(Press-News.org) A new study suggests that populations of artificial intelligence (AI) agents, similar to ChatGPT, can spontaneously develop shared social conventions through interaction alone.
The research from City St George’s, University of London and the IT University of Copenhagen suggests that when these large language model (LLM) artificial intelligence (AI) agents communicate in groups, they do not just follow scripts or repeat patterns, but self-organise, reaching consensus on linguistic norms much like human communities. The study has been published today in the journal, Science Advances.
LLMs are powerful deep learning algorithms that can understand and generate human language, with the most famous to date being ChatGPT.
“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.”
In the study, the researchers adapted a classic framework for studying social conventions in humans, based on the “naming game” model of convention formation.
In their experiments, groups of LLM agents ranged in size from 24 to 200 individuals, and in each experiment, two LLM agents were randomly paired and asked to select a ‘name’ (e.g., an alphabet letter, or a random string of characters) from a shared pool of options. If both agents selected the same name, they earned a reward; if not, they received a penalty and were shown each other's choices.
Agents only had access to a limited memory of their own recent interactions—not of the full population—and were not told they were part of a group. Over many such interactions, a shared naming convention could spontaneously emerge across the population, without any central coordination or predefined solution, mimicking the bottom-up way norms form in human cultures.
Even more strikingly, the team observed collective biases that couldn’t be traced back to individual agents.
“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.”
In a final experiment, the study illustrated how these emergent norms can be fragile: small, committed groups of AI agents can tip the entire group toward a new naming convention, echoing well-known tipping point effects – or ‘critical mass’ dynamics – in human societies.
The study results were also robust to using four different types of LLM called Llama-2-70b-Chat, Llama-3-70B-Instruct, Llama-3.1-70BInstruct, and Claude-3.5-Sonnet respectively.
As LLMs begin to populate online environments – from social media to autonomous vehicles – the researchers envision their work as a steppingstone to further explore how human and AI reasoning both converge and diverge, with the goal of helping to combat some of the most pressing ethical dangers posed by LLM AIs propagating biases fed into them by society, which may harm marginalised groups.
Professor Baronchelli added: “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. Understanding how they operate is key to leading our coexistence with AI, rather than being subject to it. We are entering a world where AI does not just talk—it negotiates, aligns, and sometimes disagrees over shared behaviours, just like us.”
The peer-reviewed study, “Emergent Social Conventions and Collective Bias in LLM Populations,” is published in the journal, Science Advances.
ENDS
Notes to editors
For media enquiry, including request for a copy of the research paper shared in confidence ahead of the embargo lifting, please contact corresponding author, Professor Andrea Baronchelli (a.baronchelli.work@gmail.com, andrea.baronchelli.1@citystgeorges.ac.uk), cc’ing press officer, Dr Shamim Quadir (shamim.quadir@citystgeorges.ac.uk).
Link to research paper once embargo lifts
https://doi.org/10.1126/sciadv.adu9368
Media Contact
For media enquiry, contact Dr Shamim Quadir, Senior Communications Officer, School of Science & Technology, City St George’s, University of London: Tel: +44 (0) 207 040 8782 email: shamim.quadir@citystgeorges.ac.uk.
Expert Contact
Contact corresponding author, Andrea Baronchelli, Professor of Complexity Science, Department of Mathematics, School of Science & Technology, City St George’s, University of London: Tel: +44 (0) 207 040 8124 email: andrea.baronchelli.1@citystgeorges.ac.uk, a.baronchelli.work@gmail.com
About the academics
Professor Andrea Baronchelli is a world-renowned expert on social conventions, a field he has been researching for two decades. His pioneering work includes the now-standard naming game framework, as well as groundbreaking lab experiments showing how humans spontaneously create conventions without central authority, and how those conventions can be overturned by small committed groups.
About City St George’s, University of London
City St George’s, University of London is the University of business, practice and the professions.
City St George’s attracts around 27,000 students from more than 170 countries.
Our academic range is broadly-based with world-leading strengths in business; law; health and medical sciences; mathematics; computer science; engineering; social sciences including international politics, economics and sociology; and the arts including journalism, dance and music.
In August 2024, City, University of London merged with St George’s, University of London creating a powerful multi-faculty institution. The combined university is now one of the largest suppliers of the health workforce in the capital, as well as one of the largest higher education destinations for London students.
City St George’s campuses are spread across London in Clerkenwell, Moorgate and Tooting, where we share a clinical environment with a major London teaching hospital.
Our students are at the heart of everything that we do, and we are committed to supporting them to go out and get good jobs.
