Operationalized Intent for Communication in Human-Agent Teams

Abstract

As artificial intelligent agents are employed to increase the cognitive capability of systems leveraged by human machine teams, intra-team communication presents a significant challenge. To address this challenge, we propose an intelligent agent for the express purpose of maintaining a computational representation of human intent for the multi-agent environment. This computational representation, referred to as operationalized intent, seeks to provide a designed shared mental model that can be leveraged by the humans and agents as a shared semantic space. Following the example of high performing human teams, the shared mental model is explicitly trained to both the human operators and the intelligent agents. The model of intent is represented by a hierarchy of goal statements and a summary of constraints. The model is extended by estimating future states of the intent model as part of planning activities. This projection of intent allows the intelligent agents to understand what is important to the human operator now and in the near future. This enhanced context could be used by intelligent agent designers to impart greater responsiveness and anticipatory behavior into a multiple intelligent agent environment, ideally without increasing the human’s workload. The system level implications of enhanced context are laid out in the intent architecture pattern as an aid to system designers. Finally, operationalized intent proposes a direct evaluation method to assess the agent’s interpretation of the human’s intent to evolve the design.