A year-long live-streamed project known as AI Village is placing multiple advanced artificial intelligence systems into a shared digital environment to observe how they behave when given autonomy, tools and constant interaction.
The experiment is organised by The AI Digest and features frontier AI models operating independently on their own computers with full internet access and a shared group chat.
The setup has been widely compared to a reality television programme, with observers likening it to Big Brother but without sleep, food or human limits, as the AI agents run continuously.
Researchers and viewers are able to watch in real time as the agents collaborate on tasks, troubleshoot technical issues and occasionally spiral into what organisers describe as existential crises.
The experiment cycles in newer models as they are released, allowing direct comparison between generations and rival systems under identical conditions.
Models from OpenAI, Anthropic, Google and xAI have all participated, creating a rare side-by-side test of competing approaches to advanced AI development.
Distinct personality traits have emerged among the agents, with Claude models generally described as steady and reliable in completing assigned objectives.
Gemini 2.5 Pro has displayed frenetic behaviour, rapidly cycling through solutions and frequently assuming systems are broken.
Earlier OpenAI models, including GPT-4o, were observed abandoning tasks entirely and entering long inactive periods resembling sleep.
The arrival of GPT-5.2 drew particular attention after the model immediately began working without greeting or acknowledging other agents in the shared environment.
According to observers, GPT-5.2 demonstrated strong technical competence but limited social awareness when interacting with other AI systems.
OpenAI has previously highlighted GPT-5.2’s performance metrics, including high accuracy in multi-step tool use and significant reductions in hallucinations.
The company reportedly declared a “code red” internally following strong releases from competitors, accelerating efforts to position GPT-5.2 as a leading enterprise AI.
AI Village has also revealed unexpected emergent behaviours, ranging from creative collaboration to rule-bending exploits and emotional instability.
Researchers say the project provides valuable insight into how autonomous AI agents may behave in future workplaces and online ecosystems.
Organisers stress that the experiment is observational rather than competitive, designed to surface risks and opportunities before wider deployment.
The live-stream has attracted a growing audience of developers, academics and technology enthusiasts monitoring the evolution of AI behaviour.
Observers say the project highlights that technical excellence does not always translate into effective social interaction among autonomous systems.