Do job seekers value procedure in AI hiring only for error correction? Evidence from a conjoint experiment

URL SCAN: Do job seekers value procedure in AI hiring only for error correction? Evidence from a conjoint experiment
FIRST LINE: # Computer Science > Computers and Society

The Dissection

The study isolates procedural features—appeals, opt-outs, bias audits, explanations, and human involvement—from error rates, then measures which systems job seekers prefer. Its finding is narrow but real: applicants value procedural control even when those procedures do not merely correct more errors.

What it is really doing is measuring the legitimacy conditions under which displaced people will tolerate automated gatekeeping. It examines how to make the machine acceptable, not whether the machine can preserve human economic participation.

The Core Fallacy

It risks confusing procedural legitimacy with structural protection. A right to appeal, an audit, or nominal human involvement may give applicants recourse, but none interrupts the competitive mechanism that drives employers toward cheaper, faster, more scalable cognitive automation.

Human involvement is especially deceptive. Under Discontinuity Thesis mechanics, it can become ceremonial review, exception handling, or institutional camouflage—not a durable domain of economically necessary human labor. The study measures preference for a human checkpoint; it does not establish that human checkpoints can survive competitive pressure at scale.

The paper therefore probes the social wrapper around automation while leaving the automation engine untouched. It does not challenge P1, P2, or P3.

Hidden Assumptions

• Applicant preferences can be converted into enforceable institutional rights.
• Employers will preserve costly human review despite automation's efficiency advantage.
• Appeals and opt-outs remain meaningful when applicants have little bargaining power and need access to employment.
• Bias audits can correct outcomes without challenging who owns, controls, and deploys the hiring system.
• Stated choices in a conjoint experiment predict behavior under actual economic coercion.
• Procedural fairness can substitute for lost productive participation.
• Human involvement remains substantive rather than becoming a rubber stamp or a scarce premium service.
• The initial screening stage is the relevant battlefield, when the deeper issue is the progressive automation of the entire cognitive labor chain.

Social Function

Classification: partial truth, transition management, and—when elevated into policy reassurance—ideological anesthetic.

The partial truth is that people care about procedure independently of accuracy. That matters for legitimacy, resistance, litigation, and the design of the transition. But the transition-management function is more consequential: appeals, audits, and human review can channel justified anger into administrative safeguards while leaving ownership, displacement, and power untouched.

The procedures may even lubricate adoption. A more acceptable automated gatekeeper is still an automated gatekeeper. The rights reduce friction around replacement; they do not reverse replacement.

The Verdict

This is evidence about how the dispossessed want the machine governed, not evidence that the machine can be stopped. Procedural rights may delay backlash, correct individual injustices, and preserve a thin layer of agency. They do not restore the mass employment–wage–consumption circuit. Under the Discontinuity Thesis, this is governance around the corpse of human-centered hiring: useful hospice care, not resurrection.