Pierre Marquis (editor) - A Guided Tour of Artificial Intelligence Research: Knowledge Representation and Reasoning: Volume I: Knowledge Representation, Reasoning and Learning

This Springer imprint is published by the registered company Springer Nature Switzerland AG

The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

The first volume of the Artificial Intelligence (AI) guided tour describes how onecan enable a computer system to reason. As this book is a guided tour, it considersthe many topics developed by AI researchers. Naturally, this is essential for allthose who want to make progress in our field; however, this does not imply thatreading these books should be restricted to AI scientists. In fact, they do notdescribe a succession of achievements, but the general principles that allow therealization of the most noteworthy results. This overall perspective gives usefulideas to all scientists, although they do not consider themselves as belonging in themainstream of Artificial Intelligence, even if they do not try to solve problems on acomputer.

AI is interested in its main goal: solving problems that were previously onlysolved by living beings, and particularly by human ones. However, it turns out that,while doing so, AI has often discovered methods and ideas useful for other scientificdisciplines. Thus, in several domains of Cognitive Science, new approachescame from ideas widely used in AI: For instance, apart from the statisticaldescription of the life of an anthill, some have begun to model every ant, consideredas a small automaton. In this book, cognitive scientists will certainly be interested inthe study of trust and emotions for a cognitive agent. For their part, philosophersand logicians will read the chapter on deontic logics, which specify in a rigorous,unadorned language, concepts such as obligatory, permissible, optional, and ought.They could also see how it is possible to reason even when there is contradictoryinformation. The discovery of a contradiction is not always a total disaster, since itcan come from a small amount of information, where one can clean up the mess.Belief revision can restore the consistency of knowledge when new data areinconsistent with what was already known.

Economists and sociologists will look with interest at the methods for collectivedecision, where several agents must cooperate to find a common decision; thesemethods are very often helpful, for choosing the president of a political party aswell as for the choice of a restaurant by a group of friends. For making suchdecisions, IA researchers have experimented with various kinds of votes and auctions;they defined equity in the sharing of resources. They have found methods and concepts useful when there are several artificial agents, but these results can also beused when a group of humans is taking a collective decision. However, I believethat computer specialists will be the most interested community: For them, nearlyall the chapters of this book will be very useful. Indeed, the distinction betweencomputer and AI scientists is often very tenuous: AI researchers become computerspecialists when they implement a system that must obtain excellent results.Conversely, without realizing it, computer specialists may use AI methods whenthey are developing a system for solving a particular problem: It is natural towonder which methods a human being is using for solving it, and to implementthem in the system.

Several areas of computer science have been developed for the first time by AI researchers: we consider them because human beings manage to obtain good results when they use them. For us, the natural approach is to ask: why cant a computer program do the same? For other people, it seems impossible or too difficult for the present state of the art. Being the first to consider a problem, we cannot be prevented to find new methods for solving it! It happens that these methods may be useful in new applications, and the computer scientists should put them in their tool box, so that they will think about using them when an opportunity presents itself.

In particular, the beginning of this volume title mentions an important problemfor computer specialists: Knowledge Representation. Several aspects of this issueare considered, such as the representation of preferences and of uncertainty.Ontologies are very important for semantic Web applications: They provide aformal knowledge representation for their domain, and they allow their management,acquisition, retrieval, etc. Knowledge engineering gives methods for findingknowledge for a particular problem, especially by a collaboration with humanexperts. These chapters can help a computer scientist to find ideas for developingfuture systems.

The second part of the title is Formalization of Reasoning. This capacity isimportant: If a system can reason, it is more general. Indeed, it is no longer necessaryto anticipate all the possible situations: By reasoning, a system can automaticallyfind the right action in an unexpected situation. General systems offer adual advantage: Firstly, fewer programs must be written; a general system does thejob of several specific systems. Secondly, results are obtained far more quickly: Ageneral system may be adapted to a new application without the need for writingmore programs. Many kinds of reasoning are presented; I will choose some of themand show that they are important and go beyond AI.