The Unseeing State

On the US Government’s response to the AI revolution

Twice in the last 100 years the US Government mobilized the American nation at scale. The first occurred in the 1930s, under the administration of President Franklin Roosevelt, in response to the Great Depression. The second took place a decade later, in the 1940s, after the US entered World War II.

These were national mobilizations that reached into every corner of American life. Few lives were untouched. Each became a model — imperfect, but durable — of federal action under extreme pressure.

The costs, consequences, and benefits of these national mobilizations still echo across the generations. Academics may debate their precise influence. But it is more likely than not that the US, still wealthy and immensely powerful, has — however imperfectly — endured until 2026 because of those ambitious and far-sighted federal programs, not in spite of them.

A greater challenge now faces the US Government. The economic and national security implications may prove comparable in scale — though different in form — to those faced in 1933 and 1941. There is a global revolution looming. It is coming first to the US but in time will spread to every other nation-state. The revolution is ushering in a new age of thinking machines. The AI revolution, perhaps the final global revolution solely led by humans, has begun.

Some commentators dispute whether the revolution has already commenced. Discernible AGI, they note, has not yet been attained; AI platforms, LLMs most of all, are still unthinking and stilted technologies. AI, these voices say, is still little more than gussied up circuit boards. Job losses publicly linked to AI are often, once examined, traditional corporate restructuring and cost-cutting.

There is some truth to this skepticism. Asking a cutting-edge AI assistant from a leading frontier lab to tell a joke instantly reveals the limits of the billions spent on AI development. Even the most advanced LLM models still regularly “hallucinate” and perform soft tasks poorly. Many US companies are still hiring for roles supposedly under threat from AI. Current robots are barely capable of even the simplest manual tasks.

What the skeptics misjudge is not where AI is today — but how quickly it is moving. Foundational elements of AI are now improving logarithmically if not exponentially: the amount of computation used to train frontier AI models has doubled every six months since 2020; the computing power of the most advanced chips is doubling every 10 months; the cost to perform inference or run a model at a fixed price has been halving roughly every two months. The amount of data an AI model can consider at once has grown by 20x per year since 2023.

Two months ago, the first autonomous research AI tools, or agents, were launched to independently conduct scientific research. Not by a government bureaucracy. Not by a technology behemoth. Just one individual with a computer; the Karpathy Loop demonstrated how to conduct continuous AI training without human intervention. Agents are now the interface with the deeper layer of digital architecture, while manual coding is increasingly unnecessary. As agents grow in numbers and capability, the digital and the physical worlds will become more and more integrated across every facet of human experience.

The agentic AI expansion is a precursor to the arrival of AI-enabled humanoid robots, which will dramatically alter many fields of human activity. In the last year, factory manufacturing costs for humanoid robots reportedly dropped by 40%, enabling robotic companies to accelerate plans for their deployment. Tesla expects to manufacture 5,000 Optimus robots next year, and projects annual production in the millions within several years. At least half a dozen other US companies plan to launch humanoid robot production in the coming 12-24 months.

Nor is the impact limited to the civilian commercial sphere. Military tactics are already being transformed by the use of AI: in Ukraine today, a much smaller country is fighting, and in certain respects outpacing, the 2^nd largest military in the world because of how it has adopted AI-enabled weapon systems. AI-enabled theater and battlefield intelligence, targeting, and coordination has already been incorporated at scale by the US military. The use of autonomous robots to perform specific military missions will soon become common, and within a matter of months prototypes of militarized humanoid robots are expected; two humanoid robots designed by a California-based robotics company are already operating in Eastern Ukraine and are being tested in combat environments.

This rate of development, and the ensuing economic disruption and social dislocation, is uncontrolled. It is also irreversible. Like a thousand tiny glacial streams descending to an alluvial plain, every day the torrent they feed grows stronger. The only limiting factors are hardware availability and energy.

The New Deal and the war mobilization programs of World War II unfolded across years of concerned government programs. But the AI revolution is happening immeasurably faster. The question is whether institutions can keep pace with it.

In 2025 the US Government promulgated an initial policy response: the American AI Action Plan (AAIP). The plan includes 90 policy recommendations focused on accelerating AI innovation, building AI infrastructure and promoting AI-related security. Some initiatives have commenced, including expedited permitting for data centers and new procurement rules to ensure unbiased AI models.

However well intended, the AAIP is inherently limited. Because it is an Executive Order rather than a formal federal law it is open to challenge and may be reversed in the next administration. Moreover, no dedicated budget was attached to the AAIP’s implementation.

Alongside the AAIP is the National Policy Framework for Artificial Intelligence (NPFAI), announced in March 2026. Its chief aim is to ensure a coordinated national policy of a “minimally burdensome national standard”, thus avoiding a possible “patchwork” of different or conflicting state-based AI standards and rules. Its policy recommendations encompass privacy tools, streamlined permitting for AI data centers while protecting ratepayers from electricity cost increases, boosting national security reviews of frontier models, enabling federal data to be AI-compatible, and integrating AI training into existing educational programs.

The NPFAI suffers from similar limitations to the AAIP. Most crucially, it is a set of policy recommendations without a budget. Second, it is intended to inform Congressional action drafting relevant legislation — but there are already hundreds of draft AI-related bills in Congress, and with mid-term elections looming the chances of any significant AI legislation being quickly adopted is remote.

Which is to say that the US Government, in the face of the rapidly advancing AI revolution, has to date mostly issued budget-less policy directives, and there is little chance of substantial new AI legislation for at least a year.

