The Rational and Irrational Minds of Others

When one surveys Professor Carbonell's many and varied research contributions, two more or less constant methodological trends seem to stand out. They are the preference for an integra-tive outlook and insistence on developing computational systems. The integrative outlook stresses the multiplicity of causes in behavior and contributing factors in reasoning. It is a rather difficult position to maintain. Indeed, with ever-accelerating specialization in science, the temptation is to concentrate on a narrow research area where one has more chances to attain deeper understanding. Still, while pursuing research in planning, reasoning, language understanding or learning, Jaime, it seems to me, has maintained his original interest in developing comprehensive models of human cognitive capacities. This choice can be traced back to POLITICS (Carbonell, 1978), which addressed reasoning, memory, intentional behavior, language processing and learning. At the same time, Jaime has insisted on developing proof of concept systems and applications that made the underlying theories and models tangible. Notably, whenever possible he opted for building systems that carried explanatory power in addition to producing expected deliverables. Moreover, in the best traditions of " good old-fashioned AI " he preferred, again, whenever possible, to develop applications that aimed not only to attain ever more spectacular functionalities but to do so while modeling cognitive capacities of humans. I have singled out the trends of integration-orientedness and system-building by applying analogical reasoning: these trends can be discerned in the research my team and I have been carrying out over the years as well. Quite possibly , this is due to a historical bias: after all, I worked in Jaime's group for seven years in the 1980s and 1990s, notably, on the KBMT-89 knowledge-based machine translation project (Nirenburg et al., 1992). In what follows I will leave the etiology of this analogy aside and discuss some selected preferences and topics of our research for which POLITICS can be considered an inspiration. Outside the work on generic architectures, the prevalent trend in the field is to separate the overall task of modeling cognitive processes into autonomous constituent subfields. This tactical decision is understandable because addressing the overall problem may require more resources of all kinds than are available. Indeed, the typical explanation of this choice is " this is how science is done. " One of the celebrated examples is the separation of syntax from semantics and semantics from pragmatics in theoretical linguistics, a choice that trickled down …