Telerobotics

Hannaford's 1990s framework for telerobotic "traded control" described a system where authority over a robot's motion passes back and forth between human and computer, axis by axis, moment by moment — a million possible configurations of shared agency on six degrees of freedom. The architecture assumed the human would always be there, hand on the master controller, choosing when to steer and when to let the machine compensate. Now look at what actually happened: the operator didn't just step back from individual axes, they left the room. Every autonomous AI agent deployed today — every coding assistant, every agentic workflow, every LLM with tool use — is a telerobotic system running in the mode Hannaford never labeled: traded control where the trade became permanent. The epistemological problem Goldberg raised in 1999 (can you trust what a telerobot shows you?) has quietly inverted. The telerobot no longer needs to convince the operator the remote environment is real. It needs to convince itself the operator still exists. The coffeepot camera at Cambridge was the first telerobot, and it watched an object that was either full or empty — a binary epistemology. We are now building systems with a million modes of operation and no one selecting among them.