Artificial Intelligence (AI) is the study of how computer systems can simulate intelligent processes such as learning, reasoning, and understanding symbolic information in context. AI is inherently a multidisciplinary field, and draws upon work in algorithms, databases and theoretical computer science.

IBM and IBM Research have been active in AI since its earliest days, when Arthur Samuels (in the 1950s) developed an expert checkers-playing program that learned from experience. Forty years later, IBM Research's chess-playing program Deep Blue made history by beating world chess champion Gary Kasparov.

AI research at IBM goes beyond game-playing programs and is at the forefront of many of the most active areas of Artificial Intelligence.

AI Methodologies

Our current AI research spans a wide variety of techniques, including learning, knowledge representation, logic programming, Bayesian reasoning, planning, and intelligent agents.

Learning includes techniques that focus on finding general patterns from a fixed set of data, and then applying these generalizations to new data. These techniques allow a system to improve its performance over time. Our research also includes optimization techniques which seek to maximize a system's performance according to a fixed metric by tuning a set of parameters on that system.

Bayesian reasoning combines probabilistic reasoning with graph-theoretical structures called Bayesian nets. Representing probabilistic information in a graph enables much of the intensive computation of probabilities to be performed efficiently, using graph algorithms. Bayesian nets are particularly useful for reasoning about and representing causal information.

Knowledge representation provides methods for capturing the complexity of the information in our environment. The focus is on developing representations, such as semantic networks, that facilitate reasoning with the information. We are also studying common sense reasoning methods, such as plausible reasoning, or reasoning from the absence of information.

Logic programming exploits the equivalence between theorem proving and computation, and provides methods for specifying the behavior of a system in declarative terms. Researchers at IBM helped pioneer constraint logic programming and are leading the effort to represent prioritizations in logic programs.

Planning techniques focus on finding a sequence (or some other structure) of actions that can achieve a particular goal. Our algorithms for efficient search and evaluation of states are crucial to our success in game-playing systems.

AI Projects

AI techniques are being used to solve challenging problems and to advance the state-of-the-art in a wide variety of areas, including e-commerce, supply chain management, intelligent tutoring, data mining, knowledge management, exploratory vision, and handwriting recognition.

e-Commerce: The explosion of business transactions over the Internet raises many research problems. Because many transactions are automated, and are accomplished through intelligent agents, it is necessary to understand information economies, large systems of software agents buying and selling goods over the Internet. Our research agenda includes the study of information filtering, auctions, price wars, representing business rules and resolving conflicts among these rules, shopbots, information bundling, and multiagent learning.

Intelligent Tutoring: Intelligent tutoring systems construct models of a student's understanding and interact with a student based upon that model. Because learning often takes place in a group setting, we are expanding this paradigm by considering the interaction among a team of students. We are developing an intelligent tutoring system that monitors and manages teams of students as they collaborate to solve distributed, multistep problems.

Performance Management: It is necessary to manage many kinds of computer services, such as Web transactions and electronic mail, in order to guarantee good service and availability. We seek to increase the level of automation for performance management. Our work includes the study of predictive detection, which gives advanced knowledge of performance degradations, and event mining, a form of data mining which recognizes situations that require specific action to ensure good performance.

Knowledge Management: Applications of knowledge management - the management of large bodies of information for use in business organizations - include mail categorization, information retrieval, and question answering. Research includes the development of learning and classification techniques, the integration of semantic networks with classical search engines, and the development of sophisticated extensions of semantic networks.

Exploratory Vision: Intelligent vision requires the ability to understand the image that a camera or other sensory device produces. This requires providing the computer with a representation that it can manipulate, along with the background knowledge, context, and computational and inferential methods that it needs to understand the image. We are studying how vision systems can be used in consumer settings to automatically recognize and classify objects.

Other projects that are closely related to Artificial Intelligence include Knowledge Discovery and Data Mining and Natural Language Processing .

Please contact Paridhi Verma to obtain copies of the Computer Science Brochure