Creativity and Artificial Intelligence

The paper describes a transition logic, TL , and a deductive formalism for i t . It shows how various important aspects (such as ramification, qualification, specificity, simultaneity, indeterminism etc.) involved in planning can be modelled in TL in a rather natural way. (The deductive formalism for) TL extends the l inear connection method proposed earlier by the author by embedding the latter into classical logic, so that classical and resource-sensitiv reasoning coexist within T L . The attraction of a logical and deductive approach to planning is emphasised and the state of automated deduct ion briefly described. 1 I n t r oduc t i on Arti f ic ial Intelligence (A I , or intellectics [Bibel, 1992a]) aims at creating artificial intelligence. Were there no natural intelligence, the sentence would be meaningless to us. Hence understanding natural intelligence by necessity has always been among the goals of intellectics (and is also the goal of cognitive science). Different points of view for approaching the goal of creating artif icial intelligence have been distinguished [Kushmerick, 1996]. Logicism [Nilsson, 199l], cognitivism [Laird et a/., 1987], and situated act ion [Agre, 1995] are three out of several such points of view. In a nutshell, the logicistic point of view argues that man can describe his creations (including an artif icial intelligence) only by natural linguistic, hence logical means; thus any way towards artificial intelligence must in some sense be a logical one. This author is strongly committed to the logicistic approach. As a consequence he believes that any other approach is in fact a logicistic one in disguise. Intelligence has many features. Clearly one of them is the abil i ty to plan ahead in time. Intuitively, planning is logical reasoning of some kind. Al l the more one might expect that planning is the domain where logic and its deductive machinery excel. The fact is that it does not. There are many software systems in everyday use solving planning tasks, but to the author's best knowledge none of them is based on logic and has a deductive component. Does this imply that logic is irrelevant for planning and for artificial intelligence for that matter? While intelligence implies the ability for planning, the converse has not necessarily to be true. It very much depends on what kind of planning is meant. In a fixed and relatively restricted domain (such as text layout) planning may well be realized in a purely functional way and with standard programming techniques. But functional (or procedural) programming has its l imits as we enter more complex and unpredictable domains; in particular it wil l never be able to produce a behavior which rightly deserves to be named "intelligent" (surely as a user of computers you noticed the stupidity of text layout systems). Section 7, as well as numerous texts in the literature, give arguments for this statement. It also gives reasons which explain the resistance of the software industry to a bolder move into a logic technology for planning and for other applications. In other words, logic is essential for intelligent planning in the true sense of the term, but industry is not ready to build intelligent systems. It is not the task of intellecticians to lament about this state of affairs but rather to prepare for the coming day when the market wil l be ripe for a broader use of a truly intelligent technology and to develop the best possible technological basis for i t . In fact, if we are frank there is yet a lot to be developed before we can comfortably go out to industry and offer a coherent set of methods for dealing with the many facets of intelligence including planning. In the present paper I review the state of the art in deductive planning with an emphasis on the contributions from research groups influenced by my own work. While much of the work in deductive planning has focused on representational issues we have always approached the problem with the necessary and available deductive techniques in mind. Since the methods and systems growing out of our work have finally achieved a leading position in the deduction community by winning the CADE-96 competition in automated theorem proving with the SETHEO system [Lets et a/., 1992; Moser et a/., 1997], we are perhaps also well placed to import the best possible techniques into the planning