InnovateUS 
One of the biggest risks of AI adoption across society goes undebated: the quiet acceptance of the prediction paradigm as a logic for how we organize collective life.
In my book Predicted: How AI is Restructuring Social Life, I describe the prediction paradigm as the sociopolitical shift in which prediction, central to how AI technically functions, becomes the dominant organizing principle for how society produces knowledge, makes decisions, and structures social life.
AI systems are, at their core, probability calculators.
This is true even for generative and agentic AI. These systems produce outputs by continuously estimating the likelihood of patterns based on learned statistical relationships in data. What makes this a paradigm, rather than just a technique, is the scope of its reach.
AI as Social Infrastructure
AI systems have become social infrastructure: they mediate social relations both formal and intimate. They curate media content, predict student performance, forecast patient outcomes, screen job candidates, and increasingly structure how we draft personal messages, navigate relationships, and reflect on our own experience.
In each case, a probabilistic estimate about a likely future is embedded into how institutions allocate opportunity, manage access, and make decisions about people.
This is undemocratic in a specific sense. Elections, deliberations, and collective choices only make sense if they are uncertain, if the future is open.
The prediction paradigm encodes the premise that the future is already here, legible in the patterns of the past, and best governed by whoever controls the data and the models.
The prediction paradigm encodes the opposite premise: that the future is already here, legible in the patterns of the past, and best governed by whoever controls the data and the models, typically private entities rather than elected representatives.
When Prediction Shapes the Future Itself
The rapid rise of prediction markets illustrates this logic at its sharpest. These platforms formalize and incentivize probabilistic speculation on the outcomes of future events, creating perverse incentives to influence what is being predicted.
In April 2026, a U.S. Army Special Forces soldier was charged with using classified knowledge of a covert raid on Venezuelan President Nicolás Maduro to place bets on Polymarket, netting around $400,000 in profits. It is the first insider trading case ever brought on a prediction market.
The prediction and the outcome were no longer independent: having a financial stake in the future creates an incentive to act on it. Without governance, this is where the logic of prediction leads.
AI Already Functions Like Infrastructure
This dynamic—prediction expanding beyond its useful bounds, privately controlled, insufficiently governed—is not only a feature of prediction markets. It describes AI’s social and political impact more broadly: as infrastructure.
AI systems are already infrastructure in the most practical sense. You cannot participate in modern institutional life without being processed by them. To apply for a job, access public benefits, get a loan, enroll in a school, or navigate a healthcare system is increasingly to be subject to AI prediction.
Like roads or electrical grids, AI systems determine who gets access to what, and on what terms. They are almost entirely privately owned and governed as products rather than as public assets.
Like roads or electrical grids, these systems determine who gets access to what, and on what terms. Unlike roads or electrical grids, they are almost entirely privately owned and governed as products rather than as public assets.
That mismatch is the central problem, and it is one we have faced before.
The Railroad Playbook
The good news is that policymakers do not need a new approach to respond. There is already a precedent for what happens when a privately controlled technology becomes foundational to public life.
In the 19th century, railroads were built under conditions that closely resemble today’s AI boom: speculative capital, public subsidies, and minimal oversight. Like AI, railroads quickly became essential infrastructure.
They shaped which regions prospered, how goods moved, and who had access to markets. The infrastructure was privately owned and outpaced its governance. Monopolistic pricing and discriminatory access followed.
What Happened When the Bubble Burst
When the railroad bubble burst in 1873, the damage extended far beyond investors. Risks had been socialized while rewards remained privatized. This is a dynamic now plainly visible in the AI industry, where public investments in research, data, and infrastructure underwrite private capture of the returns.
The policy response to the 1873 bubble burst was not to dismantle railroads, but to govern them.
The policy response was not to dismantle railroads, but to govern them. The 1887 Interstate Commerce Act classified railroads as infrastructure subject to public oversight, established common carrier obligations, and created the first federal regulatory commission.
The governing logic was non-discrimination and long-term financial stability in service of the public. The goal was not to suppress the industry but to reorient it.
Governing AI as Infrastructure
The AI industry is exhibiting familiar signs of the same dynamic. Circular dealmaking among a handful of companies—where investment capital flows between AI model makers, chip manufacturers, and cloud providers in a closed loop without generating measurable evidence of economic value—inflates valuations and creates systemic dependencies.
If this bubble bursts without governance in place, the consequences will not be contained to investors. The question, as with railroads, is whether intervention happens before the correction or after it, when options are far more limited, and costs fall on already vulnerable populations.
From Utility to Practice
Applying the railroad playbook to AI begins with classification. AI can be a utility: governed, stable, and oriented toward innovation and collective benefit.
Large model providers and datacenter operators should be combined and treated as common carriers or natural monopolies entrusted with public responsibilities.
This is not a radical proposition.
Telecommunications companies have operated under exactly this dual classification for decades, and the logic is the same: when a single infrastructure becomes essential to economic and social participation, the public interest requires that access be guaranteed, stable, and non-discriminatory.
What Governance Could Look Like
Governing AI as infrastructure requires action at both the federal and state level, and the tools to do it already exist.
Governing AI as infrastructure requires action at both the federal and state level, and the tools to do it already exist.
Federal level
At the federal level, the most important step is pricing accountability.
The Commodity Futures Trading Commission (CFTC) or a designated federal agency should establish baseline pricing standards for GPU compute hours and API access, modeled on how utility commissions set electricity rates. AI companies should be required to file cost structures publicly and justify significant pricing changes.
Common carrier obligations should also extend to AI model providers and cloud operators. This would prohibit preferential pricing or access terms for commercial partners over public institutions.
Federal procurement contracts should additionally require vendors to demonstrate minimum labor protection standards across their data labeling supply chains, including for offshore contractors.
State level
At the state level, public utility commissions should classify datacenter operators as public utilities. This could enable oversight of energy consumption, water use, and local grid impact, as well as build-out of grid infrastructure. State contracting law should also build in AI procurement standards as a baseline condition.
Vendors should be required to disclose training data, error rates across demographic groups, and contestability mechanisms before any contract can be awarded.
States should also establish AI ombudsman offices that are empowered to investigate complaints from individuals denied access to public services by AI systems and require agencies to respond.
The Inflection Point
The AI sector is at a comparable inflection point to the railroads in the 1870s: infrastructure expanding ahead of governance, speculative capital concentrating in a handful of entities, and public costs accumulating without public oversight.
The regulatory tools required are not novel. They have governed transportation, energy, and telecommunications infrastructure for over a century. What remains variable is timing.
Acting before the correction allows policymakers to shape AI markets while they are still fluid. Waiting means governing in crisis, when the damage is done, and the choices are fewer.