Books‎ > ‎

The Quest for Artificial Intelligence

posted Oct 1, 2011, 3:23 AM by Fernando Berzal   [ updated Oct 1, 2011, 8:34 AM ]

The history of ideas deals with the proposal, rebuttal, and evolution of ideas over time. This is always a fascinating topic, as it makes us aware of many of our underlying assumptions, suggests alternatives we might not have considered, and makes us reflect upon the trends and controversies that shape the history of a research field. The field of artificial intelligence (AI) is particularly relevant, due to the multitude of distinct ideas that have been devised, borrowed from other fields, adapted, combined, and often discarded, but also adopted to build many practical systems now in use. Now that the field has been with us for the last 50 years or so, we are offered the opportunity to read a roughly chronological account of its evolution, from the perspective of one of its pioneers.

Nilsson considers intelligence to be “that quality that enables an entity to function appropriately and with foresight in its environment.” Throughout his historical account, he surveys many of the major milestones that have led AI to its current state, from the often-unsuccessful trials of people who were ahead of their time and its modest origins in related disciplines. Logic, statistics, philosophy, psychological theories, what we now call neuroscience, and engineering all contributed to the early experiments that led to the first AI systems built at sites such as Stanford University, the Massachusetts Institute of Technology (MIT), and Carnegie Mellon University (CMU), in the 1950s and 1960s.

Those early efforts led to robotic systems such as Shakey. Nilsson’s first-hand recollection of it is certainly enjoyable. At its time, Shakey was “the first robot system having the abilities to plan, reason, and learn; to perceive its environment using vision...[;] and to monitor the execution of its plans.” Many ideas now familiar to computer scientists, such as the A* heuristic search procedure or the STRIPS planner, were originally proposed by Nilsson and his collaborators during the late 1960s and the early 1970s. Many knowledge representation techniques, such as scripts and frames, are from the same period (semantic networks were proposed earlier).

During the following decade, research led to the development of many perceptual systems that allowed machines to interact more like humans. Some noteworthy computer vision algorithms and many natural language processing techniques fall within this category. Speech recognition systems, for instance, also introduced an assorted toolbox that would later be used in many other contexts, from hidden Markov models to blackboard architectures. Work on knowledge representation, on the other hand, led to the flourishing of expert systems during the 1980s, whose techniques are at the heart of modern business rules engines.

Significant success toward the goal of making a machine behave in ways that would be called intelligent if a human behaved so also led to overly optimistic predictions. This caused an expectations management problem that finally led to the “AI winter,” as Nilsson puts it. Apart from mentioning the funding and political issues that influenced the development of AI research projects, Nilsson complements his chronological account with a proper discussion of the “controversies that were simmering on the sidelines and within the field itself.”

Good-old-fashioned AI ultimately led to the development of new paradigms, whose growth led to the now-fragmented AI subfield, from machine learning and data mining, to knowledge engineering techniques, computer vision systems, and the statistical natural language processing techniques currently in vogue. We have already witnessed some extraordinary achievements, from world-champion-level game-playing machines, to autonomous robots and driverless automobiles. Nilsson comments on challenging problems that AI techniques have recently helped solve, and then goes on to predict what AI programs might still do. Even though he modestly acknowledges that “more accomplished historians ... have wisely avoided writing accounts that get too close to the present,” Nilsson is bold enough to predict and even advocate for the appearance of human-level AI during the current century.

Apart from the fact that one might find the final human-level AI issue too debatable to be included in a historical retrospective such as this, Nilsson provides an otherwise balanced look at what AI has been able to do during its first 50 years of existence. His personal recollections and the rationale behind many decisions, as retold by an insider, make this book a unique contribution, interesting both for the informed and for the general reader. Both kinds of readers can learn a lot from Nilsson’s book about the evolution of this now-mature research field. The book is written in a friendly conversational style, without any unnecessary mathematical formalisms, and is richly illustrated with many diagrams that depict representative AI systems and photographs of the many innovators that led to their development.

Reviewer:  Fernando BerzalReview #: CR137774 (1102-0149)




También disponible en
 
Comments