Abstract

This chapter introduces the Grounded Turing Test, a behavioral benchmark for evaluating intelligence in embodied agents. While modern large language models can convincingly mimic human dialogue and surpass aspects of Turing’s original test, they remain fundamentally disembodied. We argue that true intelligence requires a physical presence in space and time, enabling high-dimensional sensory processing and goal-directed motor actions. Building on insights from robotics, cognitive science, and information theory, we outline how language, when grounded in physical interaction, can serve as a window into intelligent behavior. The chapter proposes a framework for embodied AI that integrates human-robot communication, collaborative planning, compositional learning, and adaptive neural structures, moving beyond static datasets toward continuous, online learning. Passing the Grounded Turing Test is not necessary for intelligence—many animals exhibit intelligent behavior without language—but it is a sufficient condition for embodied artificial general intelligence.