Must Democracy Become Ceremonial to Beat China?

The AI race may reward democracies that preserve elections while insulating the systems that actually decide.

No one will propose ceremonial democracy by name. It will arrive as a series of requests, each defensible on its own terms and each defended as the price of keeping pace with China. Guarantee investors that the compute corridor survives the next election. Shield the standards body from partisan pressure. Lock the model vendor into a multiyear agreement. Classify the frontier evaluations.

Grant enough of them and the elections of the 2030s could look entirely normal. Candidates would still argue about taxes, immigration, and the character of the nation, and one party would still lose power to another. The models used by federal agencies would stay inside the same procurement vehicles, their evaluation thresholds embedded in technical standards. Long-term energy contracts would still decide where the national compute base can expand, and security classifications would still decide who receives frontier access. Automated eligibility, enforcement, and allocation systems would keep running under configurations no incoming administration could readily inspect, much less replace.

The election would be real, and so would the transfer of office. The most consequential machinery of government would be protected from both.

Critics would call this democratic decline. Its defenders would call it strategy. The United States and its allies face a Chinese state able to coordinate infrastructure, industry, standards, data, and deployment over longer political intervals. China’s “AI Plus” program is framed as a society-wide deployment effort spanning science, industry, consumption, public welfare, governance, and international cooperation. Official targets call for new-generation intelligent terminals and agents to exceed 70 percent penetration by 2027 and 90 percent by 2030. The 2026 government work report expanded the push, and the draft Fifteenth Five-Year Plan targets multimodal systems, agents, embodied AI, and swarm intelligence.[1][2]

Democracies have made durable commitments before; dams and treaties routinely outlast the governments that approve them. The harder question is whether each electoral cycle can still reopen the rules these systems keep writing once they are running. Governments that decide it cannot will find ceremonial democracy attractive. Elections continue because legitimacy still requires them. Elected officials continue because the state still needs a recognizable author. Public argument continues because consent needs a visible theater. The systems that actually coordinate the AI transition, meanwhile, move beyond ordinary political interruption.

The ballot survives. The interruption disappears.

The Case for Ceremony

The argument for this arrangement is stronger than democratic instinct allows because AI competition extends far beyond building a better model. It is a race to coordinate the systems around the model: power, data centers, chips, cloud access, procurement, standards, security permissions, and the organizations able to integrate them. These assets operate on different clocks. Elections turn every two or four years. Transmission lines, fabs, and nuclear plants take much longer. Standards can become embedded across an economy before most legislators know they exist.

No electorate can vote directly on every model-evaluation protocol or acquisition clause. Members of Congress cannot personally determine which interoperability standard will preserve substitutability across federal systems. A president cannot review the millions of classifications, thresholds, and routing decisions an AI-administered state will make each day. A political checkpoint inside each operation would destroy the speed for which the machinery was adopted.

Democracies already respond to this problem through insulation. Central banks are not managed by referendum. Judges receive institutional protection from election returns. Independent agencies, professional civil services, treaty commitments, and long-term contracts all reduce the range of decisions a temporary political majority can immediately reverse.

The vocabulary for what could come next is old. Walter Bagehot divided the English constitution of his day into dignified parts, which held the public’s reverence, and efficient parts, which actually governed: the monarchy supplied the first, and the cabinet did the work. More recently, Colin Crouch and Peter Mair described democracies in which elections persist while decisions drift toward markets, agencies, and courts.[3] Those were accounts of drift. The prospect here differs in one respect: the transfer would be chosen deliberately, as a condition of winning a great-power competition, and defended as patriotism. In Bagehot’s terms, elections would move from the efficient constitution to the dignified one.

AI extends the logic of insulation and alters it. A central bank’s independence is bounded by a statute that names its mandate. A dam, once poured, stops making decisions. An AI administrative system keeps deciding. Each model update, revised threshold, retrained classifier, or new data pipeline changes how the state treats the people passing through it, often without a new rule being written or an official choosing the change. Shielding a finished investment from electoral reversal is old democratic practice. Shielding a rule-making process that keeps running after every vote is a different act, which the familiar precedents do not license.

