# Summary: 2026-09-04_11-56-59Z_HowdoLLMsEvaluatePerceivedMoralAgency_Investigatin.md
Saved: 2026-09-06 21:45
Source: 2026-09-04_11-56-59Z_HowdoLLMsEvaluatePerceivedMoralAgency_Investigatin.md
Original paper: [arXiv:2609.05037](https://arxiv.org/abs/2609.05037)
Model: None

---

## Summary  
The paper investigates how large language models perceive moral agency in human‑AI interactions, comparing human and AI evaluations across embodied and non‑embodied agents situated in plausible smart‑city scenarios. It is the first empirical study to apply a validated perceived moral agency scale to both participants and LLMs, revealing systematic differences in agency attribution.

## Key Contributions  
- Finding 1: Humans consistently attribute higher moral agency than autonomous artificial agents.  
- Finding 2: In concrete moral dilemmas, LLMs prioritize harm severity and contextual urgency over any stable assessment of the agent itself.  
- Finding 3: Both humans and LLMs show context‑sensitivity in judgments, but AI reasoning is more situational.

## Methodology  
The authors adapted a validated perceived moral agency scale for human participants (N = 190) and applied it to multiple LLMs within simulated smart‑city contexts involving embodied robots, drones, and disembodied agents. Participants evaluated scenarios where each agent faced ethical choices; the same prompts were fed to LLMs to capture their internal reasoning.

## Results  
Human raters rated humans as moral agents (average score 8.2/10) while AAs scored lower (6.5/10). When dilemmas involved harm severity, LLMs produced higher scores for severe outcomes regardless of agent type, indicating context‑driven weighting. Human ratings also dropped with less urgent contexts.

## Significance  
These findings clarify the gap between human moral agency attribution and AI reasoning, informing design of responsible AI in public services where ethical decisions are routine.

## Related Concepts  
- Moral Agency: capacity to act ethically and be held accountable.  
- Perceived Moral Agency (PMA): subjective judgment of an agent's moral responsibility.  
- Embodied vs. Disembodied Agents: physical presence influences agency perception.  
- Contextual Urgency: immediacy of ethical stakes.