AI society

Emergence of communication between autonomous AI agents

Network optimizationThe next challenge in AI will probably not be about making faster computers, collecting more data, or designing adaptive robot embodiment. The key will be to allow for machines to communicate their internal states, in a process arguably similar to humans sharing about their emotions. The now very popular deep neural networks, even though extremely efficient at implementing complicated tasks, represent hundreds of thousands of parameters. Apart from looking at the outputs, no human can make sense anymore of how computations are really made inside those networks, or “how the AI thinks”. The next step will naturally be for the machines themselves to report the way they reach conclusions. In order for those reports to be understandable to humans and other machines, communication will need to be established, much like a natural language for AI.


In our research, we connect a population of neural networks together, with the task of teaching each other relevant information to solve different sets of tasks, using a limited medium. Our research aims to understand the underlying principles of the spontaneous emergence of communication, from the interaction between autonomous agents. From the connectivity between different AIs, emerges a society that coevolves with its environment. This society may acquire its own swarm mind, transitioning to a phase in which it is controlled by new sets of phenomena, as a result making them more and more independent from their hardware.


  1. Grouchy, P., D’Eleuterio, G. M., Christiansen, M. H., & Lipson, H. (2016). On The Evolutionary Origin of Symbolic Communication. Scientific Reports, 6, 34615.
  2. Witkowski, O., & Ikegami, T. (2016). Emergence of Swarming Behavior: Foraging Agents Evolve Collective Motion Based on Signaling. PloS one,11(4), e0152756.
  3. Witkowski, O. (2015). Evolution of Coordination and Communication in Groups of Embodied Agents. PhD thesis, Library for Engineering and Information Science & Technology, University of Tokyo.


Related Projects