We identify and address a fundamental general problem which we regard as crucial for the widespread, effective use of decision support systems (DSS) in the future: how can we substantially improve the quality of interaction, and the degree of flexible engagement, between humans and computers? Rather than seeking an answer in additional technical functionality, we propose a new paradigm for computing that is human-centered and that adopts a novel, observation-oriented approach to data modelling. We report a recent practical work (a timetabling instrument) showing an unusual degree of openness for interaction, and we give evidence that our approach can encompass conventional tools such as expert systems. D 2002 Elsevier Science B.V. All rights reserved.