My hope is that Expressive Processing will, in turn, help broaden understanding of Minstrel‘s significance for media authoring and thinking through wider issues about software.

The first is an interesting question about cognition. If I have the knowledge to understand the problem shown in pp 24, then why would I invent that solution in the first place? And yet as people we’re constantly considering solutions and then rejecting them. That would seem to suggest that people have different processes for “creating” solutions and “evaluating” them, and that those processes use different sets of knowledge. That’s a pretty odd and surprising result when you think about it.

Second, and probably more relevant to your discussion here, is the question of how important the hand-encoding of the knowledge is to a brittle scruffy AI system. One defense of systems with very limited domains (like Minstrel) is “real machine learning would solve the problem of limited knowledge.” But even if we had robust machine learning that could (say) read a bunch of King Arthur books and create a large case-based memory, who can say whether Minstrel could operate as well on those memories as it does on the ones I hand coded? (I’d argue that it could, but the point is that there’s really nothing in Minstrel to convince anyone of that.)

I should probably stop finding faults in my own work :-)