The American government is already acquiring AI through this quieter layer, and its first safeguards sit in the same place. An April 2025 Office of Management and Budget memorandum told agencies to write data portability, knowledge transfer, intellectual-property rights, and protections against vendor lock-in into their contracts, and to work with vendors on data and rights when a contract is not renewed.[4] The memorandum is an attempt to keep the exit open. By April 2026, the Government Accountability Office reported that federal agencies had more than doubled their reported AI use between 2023 and 2024. Across thirteen acquisitions at four agencies, GAO found both agency-directed and vendor-driven adoption; in some cases, industry introduced capabilities in the absence of specific AI requirements. The four agencies were not systematically collecting lessons from their AI purchases.[5]

The Pentagon has already watched such an exit fail to open. The F-35’s Autonomic Logistics Information System, known as ALIS, runs the jet’s maintenance, supply chain, and mission planning, and the government relied heavily on Lockheed Martin to run the program’s sustainment. By 2020, corrupt or missing records in ALIS were flagging flight-ready aircraft for grounding, and the Pentagon announced a replacement called ODIN before deciding what roles it and its contractor would play in the new system.[6] In 2021 the program paused ODIN’s software effort, and its program executive officer told Congress the office had underestimated how hard retiring ALIS would be.[7] In 2023, GAO reported that the department still lacked the technical data it would need to take more of the work in house.[8] In June 2024, four years after announcing the replacement, the program began fielding what was planned as the final version of ALIS.[9] The department held the formal authority to change course throughout; it lacked the data, documentation, and transition capacity to use it. Newer AI services reproduce part of the problem. GAO found that agencies often buy AI as an ongoing service and that data rights were a recurring difficulty: FEMA officials could not share one damage-assessment model’s outputs with state partners because the rights had not been secured at award, even though the agency remained free to stop using the system, and later did.[5]

Congress kept every one of its powers. The consequential choices moved into instruments that electoral politics rarely reaches. The contract specifies the data rights. The standard defines acceptable performance. The vendor update changes the system’s behavior. The agency official signs the deployment. Congress later asks who was responsible.

At this point, the elected state begins to resemble human-in-the-loop governance at regime scale. The human remains visible at the end of a chain whose decisive commitments were made upstream, and must remain recognizable enough to authorize the system, defend it, and absorb blame when it fails; understanding or controlling it is optional. Ceremonial democracy is the political form of that arrangement. Elections choose who represents the state and supply a public author after operational authorship has moved elsewhere.

Why It Could Work

Ceremonial is a description, and it carries no verdict of uselessness. Political ceremony performs real functions. Elections designate winners without civil conflict. Legislatures give public language to distributive bargains. Presidents personify actions produced by institutions too large for any person to direct. Even when an officeholder controls little, the office supplies continuity, attribution, and a focal point for consent.

An AI state will need these functions more, not less. Machine administration can optimize, route, classify, and execute, yet technical competence alone cannot explain why one distribution of sacrifice should be accepted as legitimate. A model may identify the cheapest place for a transmission corridor while giving the people living there no reason to experience the decision as theirs. A procurement system may choose the highest-performing vendor and still say nothing about whether dependence on that vendor is an acceptable form of national vulnerability.

Ceremonial democracy would divide the work: the public supplies direction and periodic consent, protected institutions execute, and elected officials keep the public role while surrendering the operational one. The attraction is obvious. America would keep the peaceful succession, public legitimacy, and alliance appeal of democracy while acquiring some of the temporal continuity associated with China. Populists could campaign against the machine without being able to dismantle it. Governments could change without reopening the compute base, the grid plan, the standards architecture, or the security relationships on which national capacity depends.

This scenario is where many arguments for taking a technically important question “out of politics” lead. The question is how many questions can be removed before democratic government becomes the name of the surface rather than the structure beneath it.

What China Actually Has

The comparison with China is where the case begins to break. China’s advantage is often described as freedom from elections, a description that misses its deeper asset: concentrated authorship. Party organs, ministries, provincial governments, state firms, and private platforms operate inside a hierarchy capable of communicating priorities and enforcing correction. The system may be opaque from outside, yet from within it has an owner. When the Chinese state changes direction, it can force platforms to alter behavior, redirect finance, replace officials, reorganize an industry, or abandon a policy that has become too costly.

Correction has three parts, and concentration guarantees only one of them. The center can order a correction. Whether it has the information to choose the right one, and the capacity to carry it out, are separate questions. The abrupt end of zero-COVID in December 2022 displayed all three at once: authority sufficient to abandon a signature policy in a matter of weeks, information that arrived late, and an exit carried out with little preparation.

Research on Chinese policy experimentation shows where the information problem comes from. Shaoda Wang and David Yang found that China’s political structure enabled experimentation at extraordinary scale while also producing biased learning. More than 80 percent of the experiments they studied selected relatively advantaged localities, and local officials devoted effort to pilots that could not be replicated during national rollout. Centralized coordination made experimentation possible. Hierarchical incentives also corrupted the information it produced.[10]

This is the ceiling on concentrated coordination: compression increases execution while narrowing feedback. Error can travel farther before correction because the actors reporting upward are also judged by how well they satisfy the direction coming down.

A ceremonial democracy would import the compression while weakening the authorship. The bureaucracy could claim that the vendor controls the model. The vendor could claim that the contract controls deployment. The standards body could claim that it merely supplies a technical benchmark. The elected official could claim that experts made the decision. The public could remove the official and discover that the underlying arrangement remains.

