As the boundaries of autonomous agents and multi-agent systems continue to expand, there is an increasing need for agents to interact with humans. In fact, the field of multi-agent systems has matured from conceptual models to applications within the real-world (e.g., Energy and sustainability, disaster management, or health care). One significant challenge that arises when transitioning these conceptual models to applications is addressing the inevitable human interaction. To this end, this workshop examines major challenges at the intersection of human-agent systems. In particular, we focus on the challenges of designing and modelling human-agent interaction. While the former takes a human-centric view of human-agent systems and focuses on the design of human-agent coordination mechanisms, trust issues in human-agent interaction, interaction techniques, and human activity recognition, the latter is concerned with finding better models of human behaviour in a variety of settings so that autonomous and multi-agent systems can appropriately interact with human agents (e.g., agent-human negotiation strategies or health care agents encouraging physical therapy for a variety of recovering patients).
Hence, this workshop aims to establish a forum for researchers to discuss common issues that arise in designing and modelling human-agent interaction in different domains.
In designing multi-agent systems applications where such applications involve humans, it is important to consider the key principles by which the interaction between agents and humans will be established. In particular, the technical issues to be addressed by researchers, and which will be the key discussion points, at this workshop include but are not limited to:
- Flexible autonomy – how should the delegation of tasks to agents be performed such that the right degree of autonomy be given to individual agents or teams of such agents to optimise the performance of tasks by a human controller or teams of humans interacting with (teams of) agents.
- Trust between humans and agents – when humans delegate tasks to agents or vice-versa, they need to be able to capture the uncertainty in the other party being able to correctly complete tasks. Such uncertainty may be modelled using past interactions (trust) or information gathered from other agents (reputation). Developing effective trust and reputation models and mechanisms for human-agent interaction is therefore key to establishing long term relationships between humans and agents.
- Presentation and interaction techniques – to allow users to understand the actions of large collections of agents as they reason and act on behalf of users.
- Human activity recognition – to recognise human activity to allow agents to reason about human interaction so they might exploit this understanding to augment and support the action of users.
- Enhanced models of human behaviour and theory of human behaviour
- Relationships between human behaviour models and their assumption
- Comparison of approaches in applying models of human behaviour (e.g., bounded rational or psychological models)
- Applications of human behaviour models
- Cooperative and competitive agent-human systems
- Behavioural game theory
- Techniques for learning human behaviour (e.g., machine learning, crowdsourcing, and human computation)
- Evaluation techniques for models of human behaviour
- Techniques for model selection or augmenting agent learning through human modelling
- Benchmarks and evaluation methodologies for evaluating agent-human interactions
January 30, February 9, 2013 – Submission deadline.
– March 8, 2013 – Notification of acceptance.
– May 6-7, 2013 – Workshop takes place.