Vertesi, J., danah boyd, Taylor, A., & Shestakofsky, B. (2026). Reckoning with the political economy of AI: Avoiding decoys in pursuit of accountability. arXiv [cs.CY].
Summary
Vertesi, boyd, Taylor, and Shestakofsky argue that AI is best understood not as a technological artifact but as a “Project of AI”—a world-building, political-economic endeavor through which financiers and corporations assemble networks of capital, infrastructure, and labor to restructure markets in their favor. The paper’s central provocation is that many critiques of AI, even well-intentioned ones, function as decoys: they appear to challenge AI while actually stabilizing and co-constructing the political economy that produces it. Aimed at the FAccT community, the paper offers a corrective framework that pulls accountability research away from algorithmic artifacts and toward the financial, infrastructural, and network-power relations that make “AI” possible.
Key Contributions
- Introduces the “Project of AI” as a conceptual frame foregrounding world-building and political economy over technical artifact.
- Develops a typology of five decoys—ontological, inevitability, disruption, safety, and regulatory—that diagnose how critique gets co-opted.
- Bridges communication studies (Castells’ network-making power), STS/ANT, and economic sociology (Fligstein, Beckert) into an integrated political-economy lens for FAccT.
- Proposes four anchoring frames for future accountability research: material sites of network assembly, financing as technopolitical work, global flows over bounded objects, and resisting siloed/social solutionism.
- Reframes transparency and fairness goals away from models and toward the networks and capital relations producing them.
Methods
Conceptual and theoretical synthesis rather than empirical study. The authors weave Castells on network power, Actor-Network Theory and material political economy from STS, and economic sociology of market-making, valuation, and uncertainty. They illustrate the decoy typology through contemporary cases: OpenAI, Anthropic, Microsoft, DeepSeek, Altman’s Senate testimony, the EU AI Act, and the DMA.
Findings
- Ontological decoy: Debates over “what AI really is” let brokers exploit strategic ambiguity to capture funding and reshape markets.
- Inevitability decoy: AGI timelines and geopolitical race rhetoric normalize harms, mobilize resources, and foreclose alternatives—even enrolling participatory design into legitimation.
- Disruption decoy: Framing change as workforce automation obscures corporate restructuring, offshoring, and infrastructural concentration among elites.
- Safety decoy: Existential-risk framings deflect from immediate harms and position AI firms as sole arbiters of risk.
- Regulatory decoy: Incumbents court regulation to entrench themselves; market-structure rules like the EU DMA more directly target network power than content-focused AI regulation.
- Decoys enable “network shape-shifting,” allowing capital, data, and ghost-work labor flows to evade oversight by displacing accountability across nodes.
Connections
This paper sits within a growing critical literature on AI’s discursive and political-economic scaffolding, complementing work on hype, inevitability narratives, and imaginaries such as Dodds2026-df, Stanusch2026-ec, Weinbrand2026-sf, and Wang2025-zy. Where those tend to interrogate specific rhetorical or imaginary formations, this paper offers a higher-order diagnostic—asking how even critique itself can be enrolled into stabilizing the industry it targets.
Podcast
A research-radio episode discusses this paper: 🎧 MP3 · Spotify · Apple Podcasts