modelling, monitoring and moderating human-AI interaction
Key Points
- DAIRNet Theme: Human – AI Interaction
- University of Adelaide
- Point of contact: Prof Anna Ma-Wyatt
Geospatial intelligence is currently reliant on integration of information from diverse sensors and sources. The amount of data that is available through these data gathering networks is rapidly expanding. In each case, a human operator plays a critical role in aggregating and assessing information and using it to update situational awareness and inform decisions. However, humans have a limited capacity to process information and with the advent of more sophisticated AI to assist a human, human-AI interaction will be a critically fragile step.
A key problem is how to make human-AI interaction effective, resilient and less fragile. We envisage a system in which an AI agent augments human capability by scaffolding their inferences through dialogue. In this project, we develop a novel approach that synthesises advanced modelling paradigms together with non-invasive behavioural measures to model, monitor and moderate human-AI interaction.
The figure provides an overview of our approach, integrating information from sensors and the human, using that to make inferences about tasks and tasking, and providing a framework for a human and AI agent to trade information to increase their common understanding.
The overarching long-term goal of this project is to reach the following end state:
- Formation of a transdisciplinary team combining intelligence analysts, artificial intelligence, human factors and behavioural scientists working together to continue to exploit dialogue management capabilities.
- A dialogue management plug-in that is resilient and enhances the ability of intelligence analysts to use blended information across human, information and physical realms for certain operational contexts of interest to Defence and Defence stakeholders.
- Methodologies that can be used to train analysts to use the dialogue management plug in.
- Development of a system that monitors, moderates and helps model human-AI interaction, used in an operational environment to make resilient hybrid human-AI teams.
