We provide a brief overview of recent direct preference elicitation methods: these methods ask users to answer (ideally, a small number of) queries regarding their preferences and use this information to recommend a feasible decision that would be (approximately) optimal given those preferences. We argue for the importance of assessing numerical utilities rather than qualitative preferences and survey several utility elicitation techniques from artificial intelligence, operations research, and conjoint analysis. Specifically, since the ability to make reasonable decisions on behalf of a user depends on that user's preferences over outcomes in the domain in question, AI systems must assess or estimate these preferences before making decisions. Designing effective preference assessment techniques to incorporate such user-specific considerations (that is, breaking the preference bottleneck) is one of the most important problems facing AI. In this brief survey, we focus on explicit elicitation techniques where a system actively queries a user to glean relevant preferences. Preference elicitation is difficult for two main reasons. First, many decision problems have exponentially sized outcome spaces, defined by the possible values of outcome attributes. As an illustrative example, consider sophisticated flight selection: possible outcomes are defined by attributes such as trip cost, departure time, return time, airline, number of connections, flight length, baggage weight limit, flight class, (the possibility of) lost luggage, flight delays, and other stochastic outcomes. An ideal decision support system should be able to use, for example, precise flight delay statistics and incorporate a user's relative tolerance for delays in making recommendations. Representing and eliciting preferences for all outcomes in a case like this is infeasible given the size of the outcome space. A second difficulty arises due to the fact that quantitative strength of preferences, or utility, is needed to trade off, for instance, the odds of flight delays with other attributes. Unfortunately, people are notoriously inept at quantifying their preferences with any degree of precision, adding to the challenges facing automated utility elicitation.

"An Innovative Application from the DARPA Knowledge Bases Programs: Rapid Development of a Course-of-Action Critiquer," by Gheorghe Tecuci, Mihai Boicu, Mike Bowman, and Dorin Marcu, describes a critiquing agent for military courses of action, a challenge problem set by the Defense Advanced Research Projects Agency's (DARPA) High-Performance Knowledge Bases Program. Murray Burke, the DARPA manager for this program, introduces the article by setting the context for the application. Ontologies also play a key role in the creation and management of a web portal developed by Steffen Staab and his colleagues at the University of Karlsruhe, discussed in their article, "Knowledge Portals: Ontologies at Work." "L As in past years, papers were solicited in two categories: (1) deployed applications and (2) emerging applications and technologies. Deployed applications are systems that have been in use for at least several months by individuals or organizations other than their developers, have measurable benefits, and incorporate AI technologies. Emerging applications are systems that are close to deployment and clearly show an innovative implementation of AI technologies. Papers submitted in this track can also describe efforts that examine the utility of different AI techniques for specific applications. All these case studies are of value not only to other application developers looking for guidance in applying various techniques to their own applications but also to researchers who need to understand the technical challenges provided by real-world problems. Six deployed applications and 12 emerging application papers were presented plus 2 invited talks. Although no single theme emerges from this panoply of excellent applications, they served to demonstrate that the field continues to be fertile ground for innovation.

Ramasamy Uthurusamy was the program chair, and Barbara Hayes-Roth was the program cochair. IAAI-99 was a special occasion that provided an opportunity to reflect on a decade of IAAI conferences and contemplate the potential contributions in the coming decade. In addition to the three invited talks that specifically addressed these areas, IAAI-99 showcased some exciting and innovative applications. Although all the IAAI-99 papers and talks were certainly interesting and important, we present in this special issue of AI Magazine only a select subset because of page and other limitations. We include two invited talks and four applications as a snapshot of IAAI-99.

DITORIAL AI Magazine Volume 11 Number 2 (1990) ( AAAI) In this issue, Luc Steels takes a new and insightful look at knowledgebased systems and provides a synthesis of several different approaches to analyzing expertise. It's a long article but, in my opinion, an important one. I recommend it to anyone with an interest in knowledge-level analysis of expert systems. On the same general topic of expert systems but from a different perspective is the article by Rob Weitz, who proposes a methodology for forecasting the impact of expert systems on the workplace over the near term. Finally, James Hendler, Austin Tate, and Mark Drummond present an extensive survey of AI systems and techniques for plan generation.

