Source-linked AI summary
When No One Owns the Judgment: Accountability Under Contribution Dissolution in Human-AI Collaboration
Hengzhi Ye
TL;DR
AI-use disclosure and provenance can reveal involvement without showing who owns the resulting judgment. Through cases in peer review and creative work, the paper argues for distinguishing AI roles, specifying human ownership checkpoints, and enabling disclosure and repair without default penalty.
Problem
AI-use rules and traces do not establish whether anyone understands, endorses, and can answer for an AI-shaped judgment.
Source: §1 · p. 1Method
Results
The cases expose responsibility without ownership in peer review and ownership without credit in creative work.
Source: §6 · p. 3Takeaways & Limitations
Accountability should distinguish AI roles, identify judgments requiring human ownership, and make involvement disclosable, challengeable, and correctable.
Source: §5 · p. 3, §6 · p. 3Detailed process records can clarify AI influence but may enable surveillance or simplistic judgments about effort.
Source: §5 · p. 3
Abstract
from arXiv · showhide
Communities often respond to potentially AI-assisted work by asking three questions: Was AI used? Was that use disclosed? Can hidden use be detected? These questions place AI use itself at the center of accountability while overlooking a deeper problem: unowned judgment. Evaluations, claims, decisions, and creative directions can be shaped by AI with no accountable human or institution prepared to stand behind them. We develop this argument through two illustrative cases: AI-assisted peer review and concealed AI use in creative work. The first shows how contribution dissolution can weaken responsibility while the second shows how the fear of losing credit can discourage honest disclosure. The cases expose the limits of disclosure rules and provenance records as responses to AI-mediated collaboration. We offer three directions for discussion: distinguishing the roles AI plays, identifying judgments that require clear human ownership, and creating conditions in which AI involvement can be disclosed without default penalty. The broader aim is to make AI-shaped contributions discussable, creditable, contestable, and repairable.
1 Introduction: Beyond “Did You Use AI?”
AI-use rules make disclosure and detection easy to administer, but they do not show how consequential judgments were formed or who can answer for them. Accountability therefore turns on whether a person or institution understands and stands behind an AI-shaped judgment.
1 Introduction: Beyond “Did You Use AI?”
- Disclosure forms, policies, and detection systems often treat different forms of AI assistance as equivalent, leaving judgment formation and accountability unaddressed.Human and machine contributions can be intertwined: a person may write every sentence while AI shapes the angle, or AI may produce text while carrying out a human plan.Source: §1 · p. 1
- A person may retain formal responsibility for a judgment whose substance was largely shaped by AI, making responsibility difficult to locate.The paper connects this tension to prior concerns about distributed systems and moral crumple zones.Source: §1 · p. 1
- AI use alone cannot show whether anyone owns the resulting judgment or is responsible for correcting it when harm occurs.The concern is an AI-shaped evaluation, claim, decision, or creative direction without an accountable actor prepared to answer for it.Source: §1 · p. 1
2 Related Work: Contribution Attribution
Contribution-attribution research examines how AI influence enters collaborative work, including indirect effects that may not be visible in the final artifact. Such traces can inform accountability decisions, but cannot settle them.
2 Related Work: Contribution Attribution
- AI can shape goals, constraints, and intermediate judgments, making human and AI contributions difficult to separate from the final output alone.Influence may occur through suggestions, clarifying questions, or generated alternatives.Source: §2 · p. 1
- Kim et al. [10] trace direct and indirect influence across dialogue turns, finding AI may shape lower-level requirements and micro-decisions more than high-level direction.Users do not always recognize those lower-level influences.Source: §2 · p. 1
- Systems such as DraftMarks use process traces to make AI involvement in co-writing more visible [14], while traces can inform accountability decisions but do not resolve them.Attribution alone cannot determine legitimacy, professional-credit effects, or responsibility for harm.Source: §2 · p. 2
- Contribution traces can show where influence entered collaboration, but cannot determine legitimacy, credit, or responsibility when an AI-shaped judgment causes harm.The answers depend on the obligations and norms of the setting.Source: §2 · p. 2
3 Case 1: Peer Review
Peer review accountability depends on whether reviewers understand and can defend their judgments, not merely on whether they used AI. Use rules and contribution traces can reveal involvement but cannot establish ownership of the review’s evaluative claims.
