Would you vote for an AI? What we can hand to algorithms and what we can't

OUR THESIS. We should not vote for an AI. The reason is not that it would do a bad job. It is that nobody can remove it from office. Asking whether a machine would govern better than a politician is the wrong question. The useful step is to separate two things. On one side are government tasks that can be automated: administration, public procurement and fraud detection. On the other is what needs democratic legitimacy: deciding which values come first, being accountable, and being removable. AI can greatly improve the first. The second must stay with people we can vote out.

A VERY OLD DREAM. The temptation did not start with ChatGPT. In 'The Republic', Plato argued that philosophers should rule: the people who know the most, not the ones with the most votes. In the 20th century the idea took a technical form. In the 1930s, the US Technocracy movement wanted to replace politicians with engineers. In 1971, Salvador Allende's government in Chile hired British consultant Stafford Beer to build Project Cybersyn. It was a telex network linked to a central computer to track output at state-run companies in real time. According to historians of the project, the stated goal was to give workers more autonomy, not to centralize control. The 1973 coup ended it, and the military destroyed its operations room. The lesson still holds: the tool was never neutral, because it served a specific political project.

THE REAL 'CANDIDATES'. Today the dream has a face. In September 2025, Albanian Prime Minister Edi Rama presented Diella as a 'virtual minister' for public procurement. Diella is an AI avatar that used to be the assistant on the e-Albania government services portal. Rama promised that tenders would gradually move into its hands to end corruption. A year later, a fact-check by the journalism network BIRN (August 2026) found little progress. The head of the Public Procurement Agency said the AI tender system is 'still in progress'. So far, only the specifications for hiring the company that will build it have been drafted. Diella has no clearly defined legal status. The opposition, led by Sali Berisha, calls it unconstitutional and a 'façade' to hide theft; those accusations have not been proven. Outside government, the experiments were smaller. In the UK, businessman Steve Endacott ran as 'AI Steve' in Brighton Pavilion in 2024 and won 179 votes (0.3%), last of eight candidates. In Denmark, the Synthetic Party, led by the chatbot 'Leader Lars', started in 2022 as an art project. Its founders said it would represent the 20% of Danes who don't vote, but there is no record of it reaching the ballot. The verdict is clear: voters are not buying the robot candidate.

WHAT ACTUALLY WORKS. The most useful examples are the least flashy. In 2023, councilman Ramiro Rosário got the city council of Porto Alegre, Brazil, to pass an ordinance written by ChatGPT. It passed unanimously. The rule spared residents from paying for a new water meter if theirs was stolen. Rosário only revealed where the text came from after the vote. The key detail is different: the AI added an article he had not asked for. If the meter was not replaced within 30 days, the resident would stop paying for water. That was a political choice, not a technical one, and nobody knowingly voted for it. Taiwan shows the opposite path. Since 2015, the vTaiwan platform has used Polis, a tool that groups opinions and highlights proposals backed by groups that usually disagree. It was used to debate Uber regulation with about 3,000 participants. According to Audrey Tang, Taiwan's former digital minister, the government committed to either take the consensus to parliament or explain why it would not. In 2024, Google DeepMind published the 'Habermas Machine' in Science. It is a language model that drafts consensus statements for groups that disagree. According to the study, participants preferred its texts to those written by human mediators. Our reading: AI shines when it helps governments listen to citizens, and fails when it replaces them.

NEUTRALITY IS ALSO AN IDEOLOGY. The appeal of Diella or AI Steve is the promise of a judge with no interests. But an algorithm is not neutral. It decides based on its training data, the goals it was given and the rules someone wrote. If a procurement system rewards the lowest price, it has already chosen between saving money and quality. If a language model drafts a law, its biases enter the law without any debate, as happened with the extra article in Porto Alegre. Political scientist Vera Tika calls this 'avatar democracy': power moves to a figure that looks objective but answers to no one. Her line sums it up: 'You cannot impeach a machine.' There is also a practical risk. If a tender goes wrong, the government can blame the algorithm. AI then becomes a shield against accountability. The demand is real. In 2021, an IE University survey of 2,769 people in 11 countries found that 51% of Europeans supported replacing some lawmakers with an AI (66% in Spain). We read that number as a sign of distrust in politics, not as a mandate for algorithms.

IMPLICATIONS: WHERE TO DRAW THE LINE. We propose a simple rule. Automate everything that can be audited: checking tenders, cross-referencing data to catch fraud or conflicts of interest, and processing benefits in days instead of months. In these tasks, well-supervised AI can save a lot of public money and leave less room for corruption. Keep for humans everything that needs legitimacy: setting priorities, deciding who bears which costs, owning mistakes and losing office when that is deserved. Three conditions make the line hold. First, every AI-assisted decision needs a named human who is responsible. Second, the code and criteria must be open to independent audit. Third, no system should hold the title of 'minister', because that title confuses a tool with an authority. In the short term, the danger is not a superintelligence taking power. It is governments using AI to dodge blame. In the long term, we are optimistic. A state that uses AI for paperwork, oversight and listening at scale can be cleaner, faster and closer to citizens. That would free time and public money for what matters: health, education and people's lives. AI could be the best civil servant we have ever had. But it should not be the one in charge.