The result would be China without the Party: a system insulated from elections that lacks a sovereign center able to see, own, and correct the whole. Formal authority would survive on paper while the capacity to use it spread across vendors, contracts, and standards bodies.

It would also preserve many of democracy’s delays. Courts, federalism, coalition management, procurement protests, and jurisdictional conflict would all remain. Ceremonial democracy would therefore risk combining democratic latency with authoritarian opacity, keeping the drag while losing the feedback.

When Ceremony Stops Producing Legitimacy

Democratic ceremony works because the public believes an election can eventually change something that matters. A citizen does not need to understand each provision in a procurement contract. The citizen needs a credible chain running from the vote to the people who authorize the contract, supervise its execution, and can replace it when its costs become unacceptable.

That chain can be long. It can contain experts, judges, civil servants, and independent institutions. It can make reversal deliberately difficult. Once it reaches a layer that no electoral outcome can reopen, the ceremony stops representing deferred control and starts advertising its absence.

The distinction will not remain hidden simply because the machinery is technical. People encounter upstream governance through downstream consequences. They see an application rejected, a benefit recalculated, an electricity bill increased, or a neighborhood selected for infrastructure. When elected officials repeatedly prove unable to alter the system producing those outcomes, political anger finds new targets.

Some of it moves toward courts. Some moves toward sabotage, secession, conspiracy, or street politics. Some searches for a leader willing to break the institutions that ordinary politicians say cannot be changed. The attempt to protect the machine from populism can therefore produce a more destructive populism organized around smashing the machine’s constitutional shell.

This is the instability inside ceremonial democracy. To keep elections harmless, the state must convince voters that elections matter while ensuring that they do not matter too much. Success hides the contradiction for a while. Failure reveals that the public was permitted to replace the performers but not the settlement.

At that point, maintaining ceremony requires more than ceremony: control over information, organized dependency, selective enforcement, or coercion. The regime has then left democratic vulnerability unsolved and crossed into managed political competition under democratic branding.

Democracy Above the Loop

The alternative cannot be direct public control over every technical choice; no complex state has ever operated that way. The more plausible design space separates operational insulation from constitutional abandonment. Elections remain distant from individual outputs but retain authority over the purposes, boundaries, and revision procedures of the systems producing them. Technical institutions receive time to execute without permanent ownership of the direction, and democratic control concentrates at a few high-leverage points.

The first is purpose. Legislatures determine which objectives an administrative system may optimize and which interests it must treat as constraints rather than costs. Efficiency cannot decide whether the state values equality, regional continuity, privacy, resilience, or national independence. Those are political choices even when they later appear as weights inside a technical system.

The second is jurisdiction. A machine may operate continuously inside an authorized domain, but it cannot decide for itself which people, institutions, or decisions fall within that domain. Expansion of scope remains a public act.

The third is revision. Models, thresholds, and standards change. A state that cannot reconstruct those changes cannot know whether the system still implements the mandate that authorized it. Version histories, change control, preserved records, and independent inspection supply the evidence that delegation has not silently become succession.

The fourth is replacement. Political authority can leave individual outputs alone so long as it keeps a credible ability to replace the leadership, mandate, or system at defined intervals. A termination clause is a formal right; using it requires the data, documentation, trained staff, and budget to keep operating the day after the vendor leaves. A replacement power worth having would be written in at award: government ownership of operational data, preserved model versions and configuration histories, recompetition at fixed intervals, and a funded migration plan rehearsed before anyone needs it. The power may still require a supermajority, a transition period, or compensation. Friction of that kind protects continuity without turning it into rule.

The final point is distribution. AI competition will create gains, losses, dependencies, and regions asked to absorb costs for a national project. No standards body can legitimately settle those claims through technical optimization. A democracy that gives up the distributional decision has delegated, along with implementation, the reason citizens accept the state’s authority in the first place.

This architecture would not make democratic government as fast as China, and speed may be the wrong objective. The democratic advantage lies in preserving several centers of observation, exposing error without treating disagreement as disloyalty, and changing direction without requiring regime failure. The strategic objective becomes distinguishing destructive interruption from corrective interruption, then building institutions that resist the first without disabling the second.

The Price of Correction

Any authority capable of correcting the machine state can also be used to damage it. A legislature that can reopen an AI mandate can write a foolish one. A president empowered to replace technical leadership can remove competent officials. Audit powers can become ideological weapons. Planning horizons shrink whenever a new majority threatens to reverse the last one’s work.

There is no institutional trick that eliminates this danger. The same democratic lever can interrupt a national project or rescue the country from a permanent mistake. Removing the lever solves both possibilities at once.

