About the project
This PhD project explores how artificial agents can autonomously develop symbolic communication systems, resembling human alphabets or logograms, purely through interaction. Building on previous studies in sketch-based communication, the project investigates how agents might evolve compact, meaningful symbols, potentially mirroring early human writing systems, that enable efficient exchange of information.
Studying written communication protocols that emerge between artificial agents, such as those modelled through Deep Neural Networks, provides fundamental insights into how language and symbolic systems could develop in autonomous systems purely through interaction. Previous studies on sketching as a form of communication, emphasize that symbols developed by autonomous agents can communicate essential information effectively.
This project investigates how artificial agents, leveraging differentiable sketching, can autonomously evolve symbolic communication systems that resemble human alphabets, logograms, or syllabic representations. The project draws on research from emergent communication of multi-agent systems, cognitive science, and semiotics, contributing to a better understanding of how communication conventions can form without direct human programming. Such a framework could enable studying symbolic abstraction and compositionality in written language, which has been central to human communication since the earliest logographic systems like Egyptian hieroglyphs and Chinese characters.
In artificial agents, several constraints—such as task-oriented goals, communication efficiency, and environmental cues—could drive the convergence on compact, recognizable symbols for frequently used concepts, eventually leading to a logo-graphic protocol that mirrors early written languages.
The findings could significantly impact artificial intelligence (AI) interpretability and interaction, enabling agents to communicate efficiently with humans using symbolic sketches that intuitively convey information. Furthermore, the research has applications in AI-human collaboration, potentially enhancing communication protocols in fields requiring rapid, interpretative exchanges. By investigating the role of self-organization and adaptation in symbol formation, this project may also inform cognitive science and artificial intelligence studies on language evolution and communication.
