next up previous
Next: A Non-Parameterized Evolutionary Approach Up: Engineering Creativity Previous: Engineering Creativity

BRUTUS: A Parameterized Logicist Approach

We are members of the Creative Agents project at Rensselaer,24 which extends and enhances the Autopoeisis Project. Autopoeisis, launched in 1991 with grants from the Luce Foundation, with subsequent support from Apple Computer and IBM, is devoted to building an artificial storyteller capable of generating ``sophisticated fiction." (A snapshot of the project's first stage was provided in [Bringsjord, 1992].) Though we confess that no such AI is currently on the horizon (anywhere), the agent BRUTUS.1 suggests that the dream driving Autopoeisis may one day arrive (in, say, BRUTUS.n). Though they aren't exactly Shakespearean, BRUTUS.1 is able to produce stories such as the following one.




BETRAYAL.1

Dave Atwood loved the university. He loved its ivy-covered clocktowers, its ancient and sturdy brick, and its sun-splashed verdant greens and eager youth. He also loved the fact that the university is free of the stark unforgiving trials of the business world -- only this isn't a fact: academia has its own tests, and some are as merciless as any in the marketplace. A prime example is the dissertation defense: to earn the PhD, to become a doctor, one must pass an oral examination on one's dissertation. This was a test Professor John Irons enjoyed giving.

Dave wanted desperately to be a doctor. But he needed the signatures of three people on the first page of his dissertation, the priceless inscriptions which, together, would certify that he had passed his defense. One of the signatures had to come from Professor Irons, and Irons had often said -- to others and to himself -- that he was honored to help Dave secure his well-earned dream.

Well before the defense, Dave gave Irons a penultimate copy of his thesis. The professor read it and told Dave that it was absolutely first-rate, and that he would gladly sign it at the defense. They even shook hands in Iron's book-lined office. Dave noticed that John's eyes were bright and trustful, and his bearing paternal.

At the defense, Dave thought that he eloquently summarized Chapter 3 of his dissertation. There were two questions, one from Professor Rogers and one from Dr. Meteer; Dave answered both, apparently to everyone's satisfaction. There were no further objections.

Professor Rogers signed. He slid the tome to Meteer; she too signed, and then slid it in front of Irons. Irons didn't move.

``John?" Rogers said.

Irons still sat motionless. Dave felt slightly dizzy.

``John, are you going to sign?"

Later, Irons sat alone in his office, in his big leather chair, saddened by Dave's failure. He tried to think of ways he could help Dave achieve his dream.




BRUTUS.1 can generate stories like this one because, among other reasons, it ``understands" the literary concepts of self-deception and betrayal via formal definitions of these concepts. (The definitions, in their full formal glory, and the rest of BRUTUS' anatomy, are described in [Bringsjord & Ferrucci, 1997].) To assimilate these definitions, note that betrayal is at bottom a relation holding between a ``betrayer" (sr in the definition) and a ``betrayed (sd in the definition)." Then here is a (defective) definition that gives a sense of BRUTUS.1's ``knowledge:"


Agent sr betrays agent sd iff there exists some state of affairs p such that

1
sd wants p to occur;
2
sr believes that sd wants p to occur;
3
sr agrees with sd that p ought to occur;
4'
there is some action a which sr performs in the belief that thereby p will not occur;
5'
sr believes that sd believes that there is some action a which sr performs in the belief that thereby p will occur;
6'
sd wants that there is some action a which sr performs in the belief that thereby p will occur.

BRUTUS.1 also has knowledge of story structures in the form of story grammars. For example, BETRAYAL.1 conforms to the following story grammar, taken from Thorndyke [Thorndyke, 1977]. That which flanks `+' comes sequentially; the asterisk indicates indefinite repetition; parentheses enclose that which is optional; brackets attach to mutually exclusive elements.

Rule No. : Rule
(1) :Story $\rightarrow$ Setting + Theme + Plot + Resolution

(2)

:Setting $\rightarrow$ Characters + Location + Time

(3)

:Theme $\rightarrow$ (Event)$^\ast$ + Goal

(4)

:Plot $\rightarrow$ Episode$^\ast$

(5)

:Episode $\rightarrow$ Subgoal + Attempt$^\ast$ + Outcome

(6)

:Attempt $\rightarrow$ $\left\{
\begin{array}{l}
\mbox{Event$^\ast$ } \\
\mbox{Episode}
\end{array} \right.$

(7)

:Outcome $\rightarrow$ $\left\{
\begin{array}{l}
\mbox{Event$^\ast$ } \\
\mbox{State}
\end{array} \right.$

(8)

: Resolution $\rightarrow$ $\left\{
\begin{array}{l}
\mbox{Event} \\
\mbox{State}
\end{array} \right.$

(9)

$\left.
\begin{array}{l}
\mbox{Subgoal} \\
\mbox{Goal}
\end{array} \right\}$ $\rightarrow$ Desired State

(10)

$\left.
\begin{array}{l}
\mbox{Characters} \\
\mbox{Location} \\
\mbox{Time}
\end{array} \right\}$ $\rightarrow$ State


Is BRUTUS.1 creative? Perhaps not.25 After all, BRUTUS.1 is capable of generating only a small portion of the space $\cal I$ of all interesting short-short stories: the formalisms that make up BRUTUS's soul seem to leave out much of this space.26 Even a future BRUTUS.n which includes knowledge of all presently deployed literary concepts (unrequited love, friendship, revenge, etc.), all known story grammars, and so on -- even such a system, we suspect, would inevitably fail to capture the essence of $\cal I$. Our suspicion is based on the intuition that $\cal I$ is productive, in the technical sense: A set $\Phi$ is productive if and only if (i) $\Phi$ is classically undecidable (= no program can decide $\Phi$), and (ii) there is a computable function f from the set of all programs to $\Phi$ which, when given a candidate program P, yields an element of $\Phi$ for which P will fail. Put more formally (following [Dekker, 1955], [Kugel, 1986], [Post, 1944]):

Put informally, a set $\Phi$ is productive if and only if it's not only classically undecidable, but also if any program proposed to decide $\Phi$ can be counter-exampled with some element of $\Phi$. Evidence for the view that $\cal I$ is productive comes not only from the fact that even a descendant of BRUTUS.1 would seem to leave out some of $\cal I$, but from what the Autopoeisis team has experienced when attempting an outright definition of the locution `s is an interesting short-short story:' Every time a definition of this locution is ventured, someone comes up with a counter-example. (If it's proposed that all members of $\cal I$ must involve characters in some form of conflict, someone describes a monodrama wherein the protagonist is at utter, tedious peace. If it's proposed that all members of $\cal I$ must involve one or more key literary concepts (from a superset of those already mentioned: betrayal, unrequited love, etc.), someone describes an interesting story about characters who stand in a novel relationship. And so on.) So a key question arises: What about attempts to engineer creativity without trying to pre-represent the space of the artifacts desired?


next up previous
Next: A Non-Parameterized Evolutionary Approach Up: Engineering Creativity Previous: Engineering Creativity
Selmer Bringsjord
2000-03-13