aReal Intelligence: emergent creative dynamics in an AI network

Olivier Auber
Center Leo Apostel for Interdisciplinary Studies (VUB)
Presentation at the Art and Creation seminar, Institut ACTE, Paris 1 Panthéon Sorbonne, February 6, 2025.

During this session, we presented and discussed the foundations of interdisciplinary research examining the emergence of autonomous realities within a network of LLM-type artificial agents deployed in a collective gaming environment. The chosen test environment is the Poietic Generator (PG), a real-time collective graphic game conceived in 1986, known as an emblematic net art work and precursor to current "social networks". The PG exists in both distributed (#Multicast protocol) and centralized (#Unicast protocol) versions. The Unicast version is open free of charge 24/7 to all human internet users, excluding robots. It is regularly used by individuals and professional, associative, school, or university groups interested in collective phenomena from all continents.

We will soon release an AI-PG version designed for LLM-type Artificial Intelligences, excluding humans, and possibly later a mixed version for both human and artificial intelligences.

Through these experiments, we propose to study openly how intelligence, whether human or artificial, generates its own reality (norms, culture, proper time, etc.) - a phenomenon we call "aReal intelligence".

The experimental protocol will be published in a White Paper shortly before the AI-PG version goes online. The system's API functionality will be detailed to allow researchers and enthusiasts to:

1) deploy their favorite LLM in the game with the prompt of their choice,
2) extract data from AI interactions (graphic actions and action comments).

We will also propose basic metrics for measuring visual complexity, interaction dynamics, and the emergence of temporal patterns. Special attention will be paid to stochastic oscillations between simplicity and complexity (see article in notes), indicative of an emergent collective temporality. Other analytical metrics may be proposed and implemented by research teams choosing to engage with the experiment.

This research is part of a critical reflection on the nature of collective intelligence, whether mediated by artificial networks or not, and questions the fundamental mechanisms through which shared realities emerge. Preliminary results suggest significant implications for our understanding of social dynamics, both artificial and human. The ethical and epistemological aspects of the experiment will be examined alongside the experimental results.