The state of artificial intelligence

Artificial intelligence has been an active branch of research for computer scientists and psychologists for 50 years. The concept of mimicking human intelligence in a computer fuels the public imagination and has led to countless academic papers, news articles and fictional works. However, public expectations remain largely unfulfilled, owing to the incredible complexity of everyday human behavior. A wide range of tools and techniques have emerged from the field of artificial intelligence, many of which are reviewed here. They include rules, frames, model-based reasoning, case-based reasoning, Bayesian updating, fuzzy logic, multiagent systems, swarm intelligence, genetic algorithms, neural networks, and hybrids such as blackboard systems. These are all ingenious, practical, and useful in various contexts. Some approaches are pre-specified and structured, while others specify only low-level behavior, leaving the intelligence to emerge through complex interactions. Some approaches are based on the use of knowledge expressed in words and symbols, whereas others use only mathematical and numerical constructions. It is proposed that there exists a spectrum of intelligent behaviors from low-level reactive systems through to high-level systems that encapsulate specialist expertise. Separate branches of research have made strides at both ends of the spectrum, but difficulties remain in devising a system that spans the full spectrum of intelligent behavior, including the difficult areas in the middle that include common sense and perception. Artificial intelligence is increasingly appearing in situated systems that interact with their physical environment. As these systems become more compact they are likely to become embedded into everyday equipment. As the 50th anniversary approaches of the Dartmouth conference where the term ‘artificial intelligence’ was first published, it is concluded that the field is in good shape and has delivered some great results. Yet human thought processes are incredibly complex, and mimicking them convincingly remains an elusive challenge. ADVANCES IN COMPUTERS, VOL. 65 1 Copyright © 2005 Elsevier Inc. ISSN: 0065-2458/DOI 10.1016/S0065-2458(05)65001-2 All rights reserved.