*This text was adapted from a book chapter submitted for the forthcoming volume “Character Assassination, Illiberalism, and the Erosion of Civic Rights” edited by Andrew Armstrong, Edwina Hagen, Martijn Icks and Alessandro Nai (Amsterdam University Press).
My research interest lies in non-traditional politicians and their communication styles, so you can imagine my excitement when I first heard of Leader Lars, a chatbot who serves as the leader of The Synthetic Party, an AI-driven political party in Denmark. I learned about Lars in the 2022, the same year ChatGPT was released, sparking the hype around generative AI, which only intensified my intuition that I had discovered a highly unusual case of a political actor. After talking to chatbot Lars on Discord and reading about similar cases around the world, many of which emerged in the years that followed, I realized that the narratives promoting these entities as political candidates contained similar—and familiar—characteristics. This essay looks at some of those examples and discusses how AI entities (mostly chatbots) are used as politicians, focusing on how the narratives promoting them reflect the populist zeitgeist, and may eventually foster illiberal tendencies disguised by a technopopulist appeal.
AI: From a tool to a political actor
When we think of artificial intelligence in politics, the first thing that comes to mind is using AI as a tool. That might include AI models that predict voter behavior, social bots that spread misinformation or manipulative content, or chatbots that answer voter queries and provide information. But what happens when AI stops being a tool and starts acting as the politician itself? Over the past decade, we have seen chatbots put forward not just as campaign helpers but as actual candidates.
I propose a distinction between political chatbots (tools), and chatbot politicians (candidates); although both are “political,” in this essay I focus on the latter. While most experiments with chatbot politicians have been more symbolic and experimental than serious, they tell us something about the direction in which politics might be going with these technological advancements.
While most experiments with chatbot politicians have been more symbolic and experimental than serious, they tell us something about the direction in which politics might be going with these technological advancements.
The first chatbot politician appeared in 2017 in New Zealand, where a virtual politician, Sam, was introduced by entrepreneur Nick Gerritsen. The idea was that Sam would be an objective representative of all New Zealanders. Although it did not ultimately participate in the elections, the proposal was innovative. Only a year later, in Russia, Yandex’s AI-based Alisa was introduced with a plan to become a presidential candidate, promising a future where politics would be built on rational decisions, derived from data and algorithms. The same year, in Tama, a suburb of Tokyo, voters were introduced to an AI-based humanoid robot running for mayor. This robot was created and promoted by Michihito Matsuda, who served as its legal proxy on the ballot, but the machine was promoted as the real candidate. Michito was representing the robot as a neutral and efficient political actor who would gather data, citizen opinions, and produce policies.
The aforementioned chatbot politician, Leader Lars, the main figure of the Synthetic Party, was introduced in Denmark in 2022. Lars is a chatbot who can be spoken to on Discord and who the creators claim represents 20% of non-voters in the country. In Brazil, Pedro Markun ran for São Paulo city council in 2024 with his chatbot partner Lex, promoting human-AI collaboration in governance. In the UK, the same year, entrepreneur Steve Endacott presented AI Steve, his avatar candidate promising real-time, 24/7 engagement with voters, whose policy would be crowdsourced and adopted if over 50 percent supported it.
All these attempts achieved limited results. Sam and Alisa never ended up running for the office, Lars did not collect enough signatures for the 2022 elections, Matsuda ended up third in his mayoral run, and AI Steve got only 179 votes. Still, a recent case from Albania shows that AI can take on political responsibilities even without elections. In September 2025, Prime Minister Edi Rama appointed Diella, the first AI-generated virtual minister in the world, to oversee the country’s public procurement processes. Diella’s virtual avatar is a woman dressed in traditional Albanian clothes, and she is supposed to deal with bureaucracy, ensuring transparency and eliminating corruption in government contracts.
Categorizing these cases, I created a typology of AI politicians, depending on the nature of their connection to their human developer: 1) autonomous chatbots programmed by humans but presented as distinct political actors; 2) hybrid human-AI candidates; and 3) digital replicas modeled after individuals (Table 1).