Our research is impactful, engaged and at the frontier of practice. In the last REF (2021) 86 per cent of City research was rated as ‘world-leading’ 4* (40%) and ‘internationally excellent’ 3* (46%) and 100 per cent of St George’s impact case studies were judged as ‘world-leading’ or ‘internationally excellent’. As City St George’s we will seize the opportunity to carry out interdisciplinary research which will have positive impact on the world around us.
Over 175,000 former students in over 170 countries are members of the City St George’s Alumni Network.
City St George’s is led by Professor Sir Anthony Finkelstein.
END
Groups of AI agents spontaneously form their own social norms without human help, suggests study
First-of-its-kind study suggests that groups of artificial intelligence language models can self-organise into societies, and are prone to tipping points in social convention, much like human societies.
2025-05-14
ELSE PRESS RELEASES FROM THIS DATE:
Different ways of ‘getting a grip’
2025-05-14
To the point
Different hand use: Two ancient human relatives, Australopithecus sediba and Homo naledi, had different finger bone morphologies that indicate they used different types of hand grips, both when using tools and when climbing
Internal structure of the finger bones: A. sediba had a mix of ape-like and human-like features, while H. naledi had a unique pattern of bone thickness, suggesting different loading patterns and possible grip types.
Human Evolution: Ancient human relatives adapted to their environments in diverse ways, balancing tool use, ...
Handy octopus robot can adapt to its surroundings
2025-05-14
Scientists inspired by the octopus’s nervous system have developed a robot that can decide how to move or grip objects by sensing its environment.
The team from the University of Bristol’s Faculty of Science and Engineering designed a simple yet smart robot which uses fluid flows of air or water to coordinate suction and movement as octopuses do with hundreds of suckers and multiple arms.
The study, published today in the journal Science Robotics, shows how a soft robot can use suction flow not just to stick to things, but also to sense ...
The ripple effect of small earthquakes near major faults
2025-05-14
When we think of earthquakes, we imagine sudden, violent shaking. But deep beneath the Earth’s surface, some faults move in near silence. These slow, shuffling slips and their accompanying hum—called tremors—don’t shake buildings or make headlines. But scientists believe they can serve as useful analogs of how major earthquakes begin and behave.
A new study by geophysicists at UC Santa Cruz explains how some of these tremor events can yield insights into how stress builds up on the dangerous faults above where major earthquakes ...
Mass General Brigham researchers pinpoint ‘sweet spot’ for focused ultrasound to provide essential tremor relief
2025-05-14
KEY TAKEAWAYS
Three decades ago, researchers at Brigham and Women’s Hospital pioneered MRI-guided focused ultrasound (MRgFUS) thalamotomy, a technique that offers lifechanging results for patients.
In a new study, researchers looked at results from more than 350 patients treated with MRgFUS for essential tremor to assess clinical improvements and side effects.
Their research creates a model of an optimal location for ablation, which will help make the procedure safer and more effective for patients at Mass General Brigham and around the world.
For millions of people around the world with essential tremor, everyday activities ...
MRI scans could help detect life-threatening heart disease
2025-05-14
Magnetic resonance imaging (MRI) scans of the heart could help to detect a life-threatening heart disease and enable clinicians to better predict which patients are most at risk, according to a new study led by UCL (University College London) researchers.
Lamin heart disease is a genetic condition that affects the heart’s ability to pump blood and can cause life-threatening abnormal heart rhythms. It is caused by a mutation in the LMNA gene, which is responsible for producing proteins used in heart cells. It often affects people in their 30s and 40s.
Lamin disease is rare but also often undiagnosed. About one in 5,000 people in the general population carry a potentially ...
NASA’s Magellan mission reveals possible tectonic activity on Venus
2025-05-14
Vast, quasi-circular features on Venus’ surface may reveal that the planet has ongoing tectonics, according to new research based on data gathered more than 30 years ago by NASA’s Magellan mission. On Earth, the planet’s surface is continually renewed by the constant shifting and recycling of massive sections of crust, called tectonic plates, that float atop a viscous interior. Venus doesn’t have tectonic plates, but its surface is still being deformed by molten material from below.
Seeking to better understand the underlying processes driving these deformations, the researchers studied a type of feature called a corona. ...
A step forward in treating serious genetic disorders prenatally
2025-05-14
Injecting medicine into the amniotic fluid staves off progression of spinal muscular atrophy in utero.
Evidence is mounting that clinicians can treat serious genetic disorders prenatally by injecting medicine into the amniotic fluid, thus preventing damage that begins in utero.
