This workshop concerns the "taskability" of cognitive agents, meaning the ability of an intelligent and adaptable system to accept high level, task-oriented instructions from a human and translate those goal-oriented directives into a suitably decomposed plan of sensing, reasoning, or action. The current state of the art in research for cognitive systems and agents in general allows human handlers to issue directions to agents either in terms of lower level action primitives that correspond to the agent's particular design, or constrained by a higher-order action language that is engineered to suffice as a shorthand for some sequence of the same kinds of agent-specific behaviors.
This workshop is addressing four key considerations for this emergent topic: 1) defining this new research area with sufficient clarity that research agendas can proceed productively; 2) identifying the science and technology issues that most be addressed to create systems that meet that definition of taskability; 3) decomposing that further into research and development needs that are the precursors to such research; and 4) exploring the formation of a research community on taskability. Taskability holds the potential to transform the way humans interact with intelligent systems. The ability to instruct a wide variety of agents to perform tasks that are not preprogrammed will have a profound effect on the broader AI enterprise, and eventually on the technologies that will enrich the daily lives of people.
Attendees: John Anderson (Carnegie Mellon University), Ken Forbus (Northwestern University), Kevin Gluck (Air Force Research Laboratory), Chad Jenkins (Brown University), John Laird (University of Michigan), Christian Lebiere (Carnegie Mellon University), Dario Salvucci (Drexel University), Matthias Scheutz (Tufts University), Andrea Thomaz (Georgia Institute of Technology), Greg Trafton (Naval Research Laboratory), Bob Wray (Soar Technology) The workshop was held in Ann Arbor, MI on May 12-13, 2014. The workshop brought together senior researchers interested in interactive task learning – computational systems that can learn completely new tasks from interaction with human teachers. The goal of the workshop was to explore this topic as a potential research problem and to take the first steps to build a community of researchers that would take the lead in establishing it as a research area. The body of this report introduces the rationale for the workshop and then provides descriptions of the discussions that were held at the workshop. The overall conclusion of the workshop is that this is an area of research that should be aggressively pursued. All of the participants were deeply engaged in the discussions and committed to continuing development of this area of research. We generated both concrete and conceptual steps to pursue in the immediate future. One concrete result of the workshop was that we agreed to call this area of research "Interactive Task Learning" (ITL) and that label is used throughout the rest of the report (in place of "Taskability"). There is a final report available at: www.eecs.umich.edu/~soar/sitemaker/docs/pubs/ITL_Report_NSF_1419590.pdf
