Better Together: Quantifying the Benefits of AI-Assisted Recruitment

TEXT START: Hiring algorithms have mostly scored the materials recruiters already see.

The Dissection

The paper documents a more invasive form of labor-market automation: AI is no longer merely ranking resumes; it conducts scalable interviews and manufactures new screening data. Its most important result is not that hiring improves. It is that firms can transfer more of the selection workload—and its cost—to applicants while increasing the throughput and precision of exclusion.

The “better together” framing is temporary. Human recruiters remain in the loop because the experiment studies an incumbent pipeline, not because human participation is structurally secure. The AI interview expands the gate while making the gate cheaper for firms and more burdensome for candidates.

The Core Fallacy

The paper commits a system-level category error: it treats better prediction inside a hiring funnel as evidence that the funnel—and human employment within it—becomes more valuable.

The reported gains show that AI can extract useful information where resumes fail. They do not show that AI creates more jobs, raises aggregate labor demand, or preserves human productive participation. Under the Discontinuity Thesis, this is precisely the dangerous pattern: AI improves allocation while eroding the wage-and-employment circuit that makes allocation socially meaningful.

The 75 percent noncompletion rate is not an incidental cost. It is the mechanism. Screening is being converted into applicant-supplied unpaid labor, with firms receiving more information and candidates absorbing more risk, time, and attrition.

Hidden Assumptions

• Hiring volume and human-led final interviews remain sufficiently large to matter.
• Improved AUC translates into better employment outcomes rather than more efficient rejection.
• Recruiters retain durable authority instead of becoming verification staff for machine-generated judgments.
• Applicants can repeatedly absorb unpaid AI interviews across competing platforms.
• Completion is a reliable signal of motivation rather than wealth, time availability, disability access, technical access, or desperation.
• Candidates cannot cheaply game, automate, or strategically adapt to the interview.
• Platform results generalize across occupations, firms, and economic conditions.
• Better matching produces broad social gains rather than concentrating opportunity among candidates who can survive increasingly expensive gates.

Social Function

Partial truth functioning as transition management and elite self-exoneration. The empirical result is real: AI can improve screening, especially for junior candidates whose resumes carry weak signals. But the paper packages a labor-displacing capability as benign augmentation and treats the transfer of screening costs to applicants as an efficiency finding.

Its ideological service is clear: firms can claim they are improving fairness and matching while quietly converting candidates into unpaid data providers. The human recruiter remains visible just long enough to make the system appear human-controlled.

The Verdict

This is not evidence that AI makes workers safer. It is evidence that AI makes the labor market’s sorting apparatus more powerful, cheaper for firms, and harsher for applicants. The paper captures an early transition stage: human hiring survives as a supervisory shell around automated information extraction. Once the system learns to conduct, score, verify, and rank the interview itself, “better together” becomes “human review optional.”