Authors:
Martyn Lloyd-Kelly
1
;
Fernand Gobet
1
and
Peter C. R. Lane
2
Affiliations:
1
University of Liverpool, United Kingdom
;
2
University of Hertfordshire, United Kingdom
Keyword(s):
Agents, Simulation, Dual-process Theory, Reinforcement Learning, Pattern Recognition, Chunking.
Related
Ontology
Subjects/Areas/Topics:
Agents
;
Artificial Intelligence
;
Artificial Intelligence and Decision Support Systems
;
Bioinformatics
;
Biomedical Engineering
;
Biomedical Signal Processing
;
Cognitive Systems
;
Computational Intelligence
;
Data Manipulation
;
Distributed and Mobile Software Systems
;
Enterprise Information Systems
;
Evolutionary Computing
;
Health Engineering and Technology Applications
;
Human-Computer Interaction
;
Hybrid Intelligent Systems
;
Information Systems Analysis and Specification
;
Knowledge Discovery and Information Retrieval
;
Knowledge Engineering and Ontology Development
;
Knowledge Representation and Reasoning
;
Knowledge-Based Systems
;
Machine Learning
;
Methodologies and Methods
;
Methodologies and Technologies
;
Multi-Agent Systems
;
Neurocomputing
;
Neurotechnology, Electronics and Informatics
;
Operational Research
;
Pattern Recognition
;
Physiological Computing Systems
;
Reactive AI
;
Sensor Networks
;
Simulation
;
Soft Computing
;
Software Engineering
;
Symbolic Systems
Abstract:
The dual-process theory of human cognition proposes the existence of two systems for decision-making: a slower, deliberative, ``problem-solving'' system and a quicker, reactive, ``pattern-recognition'' system. The aim of this work is to explore the effect on agent performance of altering the balance of these systems in an environment of varying complexity. This is an important question, both in the realm of explanations of expert behaviour and to AI in general. To achieve this, we implement three distinct types of agent, embodying different balances of their problem-solving and pattern-recognition systems, using a novel, hybrid, human-like cognitive architecture. These agents are then situated in the virtual, stochastic, multi-agent ``Tileworld'' domain, whose intrinsic and extrinsic environmental complexity can be precisely controlled and widely varied. This domain provides an adequate test-bed to analyse the research question posed. A number of computational simulations are r
un. Our results indicate that there is a definite performance benefit for agents which use a mixture of problem-solving and pattern-recognition systems, especially in highly complex environments.
(More)