Constitutions have never guaranteed good judgment. Their function is to prevent one moment of bad judgment from becoming immediately total. An AI state would need the same structure at a different layer: protected operating periods, staggered appointments, and foundational changes that require more than a temporary majority. These arrangements make correction expensive without making it fictional.

Ceremonial democracy offers a cleaner bargain, and strong performance could keep it stable for a long time. Prosperity, security, and competent public services might let elections survive as rituals of membership rather than instruments of correction, and ceremonial democracy would then prove more durable than its critics expect.

The danger also recedes while AI systems remain modular and replaceable. A government that can move its data, models, and workflows can still turn electoral change into operational change. In July 2025, Executive Order 14319 directed agencies to buy only large language models meeting new principles of truth-seeking and ideological neutrality, and OMB turned the order into contract terms that permit termination for noncompliance.[11] An elected executive rewrote the terms on which the government buys AI, a lever that reaches model contracts more easily than energy contracts or embedded standards. Procurement becomes constitutional when dependence hardens enough that such reach fails.

China at least keeps an author: a political center entitled to order a correction, whatever the quality of its information and its capacity to execute. A democratic alliance that insulates its machinery from elections may end up with neither China’s command nor democracy’s feedback. It would keep the delay, distribute the authorship, and remove the peaceful mechanism for changing direction.

The temptation will be to protect everything important from politics and leave elections responsible for everything visible. That may preserve the democratic surface through the AI transition. It may also produce a public authorized to replace its rulers and forbidden to alter what rules through them.

If beating China requires democracy to become purely ceremonial, China has already chosen the terms of victory.

Notes

[1] State Council of the People’s Republic of China, “China Aims for AI Application Breakthroughs in Key Sectors in Next 2 Years: Official,” August 29, 2025. The official account describes six priority fields and deployment targets for 2027 and 2030.

[2] State Council of the People’s Republic of China, “China to Advance Smart Economy, Promote Digital, Intelligent Development,” March 5, 2026. The report summarizes the government work report and draft outline of the Fifteenth Five-Year Plan for 2026 to 2030.

[3] Walter Bagehot, The English Constitution (London: Chapman and Hall, 1867); Colin Crouch, Post-Democracy (Cambridge: Polity, 2004); Peter Mair, Ruling the Void: The Hollowing of Western Democracy (London: Verso, 2013).

[4] Office of Management and Budget, “Driving Efficient Acquisition of Artificial Intelligence in Government,” M-25-22, April 3, 2025. The memorandum directs agencies to address government data, intellectual-property rights, privacy, and vendor lock-in through contract terms, encourages solicitations that favor knowledge transfer and data and model portability, and asks agencies to implement rights and data terms with vendors after deciding not to extend a contract.

[5] U.S. Government Accountability Office, Artificial Intelligence Acquisitions: Agencies Should Collect and Apply Lessons Learned to Improve Future Procurements, GAO-26-107859, April 13, 2026. GAO reviewed thirteen acquisitions across the Departments of Defense, Homeland Security, and Veterans Affairs and the General Services Administration, and reported that agencies often acquire AI as a service.

[6] U.S. Government Accountability Office, Weapon System Sustainment: DOD Needs a Strategy for Re-Designing the F-35’s Central Logistics System, GAO-20-316, March 2020; and F-35 Sustainment: DOD Needs to Address Key Uncertainties as It Re-Designs the Aircraft’s Logistics System, GAO-20-665T, July 22, 2020.

[7] “F-35 Program Office Announces a ‘Strategic Pause’ on New Logistics System,” Defense News, April 22, 2021, reporting the testimony of Lt. Gen. Eric Fick, F-35 program executive officer, before the House Armed Services Committee.

[8] U.S. Government Accountability Office, F-35 Aircraft: DOD and the Military Services Need to Reassess the Future Sustainment Strategy, GAO-23-105341, September 21, 2023.

[9] Director, Operational Test and Evaluation, FY2024 Annual Report, “F-35 Joint Strike Fighter (JSF).”

[10] Shaoda Wang and David Y. Yang, “Policy Experimentation in China: The Political Economy of Policy Learning,” Journal of Political Economy 133, no. 7 (2025): 2180-2228. The authors find that China’s institutional structure enables experimentation at unusual scale while selection and incentive effects bias the lessons carried into national policy.

[11] Executive Order 14319, “Preventing Woke AI in the Federal Government,” 90 Fed. Reg. 35389 (July 28, 2025), signed July 23, 2025; Office of Management and Budget, “Increasing Public Trust in Artificial Intelligence Through Unbiased AI Principles,” M-26-04, December 11, 2025. The memorandum makes vendor transparency obligations material contract requirements and permits termination for noncompliance.