What an Absent Dictator Reveals About Our AI Future

A ruler's power depends partly on the people he needs. AI could change how many that is.

Paul Biya has taken working from abroad further than most employers would tolerate. He is the president of Cameroon, where he has held power since 1982. He also has a well-documented attachment to Switzerland.

An OCCRP investigation published in 2018 counted at least four and a half years of private foreign visits across his first 35 years in office, not counting official travel, much of it at Geneva’s InterContinental hotel. When the findings came out, a government minister defended him on the radio: “with ICT, it is possible to pilot an organisation from wherever you are.” This year Biya left on June 7 and flew back from Geneva on August 20, more than ten weeks later. He is 93.

During that absence, the army got attention. On the 58th day, Reuters reported, decrees signed while he was abroad promoted the commander of the Presidential Guard and replaced the commanders of four of the country’s five joint military regions. Other business waited. The new government Biya promised in his New Year’s address had still not been named by October 6, according to Al Jazeera. The command of the army got his signature.

Biya’s absences offer no evidence that Cameroon runs well without him. They show that holding power and doing the daily work of government are different activities. Other people’s participation remains essential, and so does the chance that one of them objects and has to be heard. AI raises the possibility of changing that part.

The People a Dictator Still Needs

The popular picture of a dictator is a man who can do whatever he wants. In practice, his orders have to pass through people who know things he does not, control resources he cannot personally administer, and command people who do not take instructions directly from him.

A general has soldiers. A minister has an organization. The ruler can threaten them, reward them, replace them or divide them against one another. Each approach requires some ability to get the work done after the threat is carried out.

The political scientist Milan Svolik has modeled this bargaining: a dictator can try to gather power at his allies’ expense, and they can deter him with the threat of a coup. Bruce Bueno de Mesquita and Alastair Smith built The Dictator’s Handbook around the size of the group a ruler has to keep rewarded. Even a dictatorship contains people whose cooperation has a price.

Cameroon shows that the price is not the same for everyone. The political scientist Achille Mbembe, quoted in a 2018 piece by the journalist Emmanuel Freudenthal, described a president whose method is to decide as little as possible. In Freudenthal’s account, officials afraid of losing well-paid posts avoid decisions of their own and wait to be appointed. Ministers who fall from favor can be dismissed without explanation, and some have been prosecuted. Yet Freudenthal also describes high office as a reward for loyalty, part of a patronage network that by 2018 ran to more than sixty ministers and state secretaries. Biya can drop a minister more easily than he can stop using the posts to pay for loyalty.

Those are ugly constraints on power, but they make a ruler take account of something outside his own wishes. Needing other people has been one of the accidental protections against a single will: every order had to pass through someone who could slow it, bend it or refuse it. AI could remove some of that friction, and not only in dictatorships.

A Smaller Room

You are probably somewhere in a chain like this. A leader wants to know which projects are late. Departments send reports. Someone reconciles the figures and drafts a briefing. The person at the top sees a short document assembled through many judgments about what deserves attention, some of them yours.

People in that chain have influence because the person above them needs their work. Replacing whoever formats the report may be easy; replacing whoever knows which figures were padded is much harder. That knowledge is also what lets you say a plan won’t work and have the plan wait while someone listens.

Governments are already testing tools that do parts of this work. In a British trial covering 20,000 government employees, published in June 2025, participants reported saving an average of 26 minutes a day with Microsoft Copilot, most often on drafting documents, and some said it struggled with nuanced or context-heavy work that needed human judgment. These were assistants beside officials, and the trial did not establish staff replacement or any shift in who decides.

Picture one link in that chain. You run a regional office, and each quarter you tell your boss which of the late projects are really in trouble, because you know which contractor pads his estimates. Your boss does not sign off until she has heard from you. Now give her a system that reads the contracts, invoices and site reports directly and flags the padded estimate itself. Next quarter she approves the schedule without calling, and the first you hear of it is the email. She might instead use the saved time to hear more people.

