"Abduction is inference to the best explanation and has applications to diagnosis, plan recognition, natural language understanding, vision, and many other tasks. It is frequently formalized as constructing a set of assumptions that logically imply and therefore 'explain' a set of observations. ... Below is an extreme example of abduction from Eugene Ionesco's play 'Rhinoceros' from the 'Theater of the Absurd' school:

- Machine Learning Research Group University of Texas at Austin, Department of Computer Sciences, Artifical Intelligence Laboratory. Also see their related publications.

A robot that likes to play with test tubes. By David Akin. The Globe & Mail (January 17, 2004). "Special artificial intelligence software was written to give the machine the ability to formulate a scientific hypothesis about genes and the genetic makeup of yeast. In other words, it could learn and apply what it learned to solve new problems or learn new things. Once it formed a hypothesis, it had to figure out how to test its theory. This, usually, is where that human intuition comes in. Scientists, having made a particular observation about the world, are often faced with thousands or even millions of ways of figuring out how the condition they observed came to be. To do this, scientists employ what is know as abduction. Abduction is often described as the reverse of deduction. It is a form of logical reasoning in which one starts with an observation of a certain condition and then tries to isolate the variables or causes of that condition. If I have two marbles and you have three marbles, you may engage in deduction to conclude that there are total of five marbles. But if all we know is that there are five marbles and two of us, you might engage in abduction to form five possible hypotheses about how many marbles each of us has. (I have none, you have five; I have one, you have four; I have two, you have three; and so on.) Scientific problems can frequently generate thousands and even millions of such hypotheses when using this abductive technique. Human investigators frequently use intuition, instinct and a little luck to zero in on a few likely hypotheses that can be tested. The Robot Scientist, too, has a special hypothesis-generation engine built into it so that it can ask the 'how' and the 'why' questions to explain real-world phenomena."

• Related news articles on our Scientific Discovery page.

What is Abductive Inference? By Uwe Wirth, Frankfurt University. "Abductive reasoning: constitutes according to Peirce the 'first stage' of scientific inquiries (CP 6.469) and of any interpretive processes. 'Abduction' is the process of adopting an explanatory hypothesis (CP 5.145) and covers two operations: the selection and the formation of plausible hypotheses. As process of finding premisses, it is the basis of interpretive reconstruction of causes and intentions, as well as of inventive construction of theories. ... Recently the concept of abductive reasoning as reasoning to the best explanation is introduced and discussed in the field of Artificial Intelligence and expert systems (See van der Lubbe 1993). Expert systems aim at imitating the reasoning process and the human faculty to deal with uncertain information in a very efficient way. The question, however, is, how abductive inference as a pragmatic strategy of reasoning can be implemented in expert systems and whether artificial intelligence as a computational automatism can make creative guesses."

Related AI Topics Pages

• Logic
• Scientific Discovery

Other References Offline

Abduction, Reason, and Science: A Review. By Atocha Aliseda. AI Magazine 23(1): Spring 2002, 113-114. Review of Abduction, Reason, and Science: Processes of Discovery and Explanation, by Lorenzo Magnani, New York, Kluwer Academic/Plenum Publishers, 2001, 205 pages, ISBN 0- 306-46514-0. "Broadly speaking, abduction is a reasoning process invoked to explain a puzzling observation."

Abduction, Experience, and Goals: A Model of Everyday Abductive Explanation. By David B. Leake. The Journal of Experimental and Theoretical Artificial Intelligence. 7:407-428, 1995. 25 pages. Abstract: "Many abductive understanding systems generate explanations by a backwards chaining process that is neutral both to the explainer's previous experience in similar situations and to why the explainer is attempting to explain. This article examines the relationship of such models to an approach that uses case-based reasoning to generate explanations."

Learning, Bayesian Probability, Graphical Models, and Abduction. By David Poole. To appear, Peter Flach and Antonis Kakas, editors, Abduction and Induction: essays on their relation and integration, Kluwer, 1998. Abstract: "In this chapter I review Bayesian statistics as used for induction and relate it to logic-based abduction. Much reasoning under uncertainty, including induction, is based on Bayes' rule. Bayes' rule is interesting precisely because it provides a mechanism for abduction."