At the Sony Computer Science Laboratories Paris, we have pioneered new ways to address these questions by not studying human languages directly, but by discovering how robots and artificial agents can invent their own languages and concepts from the ground up.
This installation revisits one of the lab’s most iconic experiments on the origins and evolution of language: the Talking Heads experiment. In this experiment, artificial language users teleported over the internet to different locations, ranging from Paris to Tokyo, to play “language games” with each other about geometric shapes they could observe through a camera.
The agents start the experiment without any language or even concepts, but with the capacity of creating new meanings and words for expressing those meanings. While failing terribly at communication during their first interactions, the agents gradually manage to self-organize their own languages in a similar fashion as in which ant colonies are able to spontaneously form paths to food sources.
In the installation human users could create their own artificial agent, and play language games with them – both teaching them new words or learning the artificial language.
Even though this groundbreaking experiment dates from 1999, it still holds up as one of the world’s first and largest experiments involving distributed systems. It also still outperforms recent attempts at creating artificial neural systems that try to grapple with open-ended meaning spaces and open-ended linguistic environments.