Complementarity, Augmentation, or Substitutivity? The Impact of Generative Artificial Intelligence on the U.S. Federal Workforce

URL SCAN: Complementarity, Augmentation, or Substitutivity? The Impact of Generative Artificial Intelligence on the U.S. Federal Workforce
FIRST LINE: # Computer Science > Computers and Society

The Dissection

This paper converts a dynamic labor-market shock into a static competency inventory. It asks where generative AI can assist, transform, or partially automate documented federal occupational competencies, then treats those scores as evidence about the fate of occupations.

That is a narrower question than the title implies. The method evaluates current job descriptions and curated evidence through a RAG-supported language-model assessment. It does not model organizational redesign, staffing budgets, hiring freezes, attrition, managerial substitution, agentic AI systems, or the number of workers required to produce a given level of output.

The paper’s central maneuver is semantic: if AI does not perform every task in a position, the position is classified as surviving. That is not how labor substitution works. A position can become economically unnecessary while retaining human tasks if one AI-assisted worker can produce what several workers previously produced.

The Core Fallacy

The paper conflates task-level complementarity with employment-level complementarity.

AI can complement a worker on individual tasks while substituting for workers in aggregate. Augmentation is often the bridge to substitution: once AI raises the output of each remaining employee, the institution needs fewer employees. Full automation of an entire position is not required.

“Partial automation” is therefore not evidence against displacement. It is frequently the mechanism of displacement. Federal agencies can reduce labor through attrition, hiring restrictions, consolidation, reassignment, contractor reduction, and grade compression without formally declaring that an occupation has been eliminated.

The study also treats complementarity, augmentation, and substitutivity as if they were stable categories. Under rapid capability improvement, they are stages in a transition. Today’s assistant becomes tomorrow’s workflow owner, while the human remains only as a legal signature, exception handler, or accountability buffer.

Hidden Assumptions

• Current competency documentation accurately represents future production requirements.
• A task remaining human-performed implies the position remains necessary.
• Productivity gains will expand federal demand enough to preserve headcount.
• Agencies will retain existing staffing ratios despite budget and political pressure.
• Human-in-the-loop rules require one human per workflow rather than one human supervising many AI systems.
• Federal legal and institutional constraints permanently protect labor rather than merely delaying adjustment.
• Curated evidence makes LLM scoring economically valid, not merely more grounded and internally consistent.
• Generative AI is evaluated as a fixed tool rather than a rapidly improving component of automated workflows.
• Reskilling leads to durable employment instead of creating a smaller, more selective workforce.
• Near-term complementarity says anything decisive about the medium- or long-term labor equilibrium.

The abstract provides no causal test of these assumptions. It reports classifications, not observed displacement outcomes or a counterfactual staffing path.

Social Function

The paper is a partial truth functioning as transition management and ideological anesthetic.

Its limited truth is real: early federal deployment will often augment existing workers, and legal, security, procurement, and accountability structures will slow immediate substitution. But the conclusion converts that lag into reassurance. It gives institutions an evidence-flavored reason to prefer reskilling plans and human-AI collaboration while postponing the harder question: how many workers remain necessary after the collaboration works.

The RAG framework adds methodological prestige, but grounding an assessment in current documents does not solve the strategic problem. It may produce a more disciplined map of the existing bureaucracy while leaving the future production function completely unmodeled.

The Verdict

As a near-term inventory of where AI currently touches federal competencies, the study may be useful. As evidence that generative AI will preserve federal employment or disprove structural displacement, it is inadequate.

It does not test the Discontinuity Thesis’s decisive chain: durable AI superiority, institutional inability to preserve human-only economic domains, and collapse of mass productive participation. It measures the first visible phase of the transition and mistakes it for the destination.

Under that lens, “complementarity” is not a defense. It is often the loading phase of substitution: AI makes each surviving worker more productive until the institution needs fewer survivors. Federal bureaucracy supplies lag, not immunity. The paper describes the hospice interval and labels it recovery.