Two Heads Are Better Than One

Shared control — the arrangement in which a human and an AI jointly control a system — has emerged independently across domains as varied as surgery, mobility assistance, semi-autonomous driving, digital games, and creativity support. Each domain has its own terminology, its own problems, and its own design solutions. Knowledge developed in one domain is rarely available to designers working in another.

This paper surveys 55 shared-control systems across six domains and extracts a common dimension space for describing and designing them. The space has four core dimensions: the role the AI plays in the system, the degree to which the human supervises AI actions, how influence over the shared task is distributed, and how that influence is mediated between the participants.

The framework applies to both human-human and human-AI shared control, which is deliberate. The extensive literature on cooperative interfaces and computer-supported collaborative work contains decades of findings about how people coordinate shared tasks. Most of this literature has not been applied to human-AI collaboration because the framing has been different. A unified framework makes the transfer explicit.

The practical value is a design tool: a system positioned in the dimension space can be compared to other systems in adjacent positions, design patterns from other domains can be identified, and unexplored regions of the space can be treated as hypotheses about systems that do not yet exist.

Read the paper →