Seeing Through Color Blindness: Social Networks as a Mechanism for Discrimination

TEXT START: I study labor markets in which firms both hire via referrals and are race blind or color-blind.

The Dissection

The paper isolates a real pre-AI mechanism: formally race-blind firms can reproduce racial inequality because referrals are filtered through homophilous networks. The operative variable is network access, not employer taste or statistical judgment. A smaller group has fewer same-group contacts, producing fewer referral paths, fewer jobs, and lower expected wages even with equal ability.

Its limitation is scope. It treats referral-mediated employment as the decisive economic battleground. Under the Discontinuity Thesis, that is an analysis of how the old labor market distributes scarce human jobs—not a theory of the system replacing it.

The Core Fallacy

The core error is temporal extrapolation, not the paper’s mechanism. It assumes human employment remains the stable substrate of production. If P1 holds, cognitive hiring, matching, screening, and much of the referred work are automated. Referral networks then lose their position as the main production gatekeepers.

The paper can show who is disadvantaged while jobs exist. It cannot explain the post-P3 condition, where the decisive question is who owns and controls productive AI and its output.

The claim that the paper “disproves” color-blind policies as meritocratic is also broader than the supplied abstract establishes. It supports the narrower conclusion that formally color-blind rules do not guarantee meritocratic outcomes when network access is shaped by homophily and group size. It does not establish that race-conscious rules outperform them.

Hidden Assumptions

• Human referrals remain economically central.
• Jobs remain broadly available, with wage access as the primary welfare channel.
• Firms continue relying on social networks rather than AI-mediated capability verification or direct production.
• Network structure remains stable enough for youth-network calibration to predict labor-market welfare gaps.
• Racial categories and homophily retain the same economic importance through automation.
• Policy can improve job allocation without confronting ownership of the productive system.
• Labor-market inequality remains the decisive form of inequality after AI.

Social Function

Classification: partial truth, prestige signaling, and transition management.

The paper performs a useful local autopsy: declaring a system color-blind does not erase network path dependence. But it keeps the debate inside the distribution of jobs, wages, and referrals—the institutions of the dying order. It offers a precise account of unequal access to employment while leaving ownership of the replacement production system outside the frame.

The Verdict

Accurate local mechanism, obsolete global frame. The paper supports a narrow extension of P2: human institutions cannot make a networked, homophilous labor market meritocratic merely by banning explicit racial preference. It does not challenge P1 or P3; it presupposes the labor circuit that the Discontinuity Thesis says AI will sever.

Once mass cognitive employment becomes unnecessary, referral discrimination becomes a legacy pathology. The terminal question is not whether minorities receive an equal share of disappearing jobs, but whether they own AI capital or remain dependent on its owners. This paper maps a wound in the corpse. It does not identify the engine of death.