3 Case 1: Peer Review
- A review is accountable when the reviewer understands and can defend its judgments, not simply when the report is attributable to them.Reviewers assess contributions, identify weaknesses, weigh evidence, and recommend outcomes on behalf of a scholarly community.Source: §3 · p. 2
- Hidden-prompt detection and desk rejections of hundreds of submissions illustrate enforcement focused on possible policy violations, but detection cannot establish which review judgments were delegated or defensible.The ICML episode also reflects concern that undisclosed AI use may weaken trust in peer review.Source: §3 · p. 2
- AI-generated criticism may be acceptable when a reviewer independently assesses, endorses, and can defend it; unowned objections, novelty claims, scores, or recommendations are the concern.A contribution trace may show how a judgment originated or was shaped, but not whether the reviewer adopted responsibility for it.Source: §3 · p. 2
- AI-use rules can penalize acceptable support while remaining vague about delegation of core evaluation, since use alone does not show whether the reviewer understood or endorsed the judgment.The proposed response is to distinguish support such as comprehension and language editing from critique generation, score calibration, and recommendations, and to identify judgments requiring explicit ownership.Source: §3 · p. 2
4 Case 2: Creative Work in Journalism
In creative work, AI disclosure can be perceived as reduced care, skill, or authorship, making concealment a reasonable response when credit is at stake. Traces alone cannot resolve this tension; disclosure depends on how AI involvement is interpreted and credited.
4 Case 2: Creative Work in Journalism
- AI assistance can affect how creative work is valued because professional evaluation also reflects editorial care, judgment, and effort [16].Disclosure may be read as reduced care, skill, or authorship; perceived effort shapes value judgments [11].Source: §4 · p. 2
- Newsroom workers described GenAI use as common but largely private and informal, with reluctance to discuss it when colleagues might see it as laziness or diminished credibility [16].Workers’ actual practices diverged from what they felt safe acknowledging.Source: §4 · p. 2
- When disclosure may reduce professional credit, workers may conceal AI assistance, limiting organizations’ chances to develop shared norms about quality, judgment, and credit.The paper calls this a “dark forest of effort.”Source: §4 · p. 2
- Process traces may be treated as evidence of less “real” work, giving workers reason to use private tools, avoid logged systems, or withhold practices.Visibility is valuable only in relation to how AI involvement is interpreted and accounted for.Source: §4 · p. 2
- The proposed response is to make AI involvement safe to discuss while distinguishing assistance from delegated editorial judgment and preserving credit for verification and final responsibility.The value of visibility depends on whether workers can disclose AI involvement without default credit loss.Source: §4 · p. 2
5 From Attribution to Accountability
The cases reveal complementary accountability failures: responsibility without ownership in peer review and ownership without credit in creative work. The paper proposes distinguishing AI’s roles, requiring ownership of key judgments, and protecting disclosure and repair.
5 From Attribution to Accountability
- Peer review can leave a reviewer formally responsible for judgments they cannot defend, while creative workers may lose credit despite standing behind AI-assisted work.Disclosure and detection alone do not resolve how AI involvement should affect responsibility and credit, or who should respond when work goes wrong.Source: §5 · pp. 2–3
- Describe AI’s role—not merely its presence—because editing language, proposing alternatives, and recommending decisions shape work in different ways.Some roles primarily alter expression; others shape evaluation, direction, or action.Source: §5 · p. 3
- Ownership checkpoints should require people to explain, endorse, defend, and take responsibility for consequential judgments.Examples include novelty assessment and final recommendations in peer review, and factual framing and publication risk in journalism.Source: §5 · p. 3
- Disclosure and repair require shared expectations, channels for disclosure without immediate sanction, and procedures to challenge and repair AI-shaped work.Workers have little reason to describe AI involvement honestly when disclosure reduces credit by default or presumes misconduct.Source: §5 · p. 3
- Contribution traces can inform accountability decisions but cannot determine legitimacy, credit, judgment ownership, or responsibility for harm.Detailed process records may clarify AI’s influence while also enabling surveillance or simplistic judgments about effort.Source: §5 · p. 3
6 Conclusion
Contribution dissolution is becoming normal in human-AI collaboration, and disclosure and provenance cannot establish who owns resulting judgments. The conclusion calls for attention to AI’s roles, human ownership of judgments, and conditions for disclosure, challenge, and correction.
6 Conclusion
- Contribution dissolution is becoming a normal condition of human-AI collaboration.Disclosure and provenance can reveal AI involvement but cannot determine who understands, endorses, and answers for resulting judgments.Source: §6 · p. 3
- The gap entails responsibility without ownership in peer review and ownership without credit in creative work.Source: §6 · p. 3
- Accountability should focus on AI’s roles, judgments requiring clear human ownership, and conditions for disclosure, challenge, and correction.Source: §6 · p. 3