Power, Agency, and the New AI Reckoning
The conversation around artificial intelligence has, until recently, been dominated by narratives of technical prowess and breakneck innovation. But a recent high-profile panel—featuring the incisive sociologist Tressie McMillan Cottom—has thrown a sharp new light on the debate, shifting its center of gravity from questions of feasibility to those of power, distribution, and collective agency. Cottom’s intervention is timely: as the AI sector barrels forward, her thesis reframes the stakes, urging us to interrogate not just what AI can do, but who it ultimately serves.
The Mirage of Inevitable Progress
The prevailing mythos—what Cottom calls “AI-inevitability”—has long suggested that the march of generative models is both unstoppable and universally beneficial. Yet, cracks are appearing in this façade. High-profile errors by large language models, such as ChatGPT’s hallucinations and missteps, are no longer outliers but symptoms of a deeper stagnation. The industry’s faith in ever-larger parameter counts is yielding diminishing returns, with accuracy improvements plateauing even as computational and financial costs soar.
- Stagnation Signals: Marginal gains from scaling are waning, exposing the limitations of brute-force approaches.
- Infrastructure Asymmetry: Access to advanced GPUs and proprietary datasets is increasingly restricted to a handful of hyperscalers, hardening barriers to entry and consolidating power among a “computational elite.”
- Safety vs. Speed: The rush to deploy models has collided with legal and reputational risks—copyright lawsuits and publicized mistakes now serve as cautionary tales, forcing a reckoning with the limits of velocity.
Cottom’s critique is clear: the narrative of inevitable AI progress is less a neutral observation than a tool wielded by entrenched interests to cement their dominance. The future, she argues, is not preordained by technical capability, but by the choices—collective or otherwise—of those who wield and regulate these systems.
Economic Concentration and the Stakes of Governance
The economics of AI are reinforcing old patterns of concentration. Training frontier models now requires investments that only the largest firms can muster, pushing smaller players into dependency on “model as a service” offerings and deepening platform lock-in. This capital intensity, coupled with the complexity of regulatory regimes (from the EU AI Act to China’s algorithm registry), creates a landscape where only incumbents can afford to optimize across jurisdictions.
- Labor Recomposition: Contrary to apocalyptic forecasts of mass unemployment, generative AI’s near-term impact is subtler—augmenting tasks rather than replacing jobs outright. Yet, the promise of upskilling and workflow redesign remains underfunded, outpaced by boardroom expectations.
- Regulatory Arbitrage: The patchwork of global rules rewards those with the resources to navigate it, further entrenching incumbency.
Meanwhile, the risks of “Digital Redlining 2.0” loom large. If AI is optimized for profitability rather than equity, marginalized communities face algorithmic neglect or, worse, surveillance-driven exploitation. Policymakers are responding: the era of “light-touch” principles is ending, replaced by binding obligations around impact assessments, auditability, and data provenance. Institutional investors, too, are broadening their definition of fiduciary duty to include algorithmic fairness and workforce resilience.
Strategic Levers for a New AI Era
For decision-makers, the implications are profound. The locus of advantage is shifting from raw scale to governance, trust, and the capacity to anticipate second-order effects.
Key Imperatives:
- Rebalance R&D: Move beyond brute-force scaling; invest in interpretability, domain-specific models, and energy-efficient architectures.
- Build Multi-Disciplinary Governance: Integrate sociologists, ethicists, and labor economists into product councils to preempt societal risks.
- Diversify Compute Supply Chains: Hedge against GPU concentration through alliances with emerging chip designers and foster workload portability.
- Monetize Trust: Transparent data lineage, voluntary audits, and participatory design with affected communities will become durable differentiators.
- Plan for Labor Partnership: Invest in large-scale reskilling and human-in-the-loop models; productivity gains depend on complementary human capital, not its displacement.
The Contours of the Coming Decade
The AI industry stands at a crossroads reminiscent of the post-dot-com “reality check.” Funding remains robust, but the premium is shifting to demonstrable ROI and governance maturity. Regulatory harmonization is on the horizon, likely coalescing around auditability and provenance, catalyzing new markets for algorithmic assurance. Technological differentiation will rely less on sheer model size and more on specialized data, fine-tuning, and energy efficiency—opening doors for nimble innovators.
Ultimately, the decisive question is no longer “Can we build larger models?” but “Who defines the purpose, access, and guardrails of the models we build?” Leadership teams that embrace this shift—embedding agency and inclusion as design primitives—will shape not just their market share, but the very social contract that underpins the next era of AI. In this landscape, the refusal to accept deterministic futures may prove to be the most powerful innovation of all.
By
By
By
By
By
