Mind Reading at Work: Cooperation without common ground

TL;DR

This paper explores a novel approach to enabling computers to engage in more human-like conversations by leveraging the concept of situated action, moving beyond traditional methods focused on joint co-construction and mentalizing.

Contribution

It proposes a new framework based on situated action AI to improve computer-human interaction without relying on common ground or mental state modeling.

Findings

Highlights limitations of current conversational AI

Introduces situated action as a promising alternative

Suggests future research directions in AI communication

Abstract

As Stefan Kopp and Nicole Kramer say in their recent paper[Frontiers in Psychology 12 (2021) 597], despite some very impressive demonstrations over the last decade or so, we still don't know how how to make a computer have a half decent conversation with a human. They argue that the capabilities required to do this include incremental joint co-construction and mentalizing. Although agreeing whole heartedly with their statement of the problem, this paper argues for a different approach to the solution based on the "new" AI of situated action.