December 1995 (Vol. 10, No. 6) pp. 12–13
0885-9000/95/$31.00 © 1995 IEEE
Published by the IEEE Computer Society
Published by the IEEE Computer Society
Guest Editors' Introduction: Environmental Applications of AI
PDFs Require Adobe Acrobat
One of the most significant problems confronting our civilization at the end of the twentieth century concerns the status and future of the environment. Our understanding of ocean, land, and atmospheric processes is still sparse, yet decisions in this area impact not only the quality of life globally, but also our ability to sustain life. It is therefore essential that we make these decisions using the best available knowledge and tools.
Much environmental research is concerned with information processing. Environmental monitoring involves acquiring, archiving, analyzing, and retrieving enormous amounts of sensed data. Natural resource management, which examines and recommends policies on resource use, involves complex decision making that relies not only on analysis of sensed data, but also on the results of models and simulations. To carry out policy, environmental regulatory commissions must disseminate information regarding laws and regulations. They must also process and assess an incredible labyrinth of forms, while dealing at several levels with an ever-changing set of governmental regulatory policies. In turn, manufacturers and businesses must be aware of the information regarding regulatory requirements, not only to avoid costly penalties and cleanup expenses, but to incorporate environmental constraints up front during product design. In this way, subsequent manufacturing, packaging, and maintenance can follow regulatory guidelines.
Other information-related environmental tasks include robotic exploration or cleanup of toxic or dangerous environments, such as undersea oil pipelines or nuclear reactors, monitoring and control of harmful emissions in the atmosphere, and the cataloging of species, living and dead.
Providing tools for these information-centered tasks is a crucial challenge for information technology. In many environmental applications, information systems present specific challenges that AI techniques can address:
- Representing extremely vast data repositories in ways that enable reasoning and information extraction. Other representation challenges occur in simulation and modeling of complex phenomena, and modeling users or organizations for automatic information gathering and dissemination.
- Using specialized reasoning techniques (for example, in the presence of uncertainty, reasoning within a temporal context, with constraints, or with distributed evidence).
- Applying symbolic and subsymbolic reasoning in the presence of heterogeneous knowledge sources (for example, symbolic cases, databases, analytical models, and nonsymbolic data).
- Representing, indexing, and learning from data gathered in heterogeneous media.
In this issue we present five articles from the AAAI symposium, "AI & Environmental Applications," held in Seattle in August, 1994, in conjunction with the 12th National Conference on AI. The articles address a variety of interesting environmental problems. They share characteristics that we believe indicate the complexity and variety of environmental applications of AI:
- They describe a variety of application areas and AI techniques, which is typical of a field in early stages of development.
- They involve the users in different capacities, rather than provide total problem-solving capability.
They are usually embedded applications, rather than stand-alone systems. "Undersea Exploration in Antarctica Using a Robotic Submarine," by Carol Stoker, Donald Barch, Butler Hine, and James Barry, details their design of and experience with a virtual reality system controlling a remotely operating underwater vehicle in Antarctica. "Mission to Planet Earth," by Nick Short, Jr., and his colleagues at NASA's Goddard Space Flight Center, describes the AI aspects of the prominent NASA program of the same name. This program uses a commercial object-oriented database management system as the basis of knowledge representation. Stan Matwin, Daniel Charlebois, David G. Goodenough, and Pal Bhogal present a system that uses learning, planning, and plan reuse to help users query a complex, heterogeneous forestry information system. Cindy Mason from NASA Ames Research Center describes a system for detecting potential violations of the Nuclear Test Ban Treaty, based on nonmonotonic reasoning and truth maintenance. Eric Jones and Aaron Roydhouse from the Victoria University of Wellington present a case-based system that helps meteorologists access historical situations of interest.
We hope that these articles will further inform the applied-AI community about the importance of environmental applications. Furthermore, we hope that this issue spawns new applications for solving interesting environmental problems.
Cindy Mason is a principal investigator in automation sciences research at the NASA Ames Research Center. Her research projects have included the development of agent architectures and languages for a variety of intelligent network applications, including a global network of seismic monitoring stations for a comprehensive test ban treaty. She has been active in organizing workshops on environmental applications in the AI community. She can be reached at the Information Sciences Division 269-2, NASA Ames Research Center, Moffett Field, CA 94035; firstname.lastname@example.org.
Stan Matwin is a professor of computer science at the University of Ottawa, Canada. His research interests are in machine learning and its applications, and in knowledge-based systems. He is president of the Canadian Society for Computational Studies of Intelligence and head of the International Federation for Information Processing Working Group 12.2 (machine learning). He has worked for several years with Forestry Canada and the Canada Centre for Remote Sensing on applications of learning and planning in the natural resources sector. He works currently on inductive machine learning and its applications, with a special focus on combining data- and knowledge-driven learning. He is research director of a project with MDA, Inc., of Vancouver, Canada, that applies learning to the detection of oil spills on sea from satellite radar images. He can be contacted at the Dept. of Computer Science, Univ. of Ottawa, 150 Louis Pasteur, Ottawa, Ontario K1N 6N5, Canada; email@example.com.