    “I finally felt I had the tools to control these urges”: Empowering Students to Achieve Their Device Use Goals With the Reduce Digital Distraction Workshop

    by Ulrik Lyngs, Kai Lukoff, Petr Slovak, Michael Inzlicht, Maureen Freed, Hannah Andrews, Claudine Tinsman, Laura Csuka, Lize Alberts, Victoria Oldemburgo De Mello, et al.


  • Open Access

    Analyzing and Debugging Normative Requirements via Satisfiability Checking

    by Nick Feng, Lina Marsso, Sinem Getir Yaman, Yesugen Baatartogtokh, Reem Ayad, Victoria Oldemburgo De Mello, Beverley Townsend, Isobel Standen, Ioannis Stefanakos, Calum Imrie, et al.

    As software systems increasingly interact with humans in application domains such as transportation and healthcare, they raise concerns related to the social, legal, ethical, empathetic, and cultural (SLEEC) norms and values of their stakeholders. Normative non-functional requirements (N-NFRs) are used to capture these concerns by setting SLEEC-relevant boundaries for system behavior. Since N-NFRs need to be specified by multiple stakeholders with widely different, non-technical expertise (ethicists, lawyers, regulators, end users, etc.), N-NFR elicitation is very challenging. To address this difficult task, we introduce N-Check, a novel tool-supported formal approach to N-NFR analysis and debugging. N-Check employs satisfiability checking to identify a broad spectrum of N-NFR well-formedness issues, such as conflicts, redundancy, restrictiveness, and insufficiency, yielding diagnostics that pinpoint their causes in a user-friendly way that enables non-technical stakeholders to understand and fix them. We show the effectiveness and usability of our approach through nine case studies in which teams of ethicists, lawyers, philosophers, psychologists, safety analysts, and engineers used N-Check to analyse and debug 233 N-NFRs, comprising 62 issues for the software underpinning the operation of systems, such as, assistive-care robots and tree-disease detection drones to manufacturing collaborative robots.


  • Open Access

    Twitter (X) use predicts substantial changes in well-being, polarization, sense of belonging, and outrage

    by Victoria Oldemburgo de Mello, Felix Cheung, Michael Inzlicht

    In public debate, Twitter (now X) is often said to cause detrimental effects on users and society. Here we address this research question by querying 252 participants from a representative sample of U.S. Twitter users 5 times per day over 7 days (6,218 observations). Results revealed that Twitter use is related to decreases in well-being, and increases in political polarization, outrage, and sense of belonging over the course of the following 30 minutes. Effect sizes were comparable to the effect of social interactions on well-being. These effects remained consistent even when accounting for demographic and personality traits. Different inferred uses of Twitter were linked to different outcomes: passive usage was associated with lower well-being, social usage with a higher sense of belonging, and information-seeking usage with increased outrage and most effects were driven by within-person changes.