The relationship between professional and general ethics in generative AI

URL SCAN: The relationship between professional and general ethics in generative AI
FIRST LINE: # Computer Science > Computers and Society

The Dissection

This is a governance-and-measurement paper disguised as a solution to an alignment problem. It identifies a real mismatch: generic value lists do not determine what legal, medical, or translation systems should do when professional duties collide. Its proposed move is to model professional ethics as hierarchies, observe models’ situational priorities, and intervene to change them.

Under the Discontinuity Thesis, that is ethics middleware. It calibrates machine behavior while leaving untouched who owns the machine, who controls deployment, and whether the profession retains an economically necessary human role.

The Core Fallacy

The central error is level confusion: treating normative priority selection inside AI outputs as if it were control over the social system producing those outputs. P1–P3 concern cost, coordination, and labor displacement. A model that better balances confidentiality, accuracy, duty, and public interest can still automate the professional function, transfer authority to its controller, and remove the wage-bearing role entirely.

The abstract also reifies a model’s “favored ethic.” A recurring output pattern is not a moral commitment; it is the product of training data, post-training, system rules, tool access, operator incentives, and liability constraints. Changing that surface may alter behavior, but it does not alter the ownership structure or competitive pressure selecting the deployment.

Hidden Assumptions

• Professional domains will remain stable enough for their ethical hierarchies to matter.
• Professions will retain enforcement power after AI absorbs their core cognitive tasks.
• A model has a coherent, discoverable ethic rather than context-sensitive policy behavior.
• Similar-scenario testing reveals stable preferences instead of prompt, policy, data, or interface artifacts.
• Whoever intervenes in the model has legitimate authority to decide which professional ethic should prevail.
• Better ethical calibration produces better social outcomes, rather than merely making automation more acceptable and legally defensible.
• General and professional ethics can be separated from the ownership and incentive structures governing deployment.
• Subjectivity is a manageable caveat, not evidence that the intervention itself is a political choice.

Social Function

Classification: partial truth, transition management, and prestige signaling, with a secondary function as ideological anesthetic.

The partial truth is substantial: domain-specific duties really do conflict, and generic alignment language is too blunt for practice. The anesthetic enters when this conflict is framed as the central problem. Attention shifts from productive-participation collapse to finer tuning of machine conduct. Displacement is made to look governable, professional, and procedurally respectable while control of AI capital remains outside the frame.

The Verdict

Useful for auditing domain-specific model behavior. Strategically inadequate as an account of AI’s systemic trajectory. The paper maps the ethical rules of professions whose economic base is being hollowed out, then proposes to tune the replacement machinery. It can improve the handoff; it cannot stop it. Under the Discontinuity Thesis, professional ethics are becoming configurable compatibility layers around AI capital, not a defense of mass productive participation.