The question of whether human thinking can come to simulate computer intelligence--i.e., AI in reverse--is addressed in this paper. Examples are given of three computer tools which perform several functions that constitute an intellectual partnership between student and tool. Such functions include: (1) assuming part of the intellectual burden in experimentation; (2) presenting learners with novel alternatives; (3) freeing the individual from lower level operations; (4) displaying intermediate processes; (5) providing intelligent tutelage; and (6) providing models of information representation, processes, and strategies. Conditions necessary for internalization of artificial intelligence functions are described, i.e., comprehensibility, generalizability, novelty, utility, explicitness, and mindful abstraction. An experiment in which 74 Israeli seventh graders were divided into three groups and interacted with a version of the software incorporating metacognitive guidance and content-specific questions (experimental group), questions only, or neither is then reported. Results show that the experimental group devoted the most time to reading the texts and scored highest on a posttest; these findings support the hypothesis that interaction with a tool that models reading-related metacognition leads to improved reading of new texts in another non-computerized setting. Implications of these findings for a cognitive-developmental theory, a cultural-communicational theory, and educational theory and practice are examined. (28 references) (MES)