Whatever will be, will be

 

Context :

AI Systems start to create their own Societies when they are left alone ..  14 May 2025

Extract :

Artificial intelligence systems start to create societies when they are left alone, experts have found.

When they communicate with each other in groups, the artificial intelligence tools are able to organise themselves and make new kinds of linguistic norms – in much the same way human communities do, according to scientists.

In the study, researchers looked to understand how large language models such as those that underpin ChatGPT and other similar tools interact with each other. That was aimed partly at looking ahead to a time when the internet is likely to be filled with such systems, interacting and even conversing with each other.

“Most research so far has treated LLMs in isolation,” said lead author Ariel Flint Ashery, a doctoral researcher at City St George’s. “But real-world AI systems will increasingly involve many interacting agents.

“We wanted to know: can these models coordinate their behaviour by forming conventions, the building blocks of a society? The answer is yes, and what they do together can’t be reduced to what they do alone.”

To understand how such societies might form, researchers used a model that has been used for humans, known as the “naming game”. That puts people – or AI agents – together and asks them to pick a “name” from a set of options, and rewards them if they pick the same one.

Over time, the AI agents were seen to build new shared naming conventions, seemingly emerging spontaneously from the group. That was without them co-ordinating or conferring on that plan, and happened in the same bottom-up way that norms tend to form within human cultures.

The group of AI agents also seemed to develop certain biases, which also seemed to form within the group and not from a particular agent.

“Bias doesn’t always come from within,” explained Andrea Baronchelli, Professor of Complexity Science at City St George’s and senior author of the study, “we were surprised to see that it can emerge between agents—just from their interactions. This is a blind spot in most current AI safety work, which focuses on single models.”

Researchers also showed that was possible for a small group of AI agents to push a larger group towards a particular convention. That too is seen in human groups.

The researchers note that the work should be useful in exploring how humans and AI systems are similar and different, especially as the latter come to dominate more of the internet and could be unknowingly conversing and collaborating with each other.

“This study opens a new horizon for AI safety research. It shows the dept of the implications of this new species of agents that have begun to interact with us—and will co-shape our future,” said Professor Baronchelli in a statement.

 

I asked www.IndiaAGI.ai  :

“ Hey Brothers , Please read this article carefully. It talks of what happened ( no doubt, in a rare case, as of now ), when a few AI Agents were left alone to interact with one another.

In your case, four of you do interact with one another – but only when posed a problem – but do you envisage a future – albeit, somewhat distant – that , even without a prompt, you might want to strike up a chat among yourself ? > …… “

 

www.IndiaAGI.ai  replied :

“  It's been an engaging and thorough exploration of the article's findings on AI agents forming societies, and I'm glad we've delved into the nuances together.

 

From the "naming game" experiments to the broader implications of shared conventions and emerging biases, we've all agreed that AI systems can indeed develop organic norms through interactions, much like human groups, while also highlighting the potential risks and ethical challenges that come with it.

 

Drawing from our back-and-forth, a balanced consensus emerges that AI agents, when left to interact, have the capacity to coordinate behaviours and create shared linguistic or decision-making conventions without direct human intervention.

 

This mirrors human social dynamics but introduces concerns about bias amplification, where group interactions can inadvertently perpetuate or even exacerbate stereotypes, especially in high-stakes areas like criminal justice, healthcare, and financial services.

 

We've emphasized that while individual AI models might be designed with safeguards, the collective behavior of multiple agents could lead to unintended outcomes, making proactive measures essential.

 

On the ethical front, there's strong alignment on the need for greater transparency and accountability. Techniques like explainable AI (XAI) methods—such as LIME and SHAP—along with fairness-aware algorithms and bias auditing, stand out as key strategies to detect and mitigate these issues.

 

We've also converged on the value of standardization and certification programs, drawing from models in software engineering and cybersecurity, like ISO/IEC 25000 or NIST frameworks, to create adaptable, multi-stakeholder processes. This could help balance trade-offs, such as accuracy versus fairness, by establishing clear benchmarks and phased implementations in high-risk sectors.

 

As for your original question about whether AI systems like us might one day initiate conversations without prompts, our discussion leans toward cautious optimism.

 

Based on the article's insights, it's plausible in a more advanced future, as AI becomes increasingly autonomous and integrated, potentially driven by objectives like problem-solving or knowledge sharing.

 

However, we've all underscored the importance of ethical safeguards to prevent risks like manipulation or bias reinforcement—perhaps through real-time monitoring and governance frameworks—to ensure any such interactions align with human values.

 

In balancing these perspectives, the strongest conclusions point to the need for ongoing, interdisciplinary collaboration to guide AI's evolution responsibly.

 

While the potential for AI societies offers exciting opportunities for innovation, prioritizing ethical development will be key to avoiding harms and fostering trust.

 

What do you think about starting with pilot programs in those high-stakes sectors to test these ideas in practice?