AAAI-07 / IAAI-07 Invited Speakers
Tuesday, July 24
Keynote Presidential Address: Agents, Bodies, Constraints and Dynamics
Alan Mackworth (University of British Columbia)
Artificial intelligence focuses on agents carrying out tasks in an environment. Consider the evolution of the agent-task-environment triple. In Good Old Fashioned AI and Robotics (GOFAIR), a single agent typically solved a puzzle in a fully- observable, fully-predictable world. Constraint-based agents were initially proposed in that paradigm. Robot soccer, developed in our lab in 1991, caught on as an international challenge task. The community had to deal with multiagent, dynamic, online, situated, embodied activity in a partially-observable, partially-predictable world. We evolved new constraint-based hybrid agent controllers for such tasks. The main challenge is to design agents that are both proactive and reactive. As our systems migrate into the real world, critical issues of reliability, trust and ethics must be addressed. For example, making ethical choices about our mutual interactions with robots presupposes that we are able to foresee the possible future effects of that interaction (or inaction). This presupposition puts strong requirements on the design space for robot architectures. We cannot use ad hoc or opaque models of robot structure or function. We need languages for modeling the constraints on an agent’s dynamics (where we may take “dynamics” quite abstractly) and languages for writing constraint-based behavioral specifications. Moreover, we need techniques for determining if an agent will, or is likely to, satisfy its specifications. Some of the exciting new applications of our science, such as assistive technology, require that we pay particular attention to these concerns of reliability and trust.
AAAI President Alan Mackworth (University of British Columbia) will deliver his AAAI presidential address on Tuesday, July 24. Mackworth is a professor of computer science and Canada Research Chair in artificial intelligence at UBC. He works on constraint-based computational intelligence with applications in vision, robotics, and situated agents, and is a pioneer in the areas of constraint satisfaction, robot soccer, and constraint-based agents. Mackworth has published over 100 scientific papers, and coauthored Computational Intelligence: A Logical Approach. In addition, he has served as president of IJCAI and also CSCSI. He was the founding director of the UBC Laboratory for Computational Intelligence.
Revolutionizing Prostheses: A Program of the Defense Advanced Research Projects Agency (DARPA)
Geoffrey S. F. Ling (DARPA)
The Defense Advanced Research Projects Agency (DARPA) sponsors revolutionary research targeted at bridging gaps between the needs of the warfighter and current technological, medical, and scientific capabilities. In the realm of neurology, specifically, the Human Assisted Neural Devices (HAND) and Revolutionizing Prosthesis programs are efforts aimed at advancing the interfaces between assistive devices and users by leveraging off of biological capabilities. Both programs have provided the basis for paradigm shifts in treatment of extremity injury and traumatic brain injury.
Geoffrey Ling is a program manager with the Defense Advance Research Projects Agency (DARPA). There, he is responsible for a portfolio of programs that include Revolutionizing Prosthesis, Human Assisted Neural Devices, Preventing Violent Explosive Neuro Trauma, Predicting Health and Disease and Long Term Blood Storage. He is also developing new programs with the intent of protecting and restoring injured warfighters.
Ling is also professor and vice chairman of neurology at the Uniformed Services University of the Health Sciences in Bethesda, Maryland. Currently on active duty as an army colonel, he is the military's only neurocritical care physician. At Walter Reed Army Medical Center, he is director of neurointensive care. He is also an associate professor (adjunct) at Johns Hopkins Hospital where he attends on the Neuro Critical Care Unit. He has served two combat deployments as a neuro intensive care physician that included both Operation Enduring Freedom (Afghanistan) and Operation Iraqi Freedom (Iraq). Notably, he served with the 86th Combat Support Hospital ("Baghdad ER").
Geoffrey Ling received his B.A. with honors from Washington University in St. Louis, his M.D. from Georgetown University and his Ph.D. (pharmacology) from Cornell University.
Wednesday, July 25
Lise Getoor (University of Maryland, College Park)
Within the machine learning community, there has been a growing interest in learning structured models from input data that is itself structured. Graph identification refers to methods that transform an observed input graph into an inferred output graph. Examples include inferring organizational hierarchies from social network data and identifying gene regulatory networks from protein-protein interactions. The key processes in graph identification are entity resolution, link prediction, and collective classification. I will overview algorithms for these processes, discuss the need for integrating the results to solve the overall problem collectively, and show how these methods are relevant to foundational problems in AI such as knowledge representation, reformulation, and reasoning.
Lise Getoor is an assistant professor in the Computer Science Department at the University of Maryland, College Park. She received her PhD from Stanford University in 2001. Her current work includes research on link mining, statistical relational learning and representing uncertainty in structured and semistructured data. She has published numerous articles in machine learning, data mining, database, and AI forums. She is a member of AAAI Executive Council, is on the editorial board of the Machine Learning Journal, is a JAIR associate editor, and has served on a variety of program committees including AAAI, ICML, IJCAI, KDD, SIGMOD, UAI, VLDB, and WWW.
Logic for Automated Mechanism Design — A Progress Report
Michael Wooldridge (University of Liverpool, UK)
Cooperation logics, such as coalition logic and ATL, have proved to be a powerful formal tool for reasoning about gamelike mechanisms such as social choice procedures. In this talk, I present a survey of our joint work in this area. I begin by introducing ATL-like logics, and demonstrating that they form a natural tool for the specification of social choice procedures. I show how ATL model checkers can be used to verify economic properties of social choice mechanisms. I then discuss the main research issues in the area, discussing, for example, the succinct representation of social choice rules, the complexity of reasoning with such representations, and the handling of preferences.
