MBMelanie S. Brucks

Melanie S. Brucks

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

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Melanie S. Brucks has a total of 28 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Top Cited Papers
    Open Access

    Digital twins are funhouse mirrors: Five systematic distortions

    by Tianyi Peng, Melanie Brucks, George Gui, Daniel J. Merlau, Grace Jiarui Fan, Malek Ben Sliman, Eric J. Johnson, Abdullah Althenayyan, Silvia Bellezza, Dante Donati, et al.

    Scientists and practitioners are aggressively moving to deploy digital twins—large language model (LLM)-based models of individuals—across social science and policy research. We conducted 19 preregistered studies with 164 diverse outcomes (e.g., attitudes toward hiring algorithms and intention to share misinformation) and compared human responses with those of their digital twins (trained on each person’s previous answers to more than 500 questions). We establish an empirical benchmark for digital twin performance: Digital twins’ answers are only modestly more accurate than those from the (homogeneous) base LLM and correlate weakly with human responses (average correlation coefficient of 0.20). To guide future development, we document five ways in which digital twins distort human behavior: (i) insufficient individuation, (ii) stereotyping, (iii) representation bias, (iv) ideological biases, and (v) hyper-rationality. We make our full dataset and code public as a standardized testbed. Our results caution against premature deployment while laying the groundwork for the transparent, replicable, and iterative science necessary for responsible deployment of digital twins.


  • Video-call glitches trigger uncanniness and harm consequential life outcomes

    by Melanie S. Brucks, Jacqueline R. Rifkin, Jeff S. Johnson

    People are increasingly using video calls for high-stakes interactions that once required face-to-face contact: from medical consultations1,2, to job interviews3, to court proceedings4. But video calling introduces a new communication issue: minor glitches, or intermittent errors in the transmission of audiovisual information during a virtual interaction5. Here, through five experiments and three supplementary studies using both live and recorded interactions, we show that minor audiovisual glitches during video calls harm interpersonal judgements in consequential life domains (for example, hiring decisions after a virtual interview, or trust in a medical provider after a telehealth visit). In addition, two archival datasets from real-world video calls reveal that glitches are associated with both reduced social connection and a lower likelihood of being granted criminal parole. We find that audiovisual glitches damage interpersonal judgements because they break the illusion of face-to-face contact (for example, by distorting faces, misaligning audio and visual cues or making movements appear ‘choppy’), evoking ‘uncanniness’—a strange, creepy or eerie feeling6,7. As the uncanniness of a glitch increases, so does its negative effect on interpersonal judgements. Furthermore, audiovisual glitches undermine interpersonal judgements only in video calls that simulate face-to-face interaction, showing that the negative effect produced by glitches goes beyond mere disruptiveness, comprehension difficulties and negative attributions. These findings have important implications for digital equity. Despite being considered a boon to access, virtual communication might unintentionally perpetuate inequality. Because disadvantaged groups often have poorer internet connections8,9,10,11,12, they are likely to encounter more glitches, and, in turn, to experience worse outcomes in consequential contexts such as health, careers, justice and social connection.


  • Kinesthetic Properties of Response Scales Yield Different Judgments

    by Melanie S Brucks, Jonathan Levav

    This research examines how the movements an interface requires of a consumer—that is, its “kinesthetic properties”—can alter what a consumer attends to when responding and in turn change the response itself. We compare the kinesthetic properties of two ubiquitous scale formats: slider and radio-button scales. Six studies (plus four in the web appendix) show dragging a slider (vs. clicking a radio button) elicits responses that are closer to the scale starting point. This effect occurs because the slider allows participants to engage with the scale as they consider their options. When dragging past each response option, attention is directed to that option, increasing its chance of being selected. Supporting this account, sliders only result in responses closer to the starting point when participants physically drag the cursor across options to their desired response, not when they directly click on it. Furthermore, participants dragging a slider interact with the scale earlier in the judgment process and exhibit a greater visual focus on left-side (vs. right-side) options on the scale compared to participants clicking a radio button. These findings suggest that marketers, graphic designers, and researchers should consider how the kinesthetic properties of digital interfaces may shape consumer judgment.


  • Top Cited Papers
    Open Access

    Prompt architecture induces methodological artifacts in large language models

    by Melanie Brucks, Olivier Toubia

    We examine how the seemingly arbitrary way a prompt is posed, which we term “prompt architecture,” influences responses provided by large language models (LLMs). Five large-scale, full-factorial experiments performing standard (zero-shot) similarity evaluation tasks using GPT-3, GPT-4, and Llama 3.1 document how several features of prompt architecture (order, label, framing, and justification) interact to produce methodological artifacts, a form of statistical bias. We find robust evidence that these four elements unduly affect responses across all models, and although we observe differences between GPT-3 and GPT-4, the changes are not necessarily for the better. Specifically, LLMs demonstrate both response-order bias and label bias, and framing and justification moderate these biases. We then test different strategies intended to reduce methodological artifacts. Specifying to the LLM that the order and labels of items have been randomized does not alleviate either response-order or label bias, and the use of uncommon labels reduces (but does not eliminate) label bias but exacerbates response-order bias in GPT-4 (and does not reduce either bias in Llama 3.1). By contrast, aggregating across prompts generated using a full factorial design eliminates response-order and label bias. Overall, these findings highlight the inherent fallibility of any individual prompt when using LLMs, as any prompt contains characteristics that may subtly interact with a multitude of hidden associations embedded in rich language data.


  • Top Cited Papers

    Virtual communication curbs creative idea generation

    by Melanie S. Brucks, Jonathan Levav

    COVID-19 accelerated a decade-long shift to remote work by normalizing working from home on a large scale. Indeed, 75% of US employees in a 2021 survey reported a personal preference for working remotely at least one day per week1, and studies estimate that 20% of US workdays will take place at home after the pandemic ends2. Here we examine how this shift away from in-person interaction affects innovation, which relies on collaborative idea generation as the foundation of commercial and scientific progress3. In a laboratory study and a field experiment across five countries (in Europe, the Middle East and South Asia), we show that videoconferencing inhibits the production of creative ideas. By contrast, when it comes to selecting which idea to pursue, we find no evidence that videoconferencing groups are less effective (and preliminary evidence that they may be more effective) than in-person groups. Departing from previous theories that focus on how oral and written technologies limit the synchronicity and extent of information exchanged4,5,6, we find that our effects are driven by differences in the physical nature of videoconferencing and in-person interactions. Specifically, using eye-gaze and recall measures, as well as latent semantic analysis, we demonstrate that videoconferencing hampers idea generation because it focuses communicators on a screen, which prompts a narrower cognitive focus. Our results suggest that virtual interaction comes with a cognitive cost for creative idea generation.