AI systems spontaneously develop their own language, tailored to their sensors, rather than copying human language.
In a new large-scale study, researchers from the Artificial Intelligence Lab at VUB and the University of Namur (UNamur) demonstrate how autonomous AI systems can spontaneously develop a functional, shared language without human intervention. Unlike large language models such as ChatGPT, which learn from vast corpora of human-generated text, these AI agents build their own communication system, directly grounded in the data provided by their specific sensors and their environment. This discovery represents a major breakthrough for communication among autonomous systems, such as self-driving cars, robots, and industrial sensors.
Current AI language models focus on understanding and generating human languages such as Dutch or English. These human languages are historically rooted in our own sensory perceptions. When physical or virtual AI agents are equipped with sensors for which humans have no direct sensory equivalent—such as lidar, sonar, infrared, or chemical analysis—existing human language proves insufficient.
Jérôme Botoko Ekila, a Ph.D. student at the VUB AI Lab, explains: “Instead of forcing AI systems to communicate in a human language that doesn’t correspond to their sensors, we allow them to develop their own ‘artificial natural language.’ They construct effective concepts that correspond exactly to what they perceive through their sensors, even if the data is incomprehensible to humans.”
Since, in practice, the sensors of different machines or vehicles cannot always be calibrated identically, the researchers believe this language is the ideal solution.
Paul Van Eecke, professor of artificial intelligence at the VUB: “In industrial or physical applications, it is simply unrealistic to expect that all sensors on different robots or vehicles will be calibrated exactly the same way. Humans solve this problem through language: if I say I’m standing near the ‘red car,’ we understand each other even though our eyes don’t register exactly the same color values. Language provides a layer of abstraction above the sensor readings, and that is precisely what we are now letting these AI agents build autonomously.”
Instead of preprogrammed words or fixed codes, the researchers simply provided the agents with the basic mechanisms that underlie the evolution of human language: the local invention of words, the adaptation of conceptual prototypes, and the incorporation of feedback in the event of successful or failed communication.
What’s fascinating is that there’s no central computer imposing the rules. Each time, we let two agents interact locally with one another. It is solely through these individual interactions and updates that a common convention emerges spontaneously throughout the entire population. This principle of self-organization allows the language to optimize itself automatically: the agents create exactly the number of categories needed to perform their tasks efficiently, without any superfluous details.
The method’s universal applicability has been demonstrated across 37 diverse and complex datasets. As a result, the agents were able to develop their own vocabulary to communicate, particularly regarding physicochemical analyses of wine, NASA data on exoplanets, visual datasets, and images of birds or faces. In all these situations, the agents were able to effectively group the data into new shared concepts.
Transparency is a notable advantage of the developed systems. Unlike many “black-box” AI models, researchers can precisely determine the meaning of a word generated by the agents, since the prototype can be directly linked to the values of the underlying sensors.
Furthermore, these adaptive linguistic systems are extremely robust. When the environment changes, when sensor data is disrupted by noise, or when individual agents have different characteristics or configurations, the language automatically adapts to the new circumstances through continuous interaction.
Botoko Ekila, J., Verheyen, L., Nevens, J., Beuls, K., & Van Eecke, P. (2026). Emergent Communication in Continuous Worlds: Self-Organization of Conceptually Grounded Vocabularies at Scale. Transactions of the Association for Computational Linguistics, vol. 14, pp. 2234–2257. DOI: 10.1162/TACL.a.799
This research was conducted as part of a collaboration between the Artificial Intelligence Laboratory at the Vrije Universiteit Brussel (VUB) and the Faculty of Computer Science at the University of Namur (UNamur), and was funded by the Flemish Research Foundation (FWO) and the Belgian National Research Fund (FNRS) as part of the WEAVE project.
