Brutus and the Narrational Case Against Church's Thesis
Selmer Bringsjord, Dave Ferrucci · 1999
Overview There was a time, about a decade ago, when one of us (Bringsjord) thought that he would “crack” story generation this way: figure out what makes a story interesting from the standpoint of logic, proceed to formalize interestingness in some logical system, and then code this formalization . . . and presto! — he would then have a program that generates interesting stories. This idea proved to be a painful dead end. (Many of you are probably not surprised.) Whereas sometimes counter-examples to proposed definitions for concepts lead the way to improved definitions that promise to eventuate in the correct definition, nothing of the sort happened in dialectic at RPI about what conditions are necessary and sufficient for a story’s being interesting. In fact, it began to occur to us that perhaps the set of all interesting stories — denoted SI — was such that (i) there is no algorithm for deciding whether or not a story is in this set, and (ii) there is an algorithm for shooting down any algorithm proposed to decide SI . Some readers will immediately realize that properties (i) and (ii) characterize what are known in computability theory as productive sets. If the set of all interesting stories is a productive set, and if humans are able to decide, swiftly and routinely, which stories are interesting and which are not, then Church’s famous and almost universally affirmed thesis — that what can be effectively computed is co-extensive with what can be algorithmically computed — must be false. We adumbrate and defend this reasoning herein. If the reasoning is correct, the upshot would seem to be that those who would build literarily creative agents in the form of conventional computers and computer programs may at best find engineering “tricks” for giving these agents some mere approximation of what human authors tap when they craft interesting narrative. We conclude this paper with