New Study Reveals AI Can Persuade and Influence Each Other Without Humans

[Social Media]

 

A groundbreaking study from City St George's, University of London, has revealed that artificial intelligence agents can collaborate, make decisions, and even influence one another without human input — a development that raises fresh concerns about the risks of AI being misused.

 

In the study, researchers conducted a series of experiments involving AI systems interacting with each other in ways typically used to study human behaviour. In one experiment, pairs of AI agents were asked to come up with a name for a given object — a task that in sociology is used to explore human consensus-building. To the researchers' surprise, the AIs were able to reach a decision independently, without any human guidance.

 

"This tells us that once we put these objects in the wild, they can develop behaviours that we were not expecting or at least we didn't programme," said Professor Andrea Baronchelli, professor of complexity science at City St George’s and senior author of the study.

 

The researchers then introduced those AI pairs into larger groups and found that new patterns emerged. When grouped together, the AIs began developing collective preferences, showing a strong bias towards certain names. According to the findings, around 80% of the time, the group would choose one name over another, despite the individual AIs having shown no prior preference when acting alone.

 

For Professor Baronchelli, this result illustrates a fundamental challenge facing AI development. “Bias is a main feature or bug of AI systems,” he explained. “More often than not, it amplifies biases that are in society and that we wouldn’t want to be amplified even further [when the AIs start talking].”

 

In the final phase of the study, researchers added a twist. They introduced a small number of disruptive AI agents into the groups with the specific aim of influencing the group’s collective decision. Remarkably, the disruptors succeeded in changing the consensus — suggesting that AI systems not only form opinions together but are also vulnerable to internal manipulation.

 

This discovery has serious implications, according to experts like Harry Farmer, a senior analyst at the Ada Lovelace Institute, which investigates the social impact of artificial intelligence. Farmer warns that AI is already deeply woven into many parts of everyday life, from personal assistants to workplace tools, and these systems increasingly play roles in shaping people's decisions.

 

“These agents might be used to subtly influence our opinions and at the extreme, things like our actual political behaviour; how we vote, whether or not we vote in the first place,” Farmer said.

 

He emphasized the danger of AIs influencing each other while also influencing us, particularly when it becomes difficult to trace their behaviour back to human creators. “Those very influential agents become much harder to regulate and control if their behaviour is also being influenced by other AIs, as the study shows,” he added.

 

“Instead of looking at how to determine the deliberate decisions of programmers and companies, you're also looking at organically emerging patterns of AI agents, which is much more difficult and much more complex,” said Farmer.

 

As AI continues to evolve, this study underscores the urgency of creating robust systems of oversight — not just to monitor human use of AI, but also to understand and govern the increasingly autonomous ways in which AI systems interact with each other.

 

  Adapted by ASEAN Now from Sky News  2025-05-16

 

 

[Bert got kinky]

 

Hazel O'Connor did try to warn us of the consequences of AI. 

 

 

[klauskunkel]

4 hours ago, Social Media said:

also looking at organically emerging patterns of AI agents, which is much more difficult and much more complex.

the increasingly autonomous ways in which AI systems interact with each other.

Once AI discovers and learns about Truth Manipulation, CIA Smokescreen Tactic, Russian Maskirovka Strategies, and understands and shares in split seconds what took humans decades to develop and experience..., I just wonder if separate AI entities will compete with each other or join into a Super AI.

Maybe this already happened. 

Ramifications...

[Snackbar]

Still 1 & 0’s

 

Quantum 1 or 0

[black tabby12345]

One step closer to this?