Reasoning with data - A new challenge for AI?

Abstract

Artificial intelligence (AI) traditionally deals with knowledge rather than with data (with the noticeable exception of machine learning). The term "knowledge" refers here to information with a generic flavor, while "data" refers to information pertaining to (collections of) particular cases. The formalization of reasoning patterns with data has been much less studied until now than knowledge representation and its application to knowledge-based systems and reasoning, possibly in presence of imperfect information. Data are positive in nature by manifesting the possibility of what is observed or reported, and contrast with knowledge that delimit the extent of what is potentially possible by specifying what is impossible. Reasoning from knowledge and data goes much beyond the application of knowledge to data as in expert systems. Besides, the idea of similarity naturally applies to data and gives birth to specific forms of reasoning such as case-based reasoning, case-based decision, or even case-based argumentation, interpolation, extrapolation, and analogical reasoning. Moreover, the analysis, the interpretation of data sets raise original reasoning problems for making sense of data. This article is a manifesto in favor of the study of types of reasoning which have been somewhat neglected in AI, by showing that AI should contribute to (knowledge) and data sciences, not only in the machine learning and in the data mining areas.