Situation Awareness-based Agent Transparency Research Articles

Agent transparency in mixed-initiative multi-UxV control: How should intelligent agent collaborators speak their minds?

Multiple unmanned system control is a complex command and control endeavor but pairing human operators with an intelligent agent (IA) teammate can buttress the collection and synthesis of data and improve complex decision making. Effective human-autonomy teams (HATs) require human trust in IA teammates to be properly calibrated, which can be supported by communications pertaining to underlying functions of the IA, or “transparency”. One prominent guide for application of transparency is Chen and colleague's Situation awareness-based Agent Transparency (SAT) model. This effort sought to extend understanding of the application of this model by manipulating secondary transparency communication parameters: face threat (i.e., threat to a person's sense of social standing) and design of transparency communication (verbal, graphical, and iconographical). Results revealed that increasing face threat can improve reliance calibration at low transparency but may be detrimental when transparency is high. Outcomes concerning the method of transparency communication suggest that while verbal communication of transparency information is sufficient and even preferred when a low level of transparency is provided, reliance on graphical and iconographical approaches for presenting transparency information increases at a higher level of transparency.

XFlag: Explainable Fake News Detection Model on Social Media

Social media allows any individual to disseminate information without third-party restrictions, making it difficult to verify the authenticity of a source. The proliferation of fake news has severely affected people’s intentions and behaviors in trusting online sources. Applying AI approaches for fake news detection on social media is the focus of recent research, most of which, however, focuses on enhancing AI performance. This study proposes XFlag, an innovative explainable AI (XAI) framework which uses long short-term memory (LSTM) model to identify fake news articles, layer-wise relevance propagation (LRP) algorithm to explain the fake news detection model based on LSTM, and situation awareness-based agent transparency (SAT) model to increase transparency in human-AI interaction. The developed XFlag framework has been empirically validated. The findings suggest the use of XFlag supports users in understanding system goals (perception), justifying system decisions (comprehension), and predicting system uncertainty (projection), with little cost of perceived cognitive workload.

Transparency for a Workload-Adaptive Cognitive Agent in a Manned–Unmanned Teaming Application

This study focuses on the transparent design of a cognitive agent to enhance situation awareness in two aspects of a human-agent teaming application: assisted system management and mixed-initiative mission planning. Adaptive and complex agent behavior might result in the failure to comprehend resulting interventions, a decrease in trust, and a loss of overall situation awareness. This study describes and validates a concept for transparent agent design by adopting the transparency strategies proposed by the “situation awareness-based agent transparency model.” The overall objective was to improve the human operator's perception, comprehension, and projection of the agent's support. The concept was applied to the prototype of a workload-adaptive cognitive agent, which supports a helicopter crew during mission planning and execution in complex and dynamically changing multi-vehicle missions. A human-in-the-loop experiment revealed enhancements in situation awareness and performance. Subjective trust measures implied an increase in human-like characteristics of the cognitive agent. The results and the potential for further research are discussed.

Application of the GORITE BDI Framework to Human-Autonomy Teaming: A Case Study

Human-Autonomy Teaming (HAT) is of growing interest in the military sector, particularly in its application to war gaming using semi-automated computer generated forces (CGF). In these applications, one or more operators manage multiple semi-autonomous game entities. If effective collaboration (teaming) is to occur between operators and entities, then having effective interaction models is essential if the levels of trust and explanatory capability required for military operations are to be delivered. The Situation Awareness-Based Agent Transparency (SAT) Model has been identified as providing a suitable conceptual framework for such models. However, while the SAT model is informed by the Belief-Desire-Intention (BDI) model of agency, to date there has been no implementation of an interaction model at the level of desires and intentions, i.e. goals. In this paper, we propose that GORITE, a novel BDI framework that employs explicit goal representations and a shared data context for goal execution, provides a suitable platform for the development of SAT-enabled agents. The feasibility of this proposition is demonstrated through the development of a simple but representative CGF case study.

Situation awareness-based agent transparency and human-autonomy teaming effectiveness

ABSTRACTEffective collaboration between humans and agents depends on humans maintaining an appropriate understanding of and calibrated trust in the judgment of their agent counterparts. The Situation Awareness-based Agent Transparency (SAT) model was proposed to support human awareness in human–agent teams. As agents transition from tools to artificial teammates, an expansion of the model is necessary to support teamwork paradigms, which require bidirectional transparency. We propose that an updated model can better inform human–agent interaction in paradigms involving more advanced agent teammates. This paper describes the model's use in three programmes of research, which exemplify the utility of the model in different contexts – an autonomous squad member, a mediator between a human and multiple subordinate robots, and a plan recommendation agent. Through this review, we show that the SAT model continues to be an effective tool for facilitating shared understanding and proper calibration of trust in human–agent teams.

Trust between Humans and Learning Machines: Developing the Gray Box

This article explores the notion of the ‘Gray Box’ to symbolize the idea of providing sufficient information about the learning technology to establish trust. The term system is used throughout this article to represent an intelligent agent, robot, or other form of automation that possesses both decision initiative and authority to act. The article also discusses a proposed and tested Situation Awareness-based Agent Transparency (SAT) model, which posits that users need to understand the system’s perception, comprehension, and projection of a situation. One of the key challenges is that a learning system may adopt behavior that is difficult to understand and challenging to condense to traditional if-then statements. Without a shared semantic space, the system will have little basis for communicating with the human. One of the key recommendations of this article is that there is a need to provide learning systems with transparency as to the state of the human operator, including their momentary capabilities and potential impact of changes in task allocation and teaming approach.

Using agent transparency to support situation awareness of the Autonomous Squad Member

Agent transparency has been proposed as a solution to the problem of facilitating operators’ situation awareness in human-robot teams. Sixty participants performed a dual monitoring task, monitoring both an intelligent, autonomous robot teammate and performing threat detection in a virtual environment. The robot displayed four different interfaces, corresponding to information from the Situation awareness-based Agent Transparency (SAT) model. Participants’ situation awareness of the robot, confidence in their situation awareness, trust in the robot, workload, cognitive processing, and perceived usability of the robot displays were assessed. Results indicate that participants using interfaces corresponding to higher SAT level had greater situation awareness, cognitive processing, and trust in the robot than when they viewed lower level SAT interfaces. No differences in workload or perceived usability of the display were detected. Based on these findings, we observed that transparency has a significant effect on situation awareness, trust, and cognitive processing.

Agent Transparency and the Autonomous Squad Member

The present study investigated the effect of including information to support transparency, based on the Situation awareness-based Agent Transparency (SAT) model, in the interface for an autonomous agent known as the Autonomous Squad Member (ASM). In four different SAT model-based display conditions, participants used the ASM’s interface to gain information about the ASM and simulated squad’s status as they completed a route containing obstacles. Results indicated that participants had greater trust in the ASM and were most effective at maintaining situation awareness when the ASM provided outcome predictions (SAT Level 3), in addition to planned actions (SAT Level 1) and a rationale for those actions (SAT Level 2). No differences in participant workload, while monitoring the ASM, were observed between SAT display conditions.

The Effects of Agent Transparency on Human Interaction with an Autonomous Robotic Agent

We use the Situation awareness-based Agent Transparency model as a framework to design a user interface to support agent transparency. Participants were instructed to supervise an autonomous robotic agent as it traversed simulated urban environments. During this task, participants were exposed to one of three levels of information used to support agent transparency in the interface display. Our findings suggest that providing agent transparency information allows operators to properly calibrate trust without excess workload. Though, increased agent transparency information did not support operator situation awareness.