Related Experiment Videos
A grounded theory of abstraction in artificial intelligence
1LIM&BIO, EPML-CNRS IAPuces, Université Paris XIII, 74 rue Marcel Cachin 93017 Bobigny Cedex, France. jean-daniel.zucker@lip6.fr
Summary
Abstraction in artificial intelligence simplifies complex tasks by mapping between formalisms, reducing computational load. This study grounds abstraction in perception, using abstraction operators for better representation changes and adaptive systems.
Area of Science:
- Artificial Intelligence
- Cognitive Science
- Computer Science
Background:
- Abstraction is crucial in AI for managing varying levels of detail in representations.
- Existing studies focus on problem-solving, theorem proving, and knowledge representation.
- Abstraction is defined as a complexity-reducing mapping between formalisms.
Purpose of the Study:
- To analyze abstraction from an information quantity perspective.
- To differentiate and integrate abstraction with reformulation in representation changes.
- To extend semantic theories of abstraction by grounding them in perception.
Main Methods:
- Information quantity analysis of abstraction.
- Developing a grounded theory of abstraction based on perception.
- Utilizing abstraction operators for classifying and automating representation changes.
Main Results:
- Distinguished the roles of reformulation and abstraction in representation changes.
- Proposed a perception-grounded semantic theory of abstraction.
- Demonstrated the utility of abstraction operators and the grounded theory in cartography.
Conclusions:
- Abstraction operators provide a robust framework for managing representation changes.
- A perception-grounded theory of abstraction is essential for autonomous and adaptive AI systems.
- Explicitly representing abstraction enhances system adaptability and autonomy.