Understanding by Design – the synthetic approach to intelligence Daniel Bisig*+ and Rolf Pfeifer* * Dept. of Informatics, University of Zurich, 8050 Zurich, Switzerland +Inst. for Computer Music and Sound Technology, University of the Art Zurich, 8048 Zurich, Switzerland Designing and building systems for selected abilities is a core activity of research in artificial intelligence. One of the attractive points of proceeding in this manner is that we not only end up with an actual system, but along the way, we learn a lot about the phenomenon of study, e.g. spontaneous structure formation or rapid locomotion. One of the crucial questions concerns the level at which the system should be designed. In this paper, we introduce the synthetic approach to intelligence. This approach promotes design and devel- opment as a research methodology. But in contrast to classical engineering, “Understanding by Design” promotes a bottom-up design stance based on “design for emergence”. These ideas are illustrated with a number of case studies from robotics and human-computer interaction. Conceptual Background and Motivation “Understanding by Design” is the name of a particular research methodology that is increasingly employed in dif- ferent kinds of sciences. This methodology forms the main foundation of embodied AI whose research and design concepts are summarized in the text book «Understanding Intelligence» (Pfeifer and Scheier 1999, 2001) and in the more popular science book «How the Body Shapes the Way We Think – A New View of Intelligence» (Pfeifer and Bongard 2007). Very briefly, “Understanding by Design” is based on the assumption that at least some natural phenomena can best be understood by building artifacts that embody a selected set of these properties. The process of building the arti- fact and its subsequent exposure to experimentation represents a tangible form of conceptualization that leads to testable hypotheses. In addition, this approach offers the benefit that it often entails engineering innovations that possess general application potentials beyond the particular phenomenon under investigation. The methodology of “Understanding by Design” combines in itself two types of ideas, the synthetic approach and design for emergence. The synthetic approach introduces engineering practice into scientific research. Corre- spondingly, a particular phenomenon of interest (e.g. how do we recognize a face in a crowd, or how do we move and walk) is approached from an implementation perspective. Design for emergence tries to minimize designer bias and the pre-definition of the artifact’s resultant properties. This represents a novel approach uncommon both to engineering and science. The synthetic approach originated from within the framework of pre-cybernetic robotics. The term “synthetic method” was employed by psychologist Kenneth Craik to describe the process of testing behavioral theories through machine models [Craik, 1943]. The synthetic approach is not meant as a replacement but rather a comple- ment for the traditional analytical approach. The analytical sciences are very well established and have contributed