Publication 3765

In order to construct scientifically reasoning artifacts we not only have to close the loop between hypothesis generation and evaluation but also to make the system embodied. To genuinely understand scientific insights, “robot scientists” need to represent scientific knowledge within their own representational structure rather than in terms of a priori defined logical propositions. Two main features of such systems are identified: projective constructivism that reverses the flow of information processing, and cognitive canalization that reduces computational requirements.