Commander s Intent: Symbiotic Man-Machine Decision-Making

Abstract

We propose to develop a machine learning model that identifies the commander~s intent and how that intent changes over tasks, decisions, and interactions with other decision-makers by learning a minimal representation that decomposes the decision-space. We define Operational Intent as theboundary between Perspectives which represent the problem being solved and Content-Specific Context which represents the decisions and factors taken to solve the problem. Machine learning algorithms thus far have ignored intent and the process of decision-making in favor of simplicity of training and simply minimizing loss functions of the final answer. This is akin to assuming thata high school mathematics student knows how to solve a problem because they guess the correct answer, instead of checking that they had the right process. Artificial intelligence driven by machine learning models have not considered intent as a process, and therefore, have been unable to recognize perspectives and content-specific context and have been unable to generalize what to do when faced with even minor perturbations of context. Without this approach, machine learning methods will be unable to understand and model human behavior, whereas we can not only capture it, but provide reasonable explanations of why. Our decomposition is expected to yield moreaccurate predictions of behavior and can explicitly recognize what was important in making decisions, so it can then explain them. Simultaneously, this approach is expected to advance machine learning by teaching machines how to generalize like real human decision-makers do because it observes the process of making decisions. A target deliverable for this project is a method for performing assessments of Commander~s intent with respect to missions that provides individualized, proactive decision support: identifying salient mission information and health of the mission (what); describing likely decisions andcontingencies to be resilient based upon prior experience (how); and identifying when the mission and Commander~s intent drift (when and where).

Document Details

Document Type
DoD Grant Award
Publication Date
May 23, 2019
Source ID
N000141912211

Entities

People

  • Eugene Santos

Organizations

  • Board of Trustees of Dartmouth College
  • Office of Naval Research
  • United States Navy

Tags

Fields of Study

  • Computer science

Readers

  • Neural Network Machine Learning.
  • Systems Analysis and Design
  • Team-Based Human-Centered Cognitive Task Decision Making and Information Performance.

Technology Areas

  • AI & ML
  • AI & ML - DoD AI Strategy
  • AI & ML - Neural Networks
  • Space