Recent work in Human-Centered Computing has shown that people interacting with generative AI systems are susceptible to what psychology of judgment describes as the Anchoring Bias (Rosbach et al. 2026, Haag et al. 2024, Hosanagar et al. 2024). When users receive an initial estimate, explanation, or suggestion from a system, their subsequent judgments tend to adjust around that value rather than being independently generated. At the same time, a growing body of recent research suggests that large generative models themselves exhibit anchoring-like behavior (Lou et al. 2025, Valencia-Clavijo 2025). Because these systems are trained to produce predictions conditioned on context through mechanisms such as In-Context Learning within Transformer Neural Networks, elements appearing early in a prompt can systematically shift the distribution of their outputs.
In this talk, I will develop a philosophical analysis of the interaction between these two sources of bias. I will introduce a new notion, “double anchoring”, that refers to the phenomenon of human–AI interaction in which anchoring occurring in language models and human judgment combine and reinforce one another. In such cases, contextual cues anchor the model’s response, while the model’s output in turn anchors the user’s reasoning, producing a feedback loop of bias amplification within the interaction. I will argue that while double anchoring does not necessarily undermine the creative capacities of humans or artificial systems considered independently, it poses a potential threat to the creativity of their interaction. I propose that, in contexts involving generative AI, the appropriate unit of analysis for creativity is not the human or the artificial system alone but the human–AI interactive system. I will outline the implications of this phenomenon for philosophical accounts of creativity in hybrid cognitive systems and for the design of human–AI interaction.
REFERENCES
Haag, S., Schäfer, J., & Kett, H. (2024). Overcoming Anchoring Bias: The Potential of AI and Explainable AI-Based Decision Support Systems. arXiv preprint.
Hosanagar, K., & Ahn, D. (2024). Designing Human and Generative AI Collaboration. Science Advances, 10, eadn5290.
Lou, J., & Sun, Z. (2025). Anchoring Bias in Large Language Models: An Experimental Study.Journal of Behavioral and Experimental Finance (forthcoming / online first).
Rosbach, P., Schultze, T., & Kollegen. (2026). Stuck on Suggestions: Automation Bias, the Anchoring Effect, and the Role of Cognitive Load in Human-AI Decision-Making. arXiv preprint.
Valencia-Clavijo, F. (2025). Anchors in the Machine: Behavioral and Attributional Evidence of Anchoring Bias in LLMs. arXiv preprint arXiv:2511.05766.
Speaker: Mariela Destéfano