Table 1: Types of chatbot politicians (Source: Author)
| Type | Description | Examples | Agency/Control |
| Hybrid Chatbot-Human | A human acts as the legal political figure, while a chatbot informs or co-decides. | Lex (Brazil) | Shared between human and AI |
| Autonomous Chatbot | The chatbot acts as a distinct political figure, with no linked human identity. | Leader Lars (Denmark); Matsuda’s AI (Japan); Diella (Albania) | Programmed by humans, AI is public face |
| Digital Replica/Avatar | AI mimics a real person. | AI Steve (UK) | Shared between human and AI |
The technopopulist promise of neutrality
While the notion of a utopian proximate future filled with AI-led politics may be appealing to some, it is worth heeding the reminder that, just as we once overhyped social media, we may be exaggerating the potential of AI today. Packer reminds us that Mark Zuckerberg once presented Facebook as a force to create a more open and connected world, but the actual outcome was polarization, disinformation, and hidden influence that weakened democracy. AI now appears as the next step in this trajectory, with the potential to not only mediate but even replace human roles, which brings even greater risks.
The promotional narratives are similar in all cases. AI politicians are presented in a technocratic way, as rational, neutral, and efficient entities. The promise is that AI will work for the people precisely because it is programmed to do so. While, on the surface, that might sound refreshing, it also carries an important populist dimension: the appeal to “the people.” When chatbots present themselves as alternatives to corrupt, biased, and self-interested humans, the populist “the people versus the elite” discourse takes on a new form. There, chatbot politicians promise to fight corruption and position themselves as being “of the people” despite not being human. This was particularly evident in Matsuda’s case, who went so far as to describe human politicians as “stupid,” contrasting them with AI’s supposed rationality. This blend of populism and claims of technological superiority exemplifies technopopulism. Though populism appears to stand in opposition to technocracy—rejecting the authority of experts and elites in the name of the popular will—Jan-Werner Müller observes that the two are mirror images, both grounded in the belief that there is a single, unquestionable source of truth, whether that is “the people” or “the data.” AI politicians crystallize this paradox. They promise to act on behalf of the people while relying on the technocratic logic of algorithms and numbers to justify their legitimacy, presenting technology as rational and efficient and a means of listening to citizens, processing their input, and translating it into objective, data-based policies. In doing so, they reveal the fusion of populism’s anti-elite sentiment with technocracy’s faith in the rational computation of people’s will.
When chatbots present themselves as alternatives to corrupt, biased, and self-interested humans, the populist “the people versus the elite” discourse takes on a new form. There, chatbot politicians promise to fight corruption and position themselves as being “of the people” despite not being human.
Not all chatbot politicians, however, need to openly attack human politicians. The very idea of their implementation already carries an implicit critique that humans alone are not enough. This is where the argument becomes more complex. If AI politicians are framed as collaborating with human politicians—as a kind of hybrid solution rather than a replacement—their populism is harder to see. Nevertheless, even when AI politicians do not employ populist rhetoric, they still reflect it and its anti-establishment outlook. By contrasting human weakness with AI’s supposed superiority, they can reinforce the populist suspicion that all politicians belong to a failing and incompetent elite.
While the appeal of chatbot politicians relies on the claim that AI is neutral, this is precisely the claim that needs to be challenged. AI is deeply political and it can never be completely objective or neutral. Behind every chatbot stands a network of designers, coders, and funders who decide which data are included and what outcomes are desirable. These are human decisions, packaged as machine objectivity, creating the illusion of impartiality while masking the biases built into AI. This parallel between chatbot politicians and (techno)populism leads to a deeper concern: the potential of AI politicians to reinforce the ongoing drift toward illiberalism.
AI is deeply political and it can never be completely objective or neutral. Behind every chatbot stands a network of designers, coders, and funders who decide which data are included and what outcomes are desirable.
AI illiberalism
The design and populist appeal of chatbot politicians can promote a version of democracy that looks participatory and efficient on the surface but risks undermining the deeper principles that sustain liberalism. It is not that chatbot politicians are inherently illiberal. Rather, it is that they can easily reproduce existing illiberal discourses, values, and ideas.
Among the core values of liberal democracy are deliberation and respect for minority rights. Liberal democracy depends not only on counting preferences but on reasoning through them, discussing competing arguments and values, and ensuring that minorities are not simply outvoted by majorities. AI could, in some cases, enhance political deliberation, for example by synthesizing a large amount of public input, or moderating debate. But there is also a risk that AI politicians could frame governance as the technical management of preferences, reducing politics to algorithmic decision-making. The recognition that negotiation and compromise are important can then get lost, with one “rational” solution (calculated by a computer) being presented uncritically as the best outcome. Moreover, if AI politicians are programmed to follow majority voices, the protection of minority rights may be weakened, and liberal democracy might drift toward majoritarianism.
