Language as a cognitive and social tool at the time of large language models

The interaction of language with perception, action, and cognition has been a recurring theme across the cognitive sciences. Recently, research has emphasized the role of language not only as a communication device but also as a cognitive tool. The rise of large language models (LLMs) and their recent integration in embodied AI systems perceiving and acting in physical and social environments (robotics) is facing similar issues and has the potential to offer insights also into the relationship between language and human cognition. Here we chart parallel developments in these fields with the aim of fostering cross-talk and collaboration. First, we discuss work in psychology and neuroscience on language grounding and the effects of language on perception, action, and other cognitive processes. Next, we explore recent work in machine learning and robotics to design language-conditioned robots that comprehend and act on linguistic instructions, which provide new mechanistic insights on language effects. We conclude by arguing that these new parallel trends of research are finally able to capture the importance of language, in its experiential aspects that involve social interaction, thought, and behavior control.