A variety of proposals for preferred methodological approaches has been advanced in the recent artificial intelligence (AI) literature Rather than advocating a particular approach, this article attempts to explain the apparent confusion of efforts in the field in terms of differences among underlying methodological perspectives held by practicing researchers The article presents a review of such perspectives discussed in the existing literature and then considers a descriptive and relatively specific typology of these differing research perspectives. Studies are reported in a wide range of publications. While some focus on the field (e.g., Artzficial Intelligence), others are concerned with different research areas (e.g., Behavzoral and Brain Sczences). Perhaps, as others have pointed out, "there are undoubtedly some views AI simply adds to the prevailing sense of confusion. AI research, which have been previously reported in .

This article provides an overview of the state of the art in this area. It begins with the dominant paradigm of assistive control, from impedance-based cooperative controller through electromyography and intention estimation. It then covers challenge-based algorithms, which provide more difficult and complex tasks for the patient to perform through resistive control and error augmentation. Furthermore, it describes exercise adaptation algorithms that change the overall exercise intensity based on the patient's performance or physiological responses, as well as socially assistive robots that provide only verbal and visual guidance. The article concludes with a discussion of the current challenges in rehabilitation robot software: evaluating existing control strategies in a clinical setting as well as increasing the robot's autonomy using entirely new artificial intelligence techniques.

At the recent AAAI conference at Stanford, it became apparent that many new AI research centers are being established around the country in industrial and governmental settings and in universities that have not paid much attention to Al in the past. What does an AI researcher want from his computing facility? What will make him most productive? In fact, the needs of the Al researcher are not very different from the needs of any other researcher in computer science, except that facility-related problems seem to become acute in AI a few years sooner than they are felt in other research areas The considerations are roughly as follows: 0 AI programs tend to be very large because they contain, in one form or another, a lot of knowledge. It follows, then, that any machine used for AI must provide a large virtual address space in order to insulate the researcher from having to think about how to chop up his task into smaller tidbits and overlays.

This editorial introduction provides an overview of artificial intelligence for computational sustainability, and introduces the special issue articles that appear in this issue and the previous issue of AI Magazine. The emerging interdisciplinary field of computational sustainability (Gomes 2009) draws techniques from computer science, information science, mathematics, statistics, operations research, and related disciplines to help balance environmental and socioeconomic needs for sustainable development. Artificial intelligence (AI) techniques play a key role in computational sustainability research, enabling the solution of sustainability problems that involve modeling or decision making in dynamic and uncertain environments. Since 2011, the main AAAI conference has included a special track on computational sustainability, encouraging AI research in this area and broader participation of sustainability researchers in the AAAI community. Sustainable solutions must balance between environmental, societal, and economic demands (United Nations General Assembly 2005).

The integration of AI and vision has been a longterm goal of both disciplines for more than three decades. This special issue illustrates some recent work on bridging the gap. Back then, the issue was anticipated relatively easily, and a full integration of the two fields was expected within a decade. Processing of images is known to result in noisy segmentation, and in general, data might not be consistent in space or time, making AI methods appealing. In early AI, the handling of noisy, partially inconsistent data was, at best, a major challenge.

ATP systems are used in a wide variety of domains: A mathematician might use the axioms of group theory to prove the conjecture that groups of order two are commutative; a management consultant might formulate axioms that describe how organizations grow and interact and, from these axioms, prove that organizational death rates decrease with age; or a frustrated teenager might formulate the jumbled faces of a Rubik's cube as a conjecture and prove, from axioms that describe legal changes to the cube's configuration, that the cube can be rearranged to the solution state. All these tasks can be performed by an ATP system, given an appropriate formulation of the problem as axioms, hypotheses, and a conjecture. Most commonly, ATP systems are embedded as components of larger, more complex software systems, and in this context, the ATP systems are required to autonomously solve subproblems that are generated by the overall system. To build a useful ATP system, several issues have to ...