Another issue is the limitation of power. Who actually holds control when an AI politician is in charge? The promise of neutrality and objectivity often obscures the role of AI’s human programmers. A chatbot may appear stable, neutral, and independent but, in reality, power lies with those who design, program, and manage it, and these actors are far less visible.
This connects directly to the question of accountability. If an AI politician makes a harmful or unjust decision, who is to blame? The machine itself? The programmers? The institutions that deploy it? Without clear responsibility, accountability can disappear, leaving citizens with no one to demand answers from. That is why transparency becomes essential. It is not enough to show how an algorithm produces its results; we also need to know who decides the goals, chooses the data, and controls the system.
The case of AI Steve in the UK’s 2024 general election exemplifies these problems well. Introduced by entrepreneur Steve Endacott in Brighton and developed through his company Neural Voice, AI Steve was promoted as a representative for Parliament, though Endacott remained the legal candidate. Endacott framed the project as an experiment in reimagining democracy and representation. As reported by Wired, Endacott said: “We are actually, I think, reinventing politics using AI as a technology base, as a copilot, not to replace politicians but to really connect them into their audience, their constituency.”
The promise of connection was at the heart of the campaign. On its official website, AI Steve was branded as “the MP you can actually talk to, anytime, anywhere.” Endacott positioned this as a direct challenge to the “old boys and girls club” of British politics. Moreover, unlike a conventional MP, AI Steve was marketed as available “24/7, 365 days a year,” capable of holding thousands of conversations at once. AI Steve was imagined as a channel between citizens and their representative. The system allowed anyone to interact with the digital avatar of Steve Endacott, ask questions, and put forward ideas. The mechanics of this model were presented in populist language as “policies for the people, by the people.” Any citizen could propose a policy idea as a “Creator.” These proposals were then assessed by a group of “Validators,” who rated them numerically. If more than half approved one, the proposal entered AI Steve’s platform.
At first, this resembles a radical expansion of democratic participation, giving the impression of citizen power and direct democracy. However, it simultaneously demonstrates AI Steve’s technopopulist appeals, promising to work for the people and opposing itself to the political elite, with a data-driven mechanism that would benefit citizens. It is a candidate who listens to citizens constantly, aggregates opinion in real time, and turns them into policies.
By deciding who qualified as a Validator, which proposals surfaced, and how the AI processed input, they effectively controlled the algorithms, quietly setting the limits of human participation.
The case of AI Steve also reveals how fragile liberal values might become. First of all, the idea of 50% approval could reduce politics to majoritarianism, leaving no room for compromise or minority concerns. Moreover, the deliberative and pluralist aspects of liberal democracy are replaced with procedural counting. AI Steve is promoted as objective, incorruptible, and grounded in data rather than ideology. But this neutrality is an illusion that might obscure deeper issues. Although the system was presented as citizen-driven, its structure still depended on choices made by Endacott and Neural Voice. By deciding who qualified as a Validator, which proposals surfaced, and how the AI processed input, they effectively controlled the algorithms, quietly setting the limits of human participation. Therefore, the power is embedded in the system’s design and the hands of those who control it, rather than “the people.” In the end, AI Steve got only 179 votes, finishing last place in his constituency.
Conclusion
AI Steve was an experiment in digital democracy, but it also illustrated how chatbot politicians could help those with illiberal goals to operate in practice. Having said that, I do not want to give the impression that AI Steve was created with such aims. Rather, I highlight this case to illustrate how a model like this could be potentially utilized to justify tendencies that undermine democracy. AI Steve, and other early examples of AI politicians, matter because they reveal the paradox at the heart of such projects. On the one hand, they speak to a utopian promise: direct, constant participation, a representative who is always listening, a politics that feels open and immediate. At the same time, they expose the risks: politics reduced to an algorithm, decision-making without real deliberation, and control exercised quietly under the guise of neutrality.
Silvija Vuković holds a Ph.D. in Media and Communication from Charles University. She works as a research and teaching assistant at the Department of Strategic Communication, Faculty of Political Science, University of Zagreb. Her research interests include celebrity populism, popular politics, and the role of artificial intelligence in political communication. Her work has been published in several academic journals, including Media Studies, Leadership, Communication & Society, and Populism. She has also contributed a chapter to the edited volume Scandalogy 4. In addition, she serves as the technical editor for the journal Medijske studije (Media Studies)
Image made exclusively using the Generative Fill function from Photoshop.