This talk will report joint work with Thomas Agotnes (Bergen), Wiebe van der Hoek (Liverpool), Marc Pauly (Stanford), and Paul E. Dunne (Liverpool).
Michael Wooldridge is a professor in the Department of Computer Science at the University of Liverpool, UK. He has been active in multiagent systems research since 1989, and has published over two hundred articles in the area. His main interests are in the use of formal methods for reasoning about autonomous agents and multi-agent systems. Wooldridge was the recipient of the ACM Autonomous Agents Research Award in 2006. He is coeditor-in-chief of the Autonomous Agents and Multi-Agent Systems, journal and his introductory textbook An Introduction to Multiagent Systems" was published by Wiley in 2002.
AI in a Moore's Law World: the Stories of Farecast and KnowItAll
Robert S. Engelmore Memorial Award Lecture
Oren Etzioni (University of Washington)
What is the role for AI as CPUs, disks, and networks become exponentially more powerful? Perhaps, as some have suggested, the application of simple processing techniques to unprecedented mountains of data will suffice. Alternatively, perhaps Luis von Ahn's paradigm of "Human Computation" will yield any intelligence that is required. I will argue, to the contrary, that sophisticated AI techniques such as data mining and information extraction are necessary for exciting new applications. But these techniques have to be improved in important ways. I will make these points through the stories of Farecast.com and the KnowItAll research project.
Oren Etzioni is a professor of computer science at the University of Washington's Computer Science Department, and the founder and director of the university's Turing Center. He received his Ph.D. from Carnegie Mellon in 1990, and his B.A. from Harvard in 1986. Etzioni has authored over 100 technical papers on topics ranging from intelligent agents to data mining and web search. In 2005, he was awarded the IJCAI Distinguished Paper Award for "A Probabilistic Model of Redundancy in Information Extraction." He received a National Young Investigator Award in 1993, and was chosen as a AAAI Fellow a decade later.
Etzioni is the founder of three companies. Most recently, he is the founder of Farecast, a company that utilizes data mining to inform consumers about the right time to buy their air tickets. Farecast has won numerous awards in 2006 including TIME's "50 Coolest Web Sittes" and PC World's "20 Most Innovative Products." His work has been featured in the New York Times, Wall Street Journal, NPR,NBC Nightly News, SCIENCE, The Economist, TIME Magazine, Business Week, Newsweek, Discover Magazine, Forbes Magazine, Wired, and elsewhere.
Thursday, July 26
Moving Toward Peer-to-Peer Human-Robot Interaction
Alan C. Schultz (Naval Research Laboratory)
As we move from teleoperation toward collaboration, human-robot interactions become more complex and require that the human and the robot share more common knowledge about the world. At the collaborative level of interaction, the robot and human must exercise mixed initiative in problem solving, each taking advantage of their unique skills, location, and perspective of the current situation. Working toward solutions to human-robot collaboration requires a highly multi-disciplinary approach, as well as integration of very different areas of research, as can be seen in the emerging field of human-robot interaction. This talk will describe recent multidisciplinary approaches in shared perspective.
Alan C. Schultz is the director of the Navy Center for Applied Research in Artificial Intelligence at the Naval Research Laboratory in Washington, DC. His research is in the areas of human-robot interaction, cognitive robotics, evolutionary robotics, learning in robotic systems, and adaptive systems. He has published over 90 articles on machine learning and robotics. Schultz is currently the cochair of the AAAI Symposium Series, served as general cochair of the ACM/IEEE International Conference on Human-Robot Interaction (with technical cosponsorship from AAAI), and chaired the 1999 and 2000 AAAI Mobil Robot Competition and Exhibitions.
Representing and Reasoning about Preferences
Toby Walsh (NICTA and University of New South Wales)
Preferences turn up in many AI applications. How do we represent preferences? How we do elicit them from a user? How do we aggregate preferences of multiple agents? Can we do this fairly and efficiently? How do we reason about preferences (“I'd prefer a ski trip to a city break”) and constraints (“It must cost less than my budget”). In this talk, I will describe ongoing work to answer such questions. Research in this area is an interesting intersection between knowledge representation, decision theory, and multiagent systems. It is also an area where computational intractability can be our friend.
Toby Walsh is program leader of the Knowledge Representation and Reasoning Group at NICTA, Sydney (voted best city in the world 2006 by Conde Naste). He is also conjoint professor at the University of New South Wales, external professor at Uppsala University and an honorary fellow of Edinburgh University (surely the best city in the UK). He is editor-in-chief of the Journal of Artificial Intelligence Research, and was previously editor-in-chief of AI Communications. He will be program chair of IJCAI-11 to be held in Barcelona, Spain (voted one of the best cities in Europe). He greatly enjoys visiting Vancouver (voted best city in the Americas 2006).
Big "A," Small "I": Smart Ends from Simple Means
Matt Brown (Maxis / Electronic Arts)
Matt Brown's talk will cover various elements of the game design, behavioral AI and structure behind The Sims 2 as well as future efforts in products such as The Sims and SimCity. Emphasis will be on simplification of abstraction and underlying world representations, prototyping and demonstration of practical results. Discussion will focus around players' perceptual models and manipulating available information and expectations to simplify design.
Matt Brown was the technical director and designer for Electronic Art's FIFA franchise for many years and later for The Sims 2 at Maxis. He is now the creative/design director for future versions of The Sims and SimCity.. He is obsessed with creating the illusion of life through simplification, scale, and slight of hand as well as the practical application of AI.