Foundations of Autonomy and Its (Cyber) Threats: From Individuals to Interdependence
Papers from the 2015 AAAI Spring Symposium
Ranjeev Mittu, Gavin Taylor, Don Sofge, William F. Lawless, Program Chair
Technical Report SS-15-02
Published by The AAAI Press, Palo Alto, California.
Approaches using artificial intelligence (AI) may soon manage complex systems with teams, including hybrid teams composed arbitrarily of humans, machines, and robots. Already, AI has been useful in modeling the defense of individuals, teams, and institutions, as well as the management of social systems such as virtual ship bridges. However, foundational problems remain in the continuing development of AI for team autonomy, especially with objective measures able to optimize team function, performance and composition.
AI approaches often attempt to address autonomy by modeling aspects of human decision-making or behavior. Behavioral theory is either based on modeling the individual, such as through cognitive architectures or, more rarely, through group dynamics and interdependence theory. Approaches focusing on the individual assume that individuals are more stable than the social interactions in which they engage. Interdependence theory assumes the opposite, that a state of mutual dependence among participants in an interaction affects the individual and group beliefs and behaviors of participants. The latter is conceptually more complex, but both approaches must satisfy the demand for predictable outcomes as autonomous teams grow in importance and number.
Despite its theoretical complexity, including the inherent uncertainty and nonlinearity exposed by interdependence, we argue that complex autonomous systems must consider multiagent interactions to develop predictable, effective and efficient hybrid teams. Important examples include cases of supervised autonomy, where a human oversees several interdependent autonomous systems; where an autonomous agent is working with a team of humans, such as in a network cyber defense; or where the agent is intended to replace effective, but traditionally worker-intensive team tasks, such as warehousing and shipping. Autonomous agents that seek to fill these roles, but do not consider the interplay between the participating entities, will likely disappoint.
Our symposium offers opportunities with AI to address these and other fundamental issues about autonomy, including its application to hybrids at the individual, group and system levels.