Dehumanizing Machines: Mitigating Anthropomorphic Behaviors in Text Generation Systems

2025-02-19

AI Summary

Increasingly anthropomorphic outputs from large language models (LLMs) risk inducing user overreliance and emotional attachment, posing significant safety concerns. Method: We introduce the first systematic taxonomy of anthropomorphism-mitigating interventions, synthesizing insights from a comprehensive literature review and large-scale crowdsourced editing data; we empirically validate interventions via controlled crowdsourcing experiments, qualitative coding, and conceptual modeling. Contribution/Results: We propose the first evidence-based intervention checklist for mitigating anthropomorphism in LLM outputs and develop a structured theoretical framework that explicates intervention mechanisms, efficacy, and trade-offs. Our framework is interpretable, empirically grounded, and quantitatively evaluable. Collectively, this work establishes a methodological foundation and actionable pathway for enhancing controllability and safety in text-generation systems—advancing responsible LLM deployment through principled, human-centered design.

Technology Category

• Natural Language Processing: Safety and Robustness
• Humans and AI: Intelligent User Interfaces
• Intelligent Robots: Human-Robot Interaction

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

Abstract

As text generation systems' outputs are increasingly anthropomorphic -- perceived as human-like -- scholars have also raised increasing concerns about how such outputs can lead to harmful outcomes, such as users over-relying or developing emotional dependence on these systems. How to intervene on such system outputs to mitigate anthropomorphic behaviors and their attendant harmful outcomes, however, remains understudied. With this work, we aim to provide empirical and theoretical grounding for developing such interventions. To do so, we compile an inventory of interventions grounded both in prior literature and a crowdsourced study where participants edited system outputs to make them less human-like. Drawing on this inventory, we also develop a conceptual framework to help characterize the landscape of possible interventions, articulate distinctions between different types of interventions, and provide a theoretical basis for evaluating the effectiveness of different interventions.

Problem

• Mitigating anthropomorphic behaviors in text systems
• Preventing harmful user dependencies on AI outputs
• Developing interventions to reduce human-like AI responses

Innovation

• Crowdsourced editing reduces anthropomorphic text
• Conceptual framework categorizes intervention types
• Empirical basis evaluates intervention effectiveness