America just turned 250 years old. The founders built self-government into a world of pamphlets and town meetings, and now we apply their political architecture to AI.
The birthday question is whether AI strengthens democracy or weakens it. Serious thinkers have lined up both sides of the great debate.
Here is my scorecard, extracted from five books and seven articles, and then the question neither side is asking: which is growing faster, power over AI or access to it?
Start with surveillance.
Yuval Noah Harari argues The Nexus that democracy is a distributed information network with self-correcting mechanisms: free radio, opposition parties, and courts that catch mistakes and correct them. Dictatorship is a centralized network that suppresses reform. For two centuries, centralization had a built-in cost, because complete surveillance required armies of human informers, and armies are expensive. AI removes costs. It watches everyone, all the time, for pennies. Evidence is no longer thought. A study in the Quarterly Journal of Economics documented a feedback loop in China: local unrest leads to government purchases of facial recognition AI, and those purchases suppress subsequent unrest. The authors titled their paper “AI-tocracy.”
The second argument is economic.
Previous technology replaced some workers, switchboard operator, road collector, while job opportunities were created for people using new machines. The desire for AI targets all employees. Daron Acemoglu and Simon Johnson presented the book, Power and Developmentto this concern, writing that “the current approach to AI is not good for the economy or democracy.” Acemoglu, the 2024 Nobel laureate, hammered home this point in Fortune in February, warning that on the current path of job destruction and growing inequality, “US democracy will not survive.”
The third argument is aimed at the self-governing machine itself.
I sounded this alarm in the Harvard Business Review back in 2019, warning that AI is poised to make reliable forgery of video, audio, and documents cheap and automated, with potentially catastrophic consequences for democracy. The deception is old. AI is industrializing it. Security expert Bruce Schneier predicts that AI will improve lobbying and write “small rules,” small provisions that quietly benefit one group, and he notes that technology often makes the powerful more powerful. He and Nathan Sanders became deeply concerned when a book written about AI opposing AI legislation appeared in the New York Times. Marietje Schaake provides the foundation of the institution The Tech Coup: non-elected companies are now carrying out functions that used to belong to the government.
The prosecutor rests. Now comes the defense.
On July 4, computer scientist Daphne Koller marked her 250th birthday, and her 37th anniversary as an immigrant, by visiting Shasta Dam. In a presentation posted that day, he argued that America’s signature success is taking what was rare and making it abundant: water to power in Shasta, electricity to a grid anyone can plug into, pocket computing. He did it himself; Coursera, which he co-founded, has put higher education in front of more than 150 million students. He wrote that AI is the next chapter, “making abundant one of the rarest resources in the world: powerful thinking.” Judgment that was once reserved for authorized experts is now for anyone who can ask the right question. Lawyers and doctors charge by the hour. The AI responds in seconds.
The economic counter comes from MIT Acemoglu’s colleague David Autor, who argues with Noema that AI can transfer expertise to workers without advanced degrees and thus rebuild an open labor market. The first evidence points his way. When a Fortune 500 company gave its customer support agents an AI assistant, productivity increased by 15% on average, and the benefits went largely to new and less skilled workers, who improved in both speed and quality. The study, published in the Quarterly Journal of Economics, found that the most experienced agents made the least profit. If the pattern holds, AI can squeeze the very gaps Acemoglu fears will widen.
Reid Hoffman and Greg Beato The Superagency states the state of hope in general terms: AI increases individual agency so widely that the real danger lies in democracies handing over their development to less benevolent actors. In PluralTaiwan’s first digital minister Audrey Tang and economist Glen Weyl describe a decade of digital tools that have gained consensus in a society divided by live laws, from horse-sharing laws to pandemic policy. Controlled trials support them. Google DeepMind researchers built an AI arbiter, tested it on 5,734 Britons with questions about Brexit and immigration, and reported in Science that participants preferred the AI team’s statements to a human arbiter, which were more clearly balanced and less biased. The groups also ended up not being too different. The city hall never fit a million people. It can happen now.
I put the two columns side by side and noticed something strange: they don’t fit together. Pessimists argue about who controls AI. Optimists argue over who will use it. Power and access are separate questions, and both fields can be right at the same time.
Koller’s dam makes a physical point. Generation is concentrated, a few turbines owned by a few. The grid is distributed, and anyone can connect. One machine does both at the same time. AI shares that nature: anyone can connect to a frontier model for $20 a month, while the frontier weights and training data centers belong to twenty-two companies.
Gutenberg adds duration. The printing press broke Rome’s monopoly on literary expression, and four centuries later built Hearst’s empire; access and power exchanges on the same machine. Both forces are real. The open question is which one goes faster, and the current battles over open weights, chip shipments, and model ownership are the battles that will help settle this question.
The founders faced the same question about concentrated power and answered it by spreading the vote, gradually, and later almost universally. Koller ended his career with an obligation fitting for the country’s 250th year: anyone who has been given more than his share owes a duty to ensure that the next rare item does not remain in short supply for long. Intelligence is the next scarce commodity. Koller’s dam has already been built, along with boundary models and training data centers. The decision before us is whether we build the grid, providing broad, affordable access to AI for all Americans.