A UC San Francisco-led study found that delivering medicine for spinal muscular atrophy (SMA) via the amniotic fluid was safe, and it helped prevent damage to nerve cells in the spinal cord, a part of the central nervous system that is responsible for movement. One experiment was done in mice with SMA — a neurodegenerative disease that causes muscular weakness, atrophy, ...
New study shows AI can predict child malnutrition, support prevention efforts
2025-05-14
A multidisciplinary team of researchers from the USC School of Advanced Computing and the Keck School of Medicine, working alongside experts from the Microsoft AI for Good Lab, Amref Health Africa, and Kenya’s Ministry of Health, has developed an artificial intelligence (AI) model that can predict acute child malnutrition in Kenya up to six months in advance.
The tool offers governments and humanitarian organizations critical lead time to deliver life-saving food, health care, and supplies to at-risk areas.The machine learning model outperforms traditional approaches by integrating clinical data from more than 17,000 Kenyan health facilities with satellite data on crop ...
Microplastics in Texas bays are being swept out to sea
2025-05-14
From tiny pellets to creepy wave-battered baby dolls, the Texas coast is a notable hot spot for plastic debris.
But when researchers from The University of Texas at Austin went searching for microplastics in sediments pulled from the bottom of Matagorda Bay and its surrounding inlets, they didn’t find much.
Most of their samples contained only tens to hundreds of microplastic particles for each kilogram of sediment. This is hundreds to thousands of times less than other bayside environments around the world.
Their findings, which were published in Environmental Science & Technology, suggest that rather than settling at the bottom ...
Loneliness increases risk of hearing loss: evidence from a large-scale UK biobank study
2025-05-14
A large-scale cohort study led by researchers from Tianjin University, Shenyang Medical College, Shengjing Hospital of China Medical University, and the Chinese University of Hong Kong has uncovered strong evidence that loneliness may independently increase the risk of hearing loss. The findings were published in Health Data Science on May 2, 2025.
Hearing loss is one of the most prevalent global health conditions, affecting more than 1.5 billion people. While physiological and behavioral risk factors are well-documented, the role of psychosocial factors such as loneliness has been underexplored. This study sought to determine ...
LAST 30 PRESS RELEASES:
From blood sugar to brain relief: GLP-1 therapy slashes migraine frequency
Variability in heart rate during sleep may reveal early signs of stroke, depression or cognitive dysfunction, new study shows
New method to study catalysts could lead to better batteries
Current Molecular Pharmacology impact factor rises to 2.9, achieving Q2 ranking in the Pharmacology & Pharmacy category in 2024 JCR
More time with loved ones for cancer patients spared radiation treatment
New methods speed diagnosis of rare genetic disease
Genetics of cardiomyopathy risk in cancer survivors differ by age of onset
Autism inpatient collection releases genetic, phenotypic data for more than 1,500 children with autism
Targeting fusion protein’s role in childhood leukemia produces striking results
Clear understanding of social connections propels strivers up the social ladder
New research reveals why acute and chronic pain are so different – and what might make pain last
Stable cooling fostered life, rapid warming brought death: scientists use high-resolution fusuline data reveal evolutionary responses to cooling and warming
New research casts doubt on ancient drying of northern Africa’s climate
Study identifies umbilical cord blood biomarkers of early onset sepsis in preterm newborns
AI development: seeking consistency in logical structures
Want better sleep for your tween? Start with their screens
Cancer burden in neighborhoods with greater racial diversity and environmental burden
Alzheimer disease in breast cancer survivors
New method revolutionizes beta-blocker production process
Mechanism behind life-threatening cancer drug side-effect revealed
Weighted vests might help older adults meet weight loss goals, but solution for corresponding bone loss still elusive
Scientists find new way to predict how bowel cancer drugs will stop working – paving the way for smarter treatments
Breast cancer patients’ microbiome may hold key to avoiding damaging heart side-effects of cancer therapies
Exercise-induced protein revives aging muscles and bones
American College of Cardiology issues guidance on weight management drugs
Understanding the effect of bedding on thermal insulation during sleep
Cosmic signal from the very early universe will help astronomers detect the first stars
With AI, researchers find increasing immune evasion in H5N1
Study finds hidden effects of wildfires on water systems
Airborne fungal spores may help predict COVID-19 & flu surges
[Press-News.org] Groups of AI agents spontaneously form their own social norms without human help, suggests studyFirst-of-its-kind study suggests that groups of artificial intelligence language models can self-organise into societies, and are prone to tipping points in social convention, much like human societies.