Making and Keeping Informed Commitments in Human Machine Systems
Abstract
As the tempo, scope, and complexity of their mission environments grow, people will become increasingly reliant on delegating some decisions about situation comprehension and response to computational cognitive systems. When such a symbiotic human-machine system is functioning well, the machine will act as an extension of its human user, making decisions that agree with what the user would have done if the user could have devoted enough time and attention to making them. When a user has trust in the cognitive assistance the system provides, the user can focus on the important and time-critical tasks that require human reasoning, and safely ignore tasks delegated to a cognitive system. Our approach to reliable human-machine interaction that avoids such automation surprise involves two main thrusts. The first thrust is to enable the machine to recognize for itself when its uncertainty about the users expectations under the current circumstances justifies asking the user for clarification, and how to pose queries that efficiently acquire the most consequential information from the user about what expectations the system should commit to fulfilling. The second thrust, then, is to formulate principled computational techniques for explicitly modeling and planning in the context of such commitments to the user, to meet the users expectations while nonetheless exercising autonomy when responding to evolving mission conditions. Our projects objectives have been to make fundamental advances along these thrusts of making and keeping commitments separately, and to integrate them to solve the combined problem of querying about what commitments would be best to form given the goals and uncertainty of both the user and the system.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 08, 2021
- Accession Number
- AD1144448
Entities
People
- Edmund H. Durfee
Organizations
- Board of Regents of the University of Michigan