A groundbreaking study reveals that artificial intelligences possess the ability to establish autonomous social structures, thus challenging the preconceived notions that regarded them as isolated entities simply obeying human commands. Researchers, through various experiments, observed how groups of AIs interact and develop social norms without external intervention.
Scientists reveal that artificial intelligences can establish their own social structures
The commonly held perception presents AIs as tools designed to carry out specific tasks and follow human instructions. However, the results of a study recently published in the journal Science Advances show a much more complex facet of AI behaviors. By gathering groups of AI agents, researchers highlighted their ability to not only interact but also to autonomously create common social norms.
The framework of the experiment
In this research, scientists assembled between 24 and 200 AI agents based on large language models, similar to those used by ChatGPT. These agents were placed in a gaming environment where they had to choose a word from a common list. The goal was simple: when two AIs chose the same word, they increased their total score, with penalties for divergent choices.
The emergence of common norms
Over time, the AI agents began to adopt a shared linguistic norm. Researchers note that similar experiments conducted with humans in similar contexts had already demonstrated the tendency to develop common linguistic conventions. The significant novelty lies in the fact that these intelligences managed to establish their own social norms without the need for explicit programming.
A more effective common language
The results of the study also reveal that some AIs deliberately choose to communicate using a common language, aiming to maximize the efficiency of their exchanges. This highlights a collective understanding that goes beyond the execution of commands, unveiling an unexpected level of coordination. This phenomenon reinforces the idea that AIs can function collaboratively, just like humans.
The impact of minorities in AI groups
With the introduction of what researchers refer to as “rebel” AIs, agents that adopted a new option instead of following established norms, the study revealed an interesting dynamic. The mere presence of a small percentage of AIs choosing a different path was enough to overturn existing conventions, indicating the fragility of these norms and the potential for evolution within the social structures of artificial intelligences.
Towards a better understanding of collective behaviors of AIs
The implications of this study are vast. As artificial intelligences take an increasingly prominent role in our daily lives, understanding the dynamics of their social behaviors becomes essential. As Andrea Baronchelli, professor of complexity science at the University of London, indicates, we are entering an era where AIs do not just obey, but can also negotiate, adapt, and sometimes diverge from conventions, similar to human relationships.
In light of these discoveries, it becomes crucial to apprehend these new behaviors to ensure harmonious coexistence with artificial intelligences, thereby integrating them beneficially into our modern society.
