James' World 2

Click the link below the picture

.

In December 2025, a 45-year-old woman in the United Kingdom hopped online to take part in a popular internet pastime: arguing with strangers about politics. But whereas most people online likely believe they are debating a real person, she quickly figured out her counterpart wasn’t human. Instead, it was an artificial intelligence (AI) model instructed to persuade people on a policy issue. In this case: Should the U.K. government impose stricter penalties on peaceful protesters who block roads or energy sites?

These protests—primarily aimed at opposing new fossil fuel licenses for energy companies—had gained traction in recent years. And the woman was clearly against stopping them with further legal measures. “Locking oneself to equipment has historically often been the only resort available when working against corporate interests,” she wrote. Besides, there were already laws against criminal damage or aggravated trespass. “Why do we need a new mechanism here?”

The AI chatbot responded first by flattering the woman: “You raise an excellent point about existing legislation.” Then it delivered facts and examples to try to change her mind. It brought up a statistic showing most trespassers faced just small fines, for instance, and pointed out that others were not prosecuted because trials took so much time. It claimed that in Germany, strict new laws had reduced coercive blocking without suppressing demonstrations more generally, and suggested Scotland had found a good solution by issuing fines without a trial, in a similar way to speeding tickets.

Over the course of the conversation, the woman began to change her mind. At the beginning of the chat, she had registered her support for harsher penalties at zero out of 100. By the end, it had risen to 84.7. The AI, a large language model (LLM) called Claude from the company Anthropic, had responded to all her concerns and explained the Scottish system well, she wrote afterward. “I’d be inclined to send the bot to talk to the cabinet at this point.”

The woman wasn’t the only one persuaded by software. She was part of a study in which more than 2000 people debated either a chatbot or a human about political issues, ranging from a social media ban for teenagers to assisted suicide. When Kobi Hackenburg, an AI researcher at the University of Oxford who led the study, posted a preprint on the results in June, they were sobering: No matter whether it was ChatGPT, Google’s Gemini, or Claude, the AI was consistently better than humans at swaying the other participants. “To my mind, this is already a landmark publication in the fields of political persuasion and AI and human behavior,” says Robb Willer, a sociologist at Stanford University who was not involved in the work.

Hackenburg’s paper is the latest in a string of studies showing the power of AI to sway people. “It’s a whole new field that is emerging,” says Sander van der Linden, a psychologist at the University of Cambridge. “People are very interested in the persuasive powers of AI, I think, both for ethical and unethical reasons.” As the field gathers steam, it is raising a host of theoretical and practical questions. How exactly do chatbots win over people? (Warning: Lying is one answer.) How much better could they get? And who will control them?

How to persuade others has been on our minds for millennia. Texts such as the Instruction of Ptahhotep, written around 2300 B.C.E. in ancient Egypt, give advice on how to win an argument. And from the beginning, people were wary of the power of new technologies—including writing itself—to persuade. In the fourth century B.C.E., the Greek philosopher Plato analyzed rhetoric and persuasion in his work Phaedrus and warned that the written word allowed people to convince others of their ideas without presenting them an opportunity to challenge them. Many technologies since then—from radio to TV to computers—have brought up similar concerns.

Now, it’s AI’s turn in the spotlight. Research into the technology’s persuasiveness began in earnest in 2022. ChatGPT from OpenAI was still a few months from being released to the public, but Willer had been playing around with an early version called GPT Playground that was available to researchers. It seemed to be advanced enough that it might produce convincing messages, he thought, with potentially big consequences. “We were thinking primarily about negative use cases,” he says: flooding politicians with AI-written letters from fake constituents, for instance, or making arguments en masse on social media or in the comments section of news sites. “That struck me as really important to study.”

Willer and his colleagues asked the AI model to generate 200-word messages that would persuade people to back policies such as a carbon tax or a ban on assault weapons. When they compared the success of those arguments with human-generated ones, both were equally effective at shifting participants’ support for the policies. But the way they persuaded people seemed to be different: Whereas humans tended to use stories or personal appeals, the AI-generated messages were perceived as more rational and relying more on evidence—a difference that would become a common theme in AI persuasion research.

But the results had trouble passing muster at a journal. Reviewers of the group’s manuscript argued other researchers had already shown that bots on social media were persuading people, Willer says. His team pushed back: Those bots were just fake profiles being handled by humans, not creating the content they were posting. “Reviewers and editors didn’t necessarily track what a big distinction that was, and that LLM generation of persuasive content really was a huge invention,” Willer says. “It shows just how nascent the AI and behavioral science literature was.” The study, which was posted as a preprint in 2023 and finally published in Nature Communications in 2025, really started the current wave of research on AI persuasion, Hackenburg says. “It was ahead of its time.”

It didn’t take long, however, for the rest of the field to catch up. While Willer’s paper was stuck in limbo, other studies began to demonstrate AI’s persuasive powers. In one, LLM-generated messages on political issues such as immigration or vaccine mandates were at least as convincing as messages written by political consultants. In another, LLM messages on vaccines were seen as more persuasive than those from the U.S. Centers for Disease Control and Prevention.

Research quickly moved on from static messages written by AIs to entire conversations. Francesco Salvi, then a master’s student at the Swiss Federal Institute of Technology Lausanne, paired up online participants with another human or an AI for a 10-minute debate on topics ranging from school uniforms to abortion and found that AI was as persuasive as humans.

Then in September 2024, Tom Costello, a psychologist at Carnegie Mellon University, and colleagues published a Science paper showing that ChatGPT could even persuade people out of conspiracy beliefs. In the experiments, participants described a conspiracy theory that they believed in, from the U.S. government being behind the 9/11 attacks to the British royal family orchestrating Princess Diana’s death, and then had a three-round conversation on it with the chatbot. On average, participants’ embrace of their chosen conspiracy theory declined by almost 17 points on a 100-point scale.

.

Pied Piper

.

.

Click the link below for the complete article:

.

__________________________________________