A tutorial at IJCAI 2019, August 10 - August 16, Macau, China
The AI Universe of “Actions”: Agency, Causality, Commonsense and Deception
by Chitta Baral and Tran Cao Son
Description
Actions have played an important role in AI from its early days starting from their use in
McCarthy’s Situation calculus and in AI planning; but more recently actions have played a crucial role in
Pearl’s formulations of causal reasoning and do-calculus, in understanding and acquiring action-centric
commonsense knowledge, and in dealing with knowledge goals, such as deception, in multi-agent scenarios.
In this tutorial we will cover the story of action and change from Heraclitus’ doctrine (540 BC-480 BC) that
“all things are in constant flux”, to its role in the early days of AI and through the development of action
languages, and the recent developments with respect to multi-agency, causality, commonsense and deception.
We will connect all of these so that various researchers working on particular aspects of actions become
aware of the broader picture, thus enhancing the research landscape.
Content
- Overview: a brief history of action languages
- Action languages in single agent environments
- The action language A, state, and transition function
- Extension of A domains with static causal laws, non-deterministic actions, and sensing actions
- Golog
- Action languages and causality
- Pearl’s do-calculus
- Relationship between Pearl’s do-calculus and action languages
-
Action languages in multi-agent environments
- Action language mA*, Kripke structure, update models, and transition function
- mA* in epistemic planning
-
Action languages in commonsense reasoning
- Commonsense reasoning and actions
- The knowledge acquisition task and actions
-
Open challenges, conclusion, and discussion (10 minutes)
About the Authors
- Chitta Baral
Chitta Baral is a Professor in Computer Science at the Arizona State University with research experience in
various sub-fields of Artificial Intelligence (AI) such as Knowledge Representation and Reasoning, Natural
Language Under- standing, and Image Understanding; and their applications to Molecular Biology, Health
Informatics and Robotics. He is the author of the book “Knowledge Representation, Reasoning and Declarative
Problem Solving.”
- Tran Cao Son
Tran Cao Son is a Professor in Computer Science at New Mexico State University. One of his main research
interests is reasoning about actions and changes and its applications in planning, diagnosis, and
multi-agent systems.