ion and delegation: Agents can be made extensible and composable in ways that common iconic interface objects cannot. Because we can "communicate" with them, they can share our goals, rather than simply process our commands. They can show us how to do things and tell us what went wrong (Miller and Neches 1987). . Flexibility and opportunism: Because they can be instructed at the level of 16 BRADSHAW goals and strategies, agents can find ways to "work around" unforeseen problems and exploit new opportunities as they help solve problems. . Task orientation: Agents can be designed to take the context of the person's tasks and situation into account as they present information and take action. . Adaptivity: Agents can use learning algorithms to continually improve their behavior by noticing recurrent patterns of actions and events. Toward Agent-Enabled System Architectures In the future, assistant agents at the user interface and resource-managing agents behind the scenes will increas...

We consider the reuse of policies for previous MDPs in learning on a new MDP, under the assumption that the vector of parameters of each MDP is drawn from a fixed probability distribution. We use the options framework, in which an option consists of a set of initiation states, a policy, and a termination condition. We use an option called a reuse option, for which the set of initiation states is the set of all states, the policy is a combination of policies from the old MDPs, and the termination condition is based on the number of time steps since the option was initiated. Given policies for m of the MDPs from the distribution, we construct reuse options from the policies and compare performance on an m + 1st MDP both with and without various reuse options. We find that reuse options can speed initial learning of the m+ 1st task. We also present a distribution of MDPs for which reuse options can slow initial learning. We discuss reasons for this and suggest other ways to design reuse options.

This paper starts out from two observations: firstly, that there are complex links between what we term intelligence and what we term creativity and, secondly, that the phenomenon of surprise has a significant role in both the genesis and evaluation of creativity, and is tightly coupled to perception.

In this paper we develop a computational framework for the measurement of different factors or abilities which are usually found in intelligent behaviours. For this, we first develop a scale for measuring the complexity of an instance of a problem, depending on the descriptional complexity (Levin LT variant) of the ‘explanation ’ of the answer to the problem. We centre on the establishment of either deductive and inductive abilities, and we show that their evaluation settings are special cases of the general framework. Some classical dependencies between them are shown and a way to separate these dependencies is developed. Finally, some variants of the previous factors and other possible ones to be taken into account are investigated. In the end, the application of these measurements for the evaluation of AI progress is discussed.

The MAS group at the DFKI is developing architectures for animated agents in virtual worlds. Within the tradition of AI there are many techniques that can be used to model the cognitive abilities of such characters, in particular the ability to learn from the environment and plan actions that lead to a well defined goal. Some have argued, however, that these techniques are too complex [Reilly, 1996], that simpler techniques with a flexible control structure are the key to making characters believable. We argue here that much of the advantages of intelligent agent architectures can be preserved, if we complement them with resourcebounded, motivational constructs. 1 INTRODUCTION The continuous increase of computing power, its concomitant ubiquity and the connectivity of the Internet are changing the interfaces computer systems offer to the user. The interfaces are now often three-dimensional, sometimes immersive, open to many users and populated with animated software entities that mode...

This is a practical and theoretical thesis in visual robotics. It describes the development and actual implementation (in a real world, real-time visual robot) of new approaches to three non-linear problems: how to ensure robustness under the harshest conditions of unforeseen reconfiguration, how to provide specialised space-variant sampling regimes according to which task is currently at hand, and how to automatically direct attention using any number of adaptive response layers in concert. Additionally, the descriptions of this practical work are preceded by the exposition of a new theoretical framework for intelligent system evaluation, which offers performance silhouettes as a schematic method. The three practical methods stem from, and are embedded in, the theoretical framework, and all have been suggested, to varying degrees, by knowledge of biological processes or capabilities- self-organisation, visual periphery sensitivity, and adaptive reduction in sensitivity, for example. The overall goal of the research is to develop extremely simple algorithms capable of operating in real-time and endowing a robot with robust essential perceptual capabilities that can operate in all environments. The outcome of the work is a visual robotic system that exhibits seemingly intelligent behaviour in complex, changing, and noisy natural environments. That it does so with minimal help from other sophisticated agents (such as computer science researchers) is a credit to its autonomy and the adaptation of its design to arbitrary environments.

www.rthomaso.eecs.umich.edu

Abstract not found

draft of 031607.0108NY This chapter is an esemplastic systematization of declarative computational cognitive modeling, a field that cuts across cognitive modeling based on cognitive architectures (such as ACT-R, Soar, and Clarion), human-level artificial intelligence (AI), logic itself, and psychology of reasoning (especially of the computational kind). The hallmarks of declarative computational cognitive modeling are the following two intertwined constraints: (1) The central units of information used in the approach are (at least in significant part) declarative in nature, and the central process carried out over these units is inference. (2) The approach to modeling the mind is top-down, rather than bottom-up. (These two points are interconnected because once one commits to (1), (2) becomes quite unavoidable, since bottom-up processing in the brain, as reflected in relevant formalisms (e.g., artificial neural networks), is based on units of information that are numerical, not declarative.) The systematization of declarative computational cognitive modeling is achieved by using formal logic, and hence declarative computational cognitive modeling, from the formal perspective, becomes logic-based computational cognitive modeling (LCCM). The chapter covers some prior research that falls under LCCM; this research has been carried out by

. The paper reports experience constructing a C++[4] class library modeling the mathematical structures of order theory. The provision of such a library has two main teaching advantages. Firstly, it allows students in 2nd and 3rd year Computer Science/Engineering to gain exposure and practice in state-of-the-art software engineering methodology using C++ --- in high demand in both industry and academe. Secondly, it allows the undergraduate artificial intelligence (AI) practical work to be extended to more challenging and realistic exercises in agent-oriented AI [19, 16, 1]. In satisfying these two teaching advantages, this approach marries the formal properties of theoretical AI with modern software engineering practice. To our knowledge, there are no other undergraduate courses in AI in Australia combining AI and C++ programming in this way --- although we would be interested to hear if there were. These ideas are fairly general, and in this sense, destined to appeal to other undergra...