The inadequacy of the US Government’s response becomes clearer when considering where it is most needed today:

• Research Prioritization: while AI has near limitless potential to transform human knowledge, and there are already countless areas of research leveraging AI tools, the US federal government has an important role in focusing and coordinating research initiatives.
• Systems Evaluation: setting standards for AI models and technologies is already a focal concern and will become more so. Evaluative tools such as the NIST AI Risk Management Framework are still nascent and under-utilized.
• The Economy of Machines: the economic activity undertaken by AI-enabled agents is set to grow rapidly, and in time may rival or exceed the traditional human-based economy. The US Government must decide how to regulate this new economy, and whether a Universal Basic Income should be implemented. 
• Education: in an age of machine super intelligence, the focus and methods used for educating people will necessarily shift. As Proteus I noted, the established pathways from university to the professions are already fraying.

This is not simply a question of speed or policy design. The US Government is structurally ill-equipped to manage this revolution. The core US administrative architecture remains, at base, an early 20^th century design: siloed departments, committee-bound legislation, incremental appropriations — all premised on a slower, less interdependent pre-machine age world.

Moreover, invoking the full power of the US bureaucracy brings risks. Large scale top-down government programs have inherent limitations beyond cost and implementation complexity. As James Scott documented in Seeing Like a State: How Certain Schemes to Improve the Human Condition Have Failed, even in advanced systems like the US, central planners are notoriously bad at addressing the full complexity of evolving economic systems and demographics.

Scott observed that states seek to make society “legible”— to render it in forms that can be measured, categorized, and administered. This legibility is a source of power, but also of limitation. What cannot be easily seen within the state’s existing bureaucratic framework is often misunderstood or ignored. Complex, adaptive realities are simplified into forms that can be managed.

AI resists this simplification.

AI does not present itself as a single problem to be solved, nor a single sector to be regulated. It is a general-purpose capability that alters the conditions under which nearly every legacy human system operates. It is, in Scott’s terms, partially illegible — not because it is unknowable, but because it does not conform to the categories through which the existing bureaucratic structure perceives and orders policies, legislation, and regulations.

The chief shortcoming of the US Government's response to date is not a failure to act, but a failure to see.

Historically the US Government has demonstrated an ability — however uneven — to recognize and respond to large-scale disruptions. During the Great Depression, the initial response was fragmented and uncertain. But within a relatively short span, the federal government reoriented itself around the problem of mass unemployment and systemic collapse. The New Deal did not simply redistribute resources; it constructed capacity. The Tennessee Valley Authority electrified entire regions. The Works Progress Administration and Civilian Conservation Corps employed millions, building roads, dams, bridges, parks — an infrastructure that would shape the country for the next century. Labor was not merely supported. It was redirected, absorbed, given form and purpose within a transformed economic landscape.

A similar dynamic appeared during World War II, at an even larger scale. The US Government did not merely increase production; it reorganized itself for it. Under the direction of the Office of War Mobilization, the economy and labor force were systematically channeled to respond to an urgent cause. In one example among many, shipyards in places like Richmond, California — one of 18 related shipyards — across the bay from San Francisco, produced Liberty ships at unprecedented speed. That production drew hundreds of thousands of workers, including a large migration of black Americans from the South. They came for better jobs, higher pay and hard skills. And in doing so the demographic and economic character of San Francisco and its surrounding cities was profoundly reshaped. War mobilization not only resulted in a military benefit. It reconfigured US society and gave citizens opportunities and meaning. In a certain way it structured and dignified their lives.

What is striking in both cases is not only the scale of mobilization, but its concreteness. The state created structures that translated disruption into meaningful activity — bridges built, ships launched, regions electrified, workers employed. Dislocation was redirected into production. An entire generation of Americans found meaningful work.

The government’s response was imperfect, often contradictory, and in places inefficient. But it was not based on aspirational policy recommendations. It reflected a recognition that the existing order could not simply be preserved; it had to be reshaped. That was both the problem, and the opportunity.

How must the US Government respond in full to the AI revolution? The techno-optimists, led by figures like venture capitalist Marc Andreessen, unwaveringly urge minimal government interference while forecasting a new golden age of wealth and innovation. On the other side are the dark imaginings of techno-skeptics like Paul Kingsnorth, who in his Against the Machine: On the Unmaking of Humanity, envisions AI as a modern-day doomsday device.

Between these extremes are sage voices that speak across eras. Hannah Arendt, in Men in Dark Times, emphasizes the search for illumination in periods of moral crisis. Ivan Illych, in Tools for Conviviality, reimagines education under pressure from technological change. Lewis Mumford, in Myth of the Machine, insists that technology should remain subordinate to human ends.

Weaving these threads of humanistic thinking together, the US Government can enjoin the fragmentation ensuing from the AI revolution with a revitalized notion of the role of government and civic life. Instead of a Tennessee Valley Authority, a new National Service program could be launched, developing national infrastructure and enhancing national security while simultaneously affording Americans a shared purpose and communal base of experience.

For now however, the pattern holds: an AI revolution of increasing velocity, and a state that does not yet see what it confronts. The signals accumulate. Between them, the gap widens. The structure of the state has not changed. The world it governs has.

A generation ago, a US Senator, decorated war hero, and Presidential candidate asked, “where is the outrage?” The question he might ask today is “where is the urgency?” As the centripetal force of AI strengthens and surges, a nation of 343 million souls waits.

"Nothing is left to chance in the ordering of the world; though all things are disposed according to order, they appear confused to those who cannot perceive it."
—Boethius, The Consolation of Philosophy