Attributions of intent and moral responsibility to AI agents
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Moral transactions are increasingly infused with decision input from AI agents. To what extent do observers believe that AI agents are responsible for their own actions? How do these AI agents' socio-psychological features affect observers' judgment of them when they transgress? With full factorial, between-participant designs, we presented participants with vignettes in which an AI agent contributed to a negative outcome either intentionally or unintentionally. We independently manipulated four features of the agent's mind: its adherence to moral values, autonomy, emotional self-awareness, and social connectedness. In Study 1 ( N = 2012), AI agents that intentionally contributed to a negative outcome consistently received harsher judgments than AI agents that contributed unintentionally. For unintentional actions, socially connected AI agents received less harsh judgments than socially disconnected AI agents. In Studies 2a-c ( N = 1507), these judgments were explained by ratings of the socially connected AI agent's ‘mind’ as less distinct from the mind of its programmers (Study 2b) and that this kind of agent also possessed less free will (Study 2c). We discuss the implications of these findings in advancing the field's understanding of the moral psychology—and design—of AI agents. • Moral judgments of AI agents are sensitive to the agents' manipulated intentionality. • AI agents contributing intentionally to negative outcomes received harsher judgments. • AI agents embedded in a human social network received more lenient judgments. • Socially connected AI agents are rated as possessing less free will. • Socially connected AI agents' ‘minds’ are rated as less distinct entities.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it