If you are looking for an answer to the question What is Artificial Intelligence? and you only have a minute, then here's the definition the Association for the Advancement of Artificial Intelligence offers on its home page: "the scientific understanding of the mechanisms underlying thought and intelligent behavior and their embodiment in machines."
However, if you are fortunate enough to have more than a minute, then please get ready to embark upon an exciting journey exploring AI (but beware, it could last a lifetime) …
This paper presents a novel approach to traffic management by coordinating driver behaviors. Current traffic management systems do not consider lane organization of the cars and only affect traffic flows by controlling traffic signals or ramp meters. However, drivers should be able to increase traffic throughput and more consistently maintain desired speeds by selecting lanes intelhgently. We pose the problem of intelligent lane selection as a challenging and potentially rewarding problem for artificial intelligence, and we propose a methodology that uses supervised and reinforcement learning to form distributed control strategies. Initial results are promising and demonstrate that intelligent lane selection can better approximate desired speeds and reduce the total number of lane changes. Introduction A large effort is under way by government and industry in America, Europe, and Japan to develop intelligent vehicle and highway systems. These systems incorporate ideas from artificial intelligence, intelligent control, and decision theory, among others, to automate many aspects of driving and traffic control. The goals of this effort are quite broad and include increased traffic throughput, fewer accidents, reduced fuel consumption, and a better driving experience. Advanced traffic management systems are designed to reduce congestion and increase overall traffic throughput. Almost all such systems maintain efficient traffic flows by controlling traffic signals or highway ramp meters, treating traffic as a single mass and normally ignoring the behavior of individual cars (Gilmore, Elibiary, Forbes, 1994; Kagolanu, Fink, Smartt, Powell, & Larson, 1995; Pooran, Tarnoff, & Kalaputapu, 1996). This view, however, misses an important component of traffic management: coordination of the cars themselves.
A Messy Genetic Algorithm is customized toflnd'optimal many-to-many matches for 2D line segment models. The Messy GA is a variant upon the Standard Genetic Algorithm in which chromosome length can vary. Consequently, population dynamics can be made to drive a relatively efficient and robust search for larger and better matches.
Some categories have a very simple structure, while others can be complex. Accordingly, learning how to properly classify items as members of category "A" or "B" can be almost trivial (e.g., the value of a single input dimension determines membership) or can be so difficult that no regularity is discovered (e.g., rote memorization of every category member is required to determine membership). Classifications are harder to master when the decision boundary (in a multidimensional space of possible inputs) is highly irregular and when there are multiple boundaries (e.g., all the members of category "A" do not fall inside one contiguous region of the input space). Difficult classification problems (problems with complex decision boundaries) typically involve categories that have a complex internal structure, perhaps consisting of multiple prototypes (i.e., category subtypes) and a number of exceptions. Linguistic analyses have demonstrated that many categories have a rich internal structure (Lakoff, 1987). Very simple learning models will fail to master difficult categorizations with complex boundaries (i.e., categories with rich internal structure). For instance, a purely linear model, like the perceptron (Rosenblatt, 1958), will be unable to master a classification when the mapping from input features to category labels is nonlinear.
These are domains in which learning must proceed exclusively with failure examples that are relatively uninformative for conventional methods. A domain theory is used to explain and then systematically perturb the observed failures so that they can be treated as if they were positive training examples. The concept induced from these "phantom" examples is exercised in the world, yielding additional observations, and the process repeats. Surprisingly, an accurate concept can often be learned even if the phantom examples are themselves failures and the domain theory is only imprecise and approximate. We investigate the behavior of the method in a stylized air-hockey domain which demands a nonlinear decision concept. Learning is shown empirically to be robust in the face of degraded domain knowledge. An interpretation is advanced which indicates that the information available from a plausible qualitative domain theory is sufficient for robust successful learning.
A duration is known as a time distance between two point events. This relationship has recently been formalized as the point duration network (PDN) in (Navarrete & Marin 1997). However, only the qualitative information about points and durations was considered. This paper presents an augmented point duration network (APDN) represent both qualitative and quantitative information about point events. We further extend APDN to capture quantitative information about durations. We propose algorithms to solve reasoning tasks such as determining satisfiability of the network, and finding a consistent scenario with minimal domains. Thus, we present an expressively richer framework than the existing ones to handle both qualitative and quantitative information about points as well as durations. Introduction Temporal knowledge can be classified into two main categories: qualitative and quantitative (or metric) information. Relationships between events (e.g., Fred arrived at work before John) are considered as a class of qualitative information while numeric distance or an event instance (e.g., Fred took 15-20 minutes to get to work) is considered as quantitative information.
Box 1738 3000 DR Rotterdam, the Netherlands YTANQFAC.FBK.EUR..NL Abstract Deontic logic, the logic of obligations and permissions, is plagued by several paradoxes that have to be understood before deontic logic can be used as a knowledge representation language. In this paper we extend the temporal analysis of Chishohn's paradox using a deontic logic that combines temporal and preferential notions. Introduction Deontic logic is a modal logic in which Op is read as'p ought to be (done).' Deontic logic has traditionally been used by philosophers to analyze the structure of the normative use of language. In the eighties deontic logic had a revival, when it was discovered by computer scientists that this logic can be used for the formal specification and validation of a wide variety of topics in computer science (for an overview and further references see (Wieringa & Meyer 1993)). The advantage is that norms can be violated without creating an inconsistency in the formal specification, in contrast to violations of hard constraints. Another application is the use of deontic logic to represent legal reasoning in legal expert systems in artificial intelligence. Legal expert systems have to be able to reason about legal rules and documents such as for example a trade contract.
The algebra (1) is a refinement of the CYCORD theory; (2) contains 24 atomic relations, hence 224 general relations, of which the usual CYCORD relation is a particular relation; and (3) is NPcomplete, which is not surprising since the CYCORD theory is. We then provide: (1) a constraint propagation algorithm for the algebra, which we show is polynomial, and complete for a subclass including all atomic relations; (2) a proof that another subclass, expressing only information on parallel orientations, is NPcomplete; and (3) a solution search algorithm for general problem expressed in the algebra. Introduction Qualitative spatial reasoning (QSI%) has become important and challenging research area of Artificial Intelligence. An important aspect of it is topological reasoning (e.g.
The correct construction of reliable situated agents is an important task in agent research nowadays. Consider for example the following plan for a robotic agent to cross a busy motor way: Wait until the road is clear then cross at. Is it a correct plan to cross a busy road 7 i The correctness of this plan depends very much on the robot's capabilities. If the robot is fast enough to be able to finish crossing the road before a car could pass by then the above plan is correct.
In abductive planning, plans are constructed as reasons for an agent to act: plans are demonstrations in logical theory of action that a goal will result assuming that given actions occur successfully. This paper shows how to construct plans abductively for an agent that can sense the world to augment its partial information. We use a formalism that explicitly refers not only to time but also to the information on which the agent deliberates. Goals are reformulated to represent the successive stages of deliberation and action the agent follows in carrying out a course of action, while constraints on assumed actions ensure that an agent at each step performs a specific action selected for its known effects. The result is a simple formalism that can directly inform extensions to implemented planners.
Representing properties of actions has been the subject of many papers and two recent books (Sandewal11995), (Shanahan 1997). One direction of work makes use of "action languages," such as,4 (Gelfond& Lifschitz 1993) and its dialects. An action language serves for describing the effects of actions on fluents. The meaning of a set of propositions in an action language can be represented by a "transition diagram." In this paper we define a new action language ¢, based on the theory of causal explanation proposed in (McCain & Turner 1997) and extended in (Lifschitz 1997a).