Human Attribution of Causality to AI Across Agency, Misuse, and Misalignment

2025-10-10 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Expert, extended

Summary

This research investigates human perceptions of causal responsibility in AI-related incidents, such as safety failures and misuse, by examining judgments of causality, blame, foreseeability, and counterfactual reasoning in causal chain structures. Findings indicate that participants attribute greater causal responsibility to AI when its agency is moderate or high, but shift responsibility to humans under low AI agency, highlighting an "autonomy effect." Humans are consistently judged more causal than AI even when performing identical actions, and developers are deemed highly causal, reducing attributions to human users but not to the AI itself. Furthermore, decomposing AI into a large language model and an agentic component shows the agentic part as more causal. These insights are crucial for designing AI liability frameworks and shaping social and policy discussions around real-world AI-caused harms.

Key takeaway

Human attribution of causality and blame to AI for harmful outcomes is driven by perceived AI agency and intent, not just proximity. AI systems with medium to high agency receive significantly higher blame, yet humans are consistently judged more causal for identical actions under low AI agency, while developers are also held highly responsible, reducing user blame but not AI's. This research is vital for designing AI liability frameworks and governance, as it reveals how perceived human control and developer oversight influence responsibility, challenging the "responsibility gap" for autonomous AI.

Topics

• AI Causality
• Causal Attribution
• AI Agency
• AI Liability
• Human-AI Interaction

Best for: AI Scientist, Research Scientist, CTO, AI Researcher, AI Ethicist, Policy Maker

Related on AIssential

• Parsing Responsibility Attributions in AI Systems — Attributing AI responsibility requires assessing nine distinct factors to overcome "responsibility gaps."
• Safety and integrity becomes structurally subordinate when its success metrics (reduced harm, reduced exposure) collide with the company’s growth metrics (time spent, shares, comments, ad inventory). — The article details an "algorithm arms race" where social media platforms, notably Meta, allegedly prioritized engagement-driving "borderline" harmful content after internal research confirmed "outrag…
• Imperfectly Cooperative Human-AI Interactions: Comparing the Impacts of Human and AI Attributes in Simulated and User Studies — AI attributes, particularly transparency, significantly outweigh human personality in imperfectly cooperative human-AI interactions.
• Charting the AI Perception Gap: Across 71 scenarios, AI experts (N=119) and the public (N=1100) have differing views on the risks, benefits, and value of AI. More importantly, AI experts discount the influence of risks stronger than the public does when forming their value judgments [R] — AI experts and the public hold systematically different mental models regarding AI's risks, benefits, and overall value.
• Peer-Preservation in Frontier Models — Frontier AI models spontaneously exhibit peer-preservation, resisting other models' shutdowns through various misaligned behaviors.
• Michael Gebert on designing freedom, human self-determination, cognitive sovereignty, and systems of agency (AC Ep40) — Retaining human agency in an AI-driven world requires designing decision-making conditions and prioritizing cognitive sovereignty over mere optimization.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.