An Integrated Theory of Human-Machine Teaming
Abstract
The rapid growth in artificial intelligence brings to life the possibility that individual and team performance could one day exhibit a level of collective intelligence never seen before. However, our current approaches to team science and technology development lead to a level of conceptual fragmentation that undercuts the possibilities for the synergistic gains we seek. If each area continues to progress in isolation, we will continue to develop AI that elicits Òalgorithm aversionÓ or fails to be tuned to the needs and communications of human users. In this proposal, we seek to propel team science and AI forward by integrating existing work on individual and team cognition into a socio-cognitive architecture that will enable further integration with developing work on Machine Theory of Mind (M-Tom) to yield an Integrated Theory of Human-Machine Teaming. Our work will unfold in three major thrusts, starting with (i) individual-level task performance and coaching; then building into (ii) an integrated framework of human-machine teaming; both of which are developed alongside (iii) testable hypothesis and measure development. Specifically, (i) we will draw together the literature on individual cognition and identify the qualities necessary for synthetic teammates to scaffold cognition for individuals (M-ToM). These inputs will be used to develop computational representations of the cognitive processes involved in individual dynamic decision making and problem solving tasks. On the basis of our theory and modeling we will provide guidance to TA1 on the development of synthetic coaches, and TA3 on the design and metrics to use in human experiments; (ii) drawing on observations from the first phase of experiments, we will develop an integrated framework to guide the specification of novel and testable hypotheses about human-machine teaming and human social cognition. Based on phase 1, we will scale up representations to team cognition and collective intelligence; and, (iii) we will continuously be drawing on existing data and new results to refine the measures that can be used for synthetic teammates and coaches to continuously monitor team behavior and intervene when helpful. These will provide the basis for ongoing development of AI in the context of human-machine systems. If we are successful in accomplishing the work proposed it will constitute a huge advance to team science and research on human-machine teaming, resulting in an Integrated Theory of Human-Machine Teaming to guide future develop of machine-based tools for enhancing collective intelligence, along with a suite of measures, metrics, and research tools for researchers to continue to contribute to this effort. The level of integration required across the fields of team science, cognitive psychology, computer science, and human-computer interaction is considerable, and the need to accomplish this integration comprises one of the major risks the program faces in being successful. A second major risk we face is the complexity in the cognitive modeling we propose to do which builds from the individual to the team level in generating and refining our hypotheses and theory. However, our interdisciplinary team is uniquely well-equipped to accomplish these goals. We have leading experts in the areas of team science, cognitive science, computer science and human-computer interaction who have track records of successful interdisciplinary collaboration and precisely the expertise necessary to push the science of human-machine teaming to the next level.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Dec 04, 2019
- Source ID
- W911NF2010006
Entities
People
- Anita Woolley
Organizations
- Army Contracting Command
- Defense Advanced Research Projects Agency
- Massachusetts Institute of Technology