Tasking Interfaces: Associate Systems that know Who’s the Boss

Keywords: UAV, shared task model, tasking intefaces

Abstract: Despite substantial technological success, “associate” systems continue to suffer from a fundamental, sociological problem— namely, that human operators of advanced automation are used to being in charge. We have recently begun adapting techniques from our associate research to the construction of “tasking” interfaces which enable the kind of interaction operators are used to having with intelligent, informed subordinates. Instead of being autonomous, or even truly mixedinitiative, “tasked” systems are always subordinate—but they know enough about the tasks in the domain that instructing them is vastly easier than instructing traditional automation systems. We use a shared task model to give advanced automation systems (e.g., Unmanned Air Vehicles—UAVs) the same task and goal understanding that the human has. When combined with a planner, the resulting system permits ‘tasking’ at all the various levels an intelligent subordinate should be able to accept: exhaustively specified plans to be performed exactly, partial plans which leave the planner free to create any full plan which includes those pieces, constraints which require the planner to stay away from certain methods or resources, or high-level goals for the automation to achieve however it thinks best. The net result is a human-machine system which is almost as capable as an associate, and which reduces human workload almost as much, yet which leaves the human more in control than an associate does.