neele engelmann

@neeleengelmann.bsky.social

618 Followers 505 Following

cognitive scientist. postdoc at center for humans and machines, MPI for human development, Berlin. Interested in moral psychology, human-AI interaction, (experimental) philosophy and other things. neeleengelmann.com

People Accurately Anticipate the Influence of Human and LLM Moral Advice Neele Engelmann1, Eva-Madeleine Schmidt1, Ali Alhosseini1, Ivar Rodríguez Hannikainen2, Mengchen Dong1 1 Center for Humans and Machines, Max Planck Institute for Human Development, Berlin 2 Department of Philosophy I, University of Granada Abstract People are often skeptical of AI in the moral domain, yet AI-generated ethical advice can influence moral judgments as much as human advice. We examined whether people accurately anticipate this influence or specifically underestimate the impact of AI moral advice. First, we developed realistic everyday moral scenarios and collected persuasive advice from both incentivized human writers (N = 92) and GPT-4. AI-generated advice was longer, more linguistically complex, more semantically uniform, more virtue-oriented, and less extreme than human advice. Next, a new sample of participants (N = 390) either revised initial moral judgments after receiving human or AI advice or indicated how much they expected such advice to influence them. Contrary to the idea of AI-specific miscalibration, participants were equally influenced by human and AI moral advice, expected similar influence from both sources, and accurately anticipated the overall extent of that influence. These findings suggest that abstract skepticism toward AI in moral contexts does not translate into reduced influence or uniquely poor insight into that influence: people may be influenceable, but they are not blindly swayed. Keywords: Moral Judgment; Metacognition; Algorithm Aversion; Advice-Taking; Human-AI Interaction
Oprah Winfrey in a red blazer with arms outstretched in an enthusiastic pose from her show. Top text: 'YOU GET A REVIEW INVITE, YOU GET A REVIEW INVITE'. Bottom text: 'EVERYBODY GETS A REVIEW INVITE'. Meme format parodying Oprah's famous 'You get a car' scene, here in the context of review invitations.
Image showing the schedule for a conference session titled **"Talks 21: AI 2"**, chaired by **Judith Sieker**, taking place in **Nob Hill C**. The session includes the following lectures: 1. **“LLMs Struggle to Reject False Presuppositions when Misinformation Stakes are High”** Authors: Judith Sieker, Clara Lachenmaier, Sina Zarrieß **Time:** 01 Aug, 09:00 2. **“Enhancing Objectivity in LLM-as-a-Judge through Perturbation Injection”** Authors: Zhihao Zhu, Haoran Liao, Yaohui Jin **Time:** 01 Aug, 09:22 3. **“Framing, not transparency, reduces cheating in algorithmic delegation”** Authors: Neele Engelmann, Lara Kirfel, Anne-Marie Nussberger, Raluca Rilla, Iyad Rahwan **Time:** 01 Aug, 09:44 4. **“Whose Values Prevail? Bias in Large Language Model Value Alignment”** Authors: Ruoxi Qi, Gleb Papyshev, Kellee Tsai, Antoni Chan, Janet Hsiao **Time:** 01 Aug, 10:06
Email starting with "Sehr geehrte Frau Engelmann, ich glaube jetzt ist komplett der Wurm drin."
large number of parked e-scooters
panel discussants at a roundtable