artificial intelligence as the most valuable player: enabling cyber-human teams to achieve decision superiority
Key Points
- DAIRNet Theme: Human – AI Interaction
- Swinburne University of Technology and La Trobe University
- Point of contact: Prof Christopher Fluke
bottom line up front
Our transdisciplinary project examines how to support Human-AI Teams to achieve decision superiority. In order to understand the role an AI team-member can play as a monitor of individual decision-making performance, we assess the cognitive and physical state (through biometric measurements, gaze-tracking, and interaction analytics) while users perform domain-specific decision-making tasks. We have demonstrated the importance of understanding the individual, which poses challenges for training an AI. We have developed user-task-biometric profiles, providing insight on data-based features that could be learnt. Through implementation and assessment of a prototype system, we are studying how to automate the use of interventions to modify cognitive state. By understanding the characteristics of effective human cognitive performance, our original aim was to integrate an AI team-member into a Human-AI Team where the AI has characteristics of the Most Valuable Player. Our research outcomes suggest that we are producing solutions where the AI also has the characteristics of a High-Performance Coach. The next steps are to examine Human-AI Teams in Defence contexts, and study the emerging concept of “Expertise Without Precedent.”
problem addressed
Human performance in time-critical decision making is influenced by a combination of mental workload, stress, situational awareness and expertise. We measure and assess cognitive factors, physical factors, and skill level through: biometrics, eye-tracking, passive sensing, interaction analytics and self-reporting. Data is gathered while a user performs a domain-specific task.
outcomes
Our first study investigated individual task performance in order to identify the most promising indicators of cognitive and physical state, which identified the need for individual user performance models. The second study resulted in the prototyping of Cognitive-Biometric Profiles, and a decision-making performance dashboard, by assessing performance while cognitively-demanding tasks where manipulated. Relevance to Defence is new/enhanced capability to monitor, manage, identify, train/retrain human decision-making performance for screen-based vigilance, sustained attention, or visual-search-style tasks.
big picture for Defence
We are now studying team performance using C3Fire as a Command-and-Control simulation environment. Utilising data collected, we are implementing, training and testing machine learning approaches to measure real-time cognitive and physical state. We continue assessment of a Decision-Making Performance Dashboard to provide oversight on individual and team performance. We will consider the implications of the complementary roles of the AI as the MVP and AI as the Coach, as a basis for “Expertise without Precedent.” The importance to Defence is potential for improved Human-AI decision-making performance. Applications include talent identification, training/retraining, rapid workforce stand-up, with potential for future operational support.
video
Watch Chris Fluke present about this project at the 2021 DAIRNet Symposium.
