Opinion: AI can shape public debate without telling a lie
A 2026 conference panel on public opinion examined how AI can shape what people see, whom they trust and which choices they consider.
At a 2026 conference on AI, liberty, security and democracy organized by the bar associations of Lucca, Verona and Trento, a panel contribution made a distinction that matters for civic life: Manipulation does not require a lie. A system can shape an opinion simply by deciding what someone sees before they have a chance to weigh the evidence.
Facebook provided an early example in 2012. For one week, it changed the feeds of 689,003 English-speaking users, showing some fewer positive posts from friends and others fewer negative ones. Users’ own posts shifted slightly in the corresponding direction. The experiment did not establish that Facebook could dictate how people felt. It showed that changing the selection of genuine posts could influence what people expressed.
AI adds a conversational element. From November 2024 to March 2025, University of Zurich researchers introduced AI-generated replies into a Reddit forum where users invite others to change their minds, without telling the community. Some accounts presented invented personal experiences. Preliminary results indicated the replies were credited with changing minds three to six times as often as the human benchmark used in the comparison. That finding is limited to a forum where people had already asked to be challenged.
Deepfakes pose a different problem: A fabricated recording can pass for evidence, but the existence of convincing fakes also gives people a way to dismiss authentic footage. A democracy needs more than a warning to distrust what we see. It needs ways for people to check evidence without having to take a single authority’s word for it.
The contribution looked to Peio, where Alpine woods and high pastures have long been managed collectively. That arrangement offers an analogy, not a ready-made policy: Some resources are too difficult for individuals to protect on their own. The practical question is what institutions would give communities a real say over how their data is used and which AI systems they adopt.






Comments