Which way it goes depends on what your influence rests on. If you matter because you can get people on the ground to cooperate, a better dashboard does little to that. If you matter because you control what reaches the top, the dashboard is aimed at your position. Headcount is a poor guide: an office can keep its whole payroll while its consequential judgments move into software. Your job could survive the change while your objection stops changing anything. Ask what would happen the next time you told the person above you that a plan won’t work: would it wait for you, or go ahead around you?

The Bargain Can Move

For a ruler, the attraction is getting more done with fewer people whose consent, loyalty or ambitions have to be managed. It has limits. A chatbot cannot make a battalion loyal, and the army is where Biya’s attention went while he was away; the civilian business he lets wait did not need software to be neglected. Automating a ministry’s paperwork does not remove the reason the post was handed out, which was to reward loyalty. A system that summarizes reports cannot guarantee that the reports describe reality.

Nor does buying software remove dependence. A ruler who reduces his dependence on ministers could increase his dependence on a small technical circle or a foreign provider. Their influence would sit in access permissions, updates and the assumptions built into the tools.

This could become another way for courtiers to filter what reaches the ruler. A confident summary may conceal an inconvenient fact more successfully than an obviously nervous aide. A presidency organized around pleasing the president could automate its existing blindness. The tools can also cut the other way: a system that reads the invoices can show an auditor the padded figure as easily as it lets a manager skip the person who knew. Which happens depends on who can see what the system shows and who has the authority to act on it.

In a 2025 report, Tom Davidson, Lukas Finnveden and Rose Hadshar of Forethought argued that AI systems loyal to one person could stand in for the soldiers and officials whose cooperation a seizure of power has always required. Luke Drago and Rudolf Laine’s The Intelligence Curse made an economic version of the case. In Cameroon, the decrees issued in an absent president’s name were about his commanders. If AI moves this bargain, it will show up as a leader going around someone he previously could not afford to lose, a regional commander or the official who keeps the real records, and getting away with it. Where officials are still needed to correct the system, the bargain survives.

Democracies have courts, elections, professional duties and organized opposition that protect the right to object whether or not an official is useful to a leader. They work best while more than one set of people still carries out decisions and can explain how to change them. A legal right to challenge a decision counts for less when nobody you can reach knows how to alter it.

When the President Does Not Return

Biya’s arrangements face a limit no Swiss hotel can postpone. In April, Cameroon restored the office of vice president, to be filled by the president’s choice. In October the post was still empty, and the law sets no deadline. In June, someone delivered a purported decree naming a vice president to the state broadcaster; the government said it had not gone through the proper process, and the person who delivered it was detained. A decree from an absent president counts only when the people who receive it agree to treat it as real.

Freudenthal ended his 2018 account by joking that even this president could not rule from beyond the grave. A dead ruler already leaves institutions, loyalists and laws, and the people interpreting them gain room to redirect his inheritance while invoking his name.

An automated administration could preserve a more detailed inheritance: the priorities used to rank cases, the criteria for distributing resources and the records from which future recommendations are generated. Software can continue applying those choices after the person who authorized them has died.

The software would not be president. The living still have to fund it, maintain it and accept its authority, and successors may replace it. More plausibly, daily work comes to depend so heavily on the inherited arrangements that changing them becomes a struggle of its own.

A system flexible enough to carry out one president’s priorities could carry out the next one’s, provided someone is left who knows how to change it. That condition fails when automation retires the people who made those decisions without recording why they made them: the successor inherits rules his staff can run and no one can explain. He could hold the title and the legal authority to reverse course while lacking the people or the understanding to do it without interrupting essential services. Whoever controls the system would then hold something the new president needs.

Biya’s absences already show a version of this without any software: the president leaves, and decrees keep arriving under his name, carried out by people who have organized their careers around him. AI could make those arrangements more capable, less dependent on particular officials, and easier to carry forward.

The next president could arrive to find his predecessor’s priorities still being applied to the files on his desk.