When consensus becomes cheap
The next challenge for democracy may not be fake information, but manufactured agreement.
TL;DR: The political danger of AI is not only that it can imitate people, but that it can lower the cost of manufacturing apparent majorities. When consensus becomes cheap, power shifts toward those who can simulate public pressure at scale.
Democracies have always lived with attempts to influence public opinion. Political actors compete to frame debates, mobilise supporters, and persuade others that their vision of society is the one people should follow. This is not a failure of democracy, but part of the process through which collective decisions are formed. Ideas spread, groups organise, and societies constantly negotiate what becomes accepted, rejected, or contested.
What changes with generative AI is the cost of this process. Historically, creating the appearance of broad agreement required resources. Movements had to mobilise people, organisations had to build networks, and media campaigns required infrastructure. Even online manipulation depended on human coordination, attention, and time. Manufacturing consensus at scale was possible, but it was difficult.
Generative AI alters this balance because it does not only automate the production of content. The familiar concern is that AI can generate unlimited text, images, or videos, making misinformation easier to produce. But the deeper transformation is the automation of coordinated behaviour: many voices appearing to interact, reinforce, criticise, adapt, and collectively shape a conversation.
This matters because humans are social learners. We rarely evaluate ideas only by considering their internal logic. We also pay attention to the social environment around them. We ask whether others find an argument convincing, whether a belief is marginal or becoming mainstream, and whether society appears to be moving in a particular direction. These signals influence how individuals behave.
Collective life depends on this process. A private belief can become public when people realise others share it. A movement can grow when isolated individuals discover that they are part of something larger. Norms emerge and stabilise because people form expectations about the behaviour of others. Consensus is never simply a collection of independent opinions. It is also a perception of what other people think.
This is why coordinated populations of AI agents represent a new political challenge. They do not need to persuade everyone directly. Their influence may come from changing the perceived social landscape: making a position appear more widespread, a reaction more spontaneous, or a shift in public opinion more inevitable than it actually is.
This is different from traditional misinformation. False information tries to change what people believe about reality. Manufactured consensus tries to change what people believe about other people. It targets the layer where individual beliefs become collective behaviour, where opinions become movements, and where societies decide what counts as normal.
The consequences go beyond any individual campaign. If creating apparent agreement becomes cheap, the distribution of influence changes. Smaller actors may acquire capabilities that previously required large organisations, while powerful actors may gain the ability to operate continuously and adaptively at unprecedented scale. The political question is therefore not only who controls information, but who can shape the perception of collective agreement.
The result may not be a world where everyone believes artificial voices. A more subtle possibility is a world where every collective reaction becomes easier to question. If any wave of support, criticism, or mobilisation could be synthetic, genuine movements can also be dismissed as artificial. The damage comes not only from deception, but from uncertainty about what is authentic.
The challenge ahead is therefore broader than detecting fake content. Democracies have always depended on disagreement and competing visions of society. They do not require everyone to agree. But they do require some confidence that public debate reflects interactions among real people and real communities.
AI does not simply create new voices. It changes the cost of creating the crowd. And when crowds can be manufactured cheaply, we need to rethink how collective opinion forms, how influence operates, and how democratic societies recognise themselves.