What a Computational Storyteller (MEXICA) Can Tell Us about Creativity?Share
Can a computer system successfully draft, evaluate, and drive a narrative? Computational creativity researchers have considered this question for decades. MEXICA: 20 Years-20 Stories is the first book of short stories produced by a creative agent capable of evaluating and making judgments about its own work.
In the preface to Rafael Pérez y Pérez’s MEXICA: 20 Years-20 Stories, D. Fox Harrell, a professor of digital media at MIT, depicts a dramatic (and hopefully fictional) story.
A computer scientist is being heckled by a mob of “true humanists” armed with pitchforks, torches and picket signs. They are angry and aggressive. They don’t want computers anywhere near Art—the holy grail of humanity. Then the disheartened computer scientist regains his courage and in a solemn salvo addresses the crowd:
“You call yourself ‘The Humanist Thrall,’ but look at me as well. Am I not a humanist myself? I’m a humanist and a computer scientist; the two can be one. I am a seeker of answers: ‘What makes a story interesting?’ ‘What makes a story novel?’ ‘What makes a story creative?’ (…) These are my questions! (…) [Computer programs] play with blocks, paint, make music and tell stories, but most importantly they reflect us humans. They enable us to think about how we do things and why we do them.” (MEXICA: 20 Years-20 Stories, page IX)
In this dramatic opening, we see an all too familiar tension between two distinct approaches to creativity. In one camp, there are people who perceive the act of creation as an ultimately human, even “magical” endeavor, spurred by the enchanting alchemy of inspiration and imaginative prowess.
But there are others—people who dare to challenge the seemingly taboo relation between computation and creation. They are exploring creativity from a different angle, and their aim is to build creative machines.
The Story of MEXICA
MEXICA: 20 Years-20 Stories contains 20 short narratives written in Spanish and English developed by the computer program MEXICA. Each story takes place in the ancient world of Mexicas (also known as Aztecs), where fictional characters interact with each other in a dramatic fashion. MEXICA is the first book produced entirely by a creative agent that is capable of evaluating and making judgments about its own work.
This is truly remarkable. The system generates the story content, reflects upon its creation, and decides when the story ends. MEXICA is also capable of reviewing a plot for consistency before presenting it to readers.
As the father of this system, Pérez y Pérez describes the creative processes in MEXICA as a continuous interplay between two phases: engagement and reflection. During engagement, the system scans and generates sequences of events. During reflection, it pauses to evaluate and modify the material.
MEXICA has the ability to break impasses; check for plot coherence; and assess the novelty (or interestingness) of the story. Then the system continues to generate a story using self-thought guidelines acquired during the reflection phase, modifying the constraints to manage the next stage of production even more effectively.
The recent publication of these 20 collected stories is the celebration of an automatic storyteller that’s been twenty years in the making. MEXICA differs from other statistical models because it is inspired by how humans actually develop fictional stories. For a better, more precise description of the process itself, please see the book’s Afterword or “MEXICA: A computer model of a cognitive account of creative writing.”
The Princess and The Jaguar Knight
The plots throughout MEXICA: 20 Years-20 Stories are congruent and engaging. By working with pre-defined phrases (not a natural language), these stories are more akin to a canvas or framework than a full story.
For my minimalistic taste, this sort of prose is not an obstacle. I found it effortless to follow the plot and fill in any gaps with my imagination. What is truly noteworthy is the drama, tension and unraveled continuity of each story.
Story 16 was my favorite. It reads like a plot for a dark, psychological thriller in which two characters, a princess and a jaguar knight, are involved in an intense, violent love/hate debacle. Here’s an excerpt:
“The princess was unable to accept that the jaguar knight did not love her.
In that instant the princess wondered why the jaguar knight was acting this way.
While trying to frighten the jaguar knight with a sudden strike, the princess injured herself. Because of the jaguar knight’s bitterness, he planned to let her die.” (MEXICA: 20 Years-20 Stories, page 38)
MEXICA, just like its human counterpart, is an author with its own easily recognized style. Minimalism mixed with gore makes for an unusually intoxicating read. Each story is simple and intense, and I can fully envision a human author using MEXICA as a co-creative agent who generates cues both coherent and dramatic for storylines.
Big Questions Asked by Computational Creativity
Coming back to Harrell’s preface with the computer scientist and angry mob of “true humanists,” here there is no question in my mind that MEXICA will likely face some snarky comments like, “Is MEXICA better than Hemingway?” and “Is art created by a computer really art?"
I find all these accusations to be naive, desperate and misguided. Our perception of art is diabolically complex, as I discussed in a previous article. The real, unique value of building computer systems capable of creation is their ability to efficiently and elegantly test hypothesis. In Pérez y Pérez’s own words:
“Through this computational model, I try to get answers to questions like: What are the processes involved in creative writing? How do we associate novel ideas during the generation of arguments? How do we evaluate a story?”
(MEXICA: 20 Years-20 Stories, page 53)
All of these questions are indeed vital for our actual understanding of creativity.
Pérez y Pérez, like other brilliant computational creativity researchers, seeks the deep understanding of the creative process itself. We all can benefit greatly from their inquiry of trials and errors. I highly appreciate Pérez y Pérez’s definition of computational creativity, too, “as the interdisciplinary study of the creative process employing computers as the core tool for reflection and generation of new knowledge.”
As a storyteller, MEXICA tells us that reducing the creative process to pure idea generation is futile. Instead, an ability to evaluate and follow a self-imposed set of constraints is what guides our creative process in a successful, coherent direction. MEXICA: 20 Years-20 Stories is not only a fun read—it contains knowledge about the process of creation itself.
A very special shout-out to my wonderful editor Ashley Turner, who graciously and patiently helps me put my thoughts together in a not-so-chaotic form.