1. IntroductionThis is a summary of recent research on the discovery process in which a commonframework for both psychological studies and machine learning (ML) approaches toscientific discovery is described. Early interest in the psychology of science can betraced to Bruner, Goodnow, & Austin (1956), Wason (1960), and Simon (1966, 1973),among others. The state of the art as of a dozen years ago is summarized in Tweney,Doherty, and Mynatt (1981). The more recent resurgence of interest in the "cognitivescience of science," particulary in ML can be attributed to Simon and his colleagues(Langley, Simon, Bradshaw & Zytkow, 1987; Kulkarni & Simon, 1988; Qin & Simon,1990, Cheng & Simon 1992; Valdes-Perez, Simon & Murphy, 1992). However, beyondthe projects that focus on the construction of computational models of the discoveryprocess (c.f., Shrager & Langley, 1990), there is a substantial amount of psychologicalresearch on various aspects of scientific discovery (e.g., Chin & Brewer, 1992; Brewer& Samarapungavan, 1991; Dunbar, 1989; Dunbar & Schunn, 1990; Gholson, Standish,Neimeyer & Houts, 1989; Giere, 1988; Gorman, 1992; Holland, Holyoak, Nisbett T Klayman & Ha, 1987; Kuhn, 1989; Schauble, 1990; Schauble, Klopfer& Raghavan, 1991; Sodian, Zaitchik & Carey, 1991; Vosniadou & Brewer, 1992).2. Scientific discovery as dual searchScientific discovery is a type of problem solving in which search takes place in twoquite distinct spaces: an hypotheses space and an experiment space (Klahr & Dunbar,1988). Each space represents accumulated knowledge, a set of operators, and a set ofgeneral heuristics for constraining search. Additionally, the process requires operatorsfor mapping between the two spaces. The SDDS ("Scientific Discovery as Dual Search")framework summarizes the way in which these component processes are organized. Itsthree major components include the following: