In the last decade we have seen the boundless enthusiasm for tutoring humans using "intelligent" machines founder, as system after system has failed to live up to expectations. For the most part, such systems can be seen to rest on similar philosophical foundations, conceptions of mind and thought rooted firmly in the Rationalist tradition that forms the cornerstone of modern science. This foundation is two-tiered, centered around an epistemology that characterizes knowledge as internalized and finite symbolic representations of reality and a metaphysics that assumes the existence of absolute truth, not only where physical reality is concerned, but also in terms of semantics. Evidence of the Representationalist mindset can be found in the underlying assumptions of many areas of science, education, and philosophy throughout history, but is especially apparent in Artificial Intelligence (Al), where the status of knowledge and meaning have become the central issues. Examples of such assumptions found in AI, and in tutoring systems in particular, include:

• The notion that meaning can exist outside of context. That meaning is absolute and finite and can therefore be described symbolically.
• That human knowledge exists as a finite, concrete interpretable structure in the brain.
• That "not knowing" is generally a problem of content. Symbolic forms are semantically detenninistic and so learning is reduced to the definition of new facts/relationships.
• A metaphor of learning based on knowledge transference; the teacher the font of knowledge and the student a vessel to be filled.

Recent work by both sociologists and educators suggests that the assumptions listed above are erroneous and situate the source of meaning firmly in its context of use. Among other things, this implies that learning can not be separated from acting physically and (especially) linguistically.