AI Use Conditions and Perspective Diversity in Ethical Decision-Making: A Pilot Study of Human Reasoning Processes

URL SCAN: AI Use Conditions and Perspective Diversity in Ethical Decision-Making: A Pilot Study of Human Reasoning Processes
FIRST LINE: # Computer Science > Human-Computer Interaction

The Dissection

This paper is not studying whether humans remain ethically sovereign. It is studying whether AI can make human deliberation look broader before humans deliver an already familiar conclusion.

The design is narrow: 29 participants, one ethical dilemma, three conditions, and two exploratory coding measures. The headline finding is that mandatory AI use increased the range of perspectives mentioned, while final judgments generally converged on disclosure and customer protection. The paper therefore frames AI as a reasoning aid that expands deliberative coverage without replacing human judgment.

That is a politically and institutionally convenient framing. It preserves the human decision-maker as the apparent locus of responsibility while importing machine-generated breadth into the process.

The Core Fallacy

The central error is confusing perspective accumulation with ethical intelligence.

A participant can mention legal, regulatory, organizational, technical, and ethical considerations without understanding any of them, weighing them correctly, or detecting that the AI has supplied generic categories. A higher Perspective Diversity Index measures a wider verbal search space, not better judgment, truth, accountability, or resistance to manipulation.

The study also treats stable final conclusions as evidence that AI broadens reasoning without altering judgment. That inference is weak. A single conventional dilemma with an intuitively attractive answer cannot test whether AI preserves human autonomy under ambiguous, high-stakes, adversarial, or materially conflicting conditions. It tests whether AI helps people produce a more elaborate justification for a socially approved answer.

Under the Discontinuity Thesis, this is the early cultural form of cognitive automation: the machine does not need to seize the final decision immediately. It first colonizes the search, framing, categorization, and justification stages that make the final decision appear human-owned.

Hidden Assumptions

• That broader consideration is inherently better rather than merely more verbose or more compliant with AI-suggested framing.
• That the three conditions are comparable despite the tiny sample and possible differences in participant motivation, AI literacy, prompting, and compliance.
• That one dilemma can represent ethical decision-making generally.
• That blind human coders can reliably measure reasoning breadth without embedding their own assumptions into CDS and PDI.
• That statistically significant PDI movement is substantively meaningful rather than an artifact of exploratory measurement, coding choices, or multiple comparisons.
• That convergent conclusions demonstrate preserved human agency rather than successful normalization of machine-mediated reasoning.
• That AI output is neutral across legal, technical, organizational, and ethical categories. It is not. The system selects what becomes salient before the human evaluates it.
• That keeping the final judgment human is the relevant control point. In practice, whoever controls the upstream framing often controls the decision.

Social Function

Primary classification: partial truth, prestige signaling, and transition management.

The partial truth is real: AI can increase the number of categories a person considers. The prestige signal is the conversion of that modest result into evidence that AI improves ethical reasoning. The transition-management function is more consequential. It normalizes mandatory AI participation in judgment while reassuring institutions that humans still retain responsibility.

This is not yet full ideological anesthetic, but it is compatible with one. It offers organizations a future-proofing narrative: automate the cognitive scaffolding, retain a human signature, and call the resulting process responsible.

The Verdict

The paper demonstrates a limited phenomenon: compelled AI use can increase coded perspective breadth in a small laboratory exercise. It does not demonstrate better ethics, durable human autonomy, or safe decision-making.

Its deeper significance is structural. AI is moving upstream from answering questions to determining which questions, categories, risks, and justifications enter human consideration. The human remains in the loop as a legitimizing interface. That is not proof that the post-WWII economic order survives; it is evidence of the transition mechanism by which cognitive labor is hollowed out while human responsibility is retained as decorative liability.