Story Grid Writing Podcast

Art, Automation, and AI — Part 2

Episode Summary

Can you do math on a sentence? Picking up from last week, Shawn Coyne argues that an LLM produces mimicked artifacts, not original ones, and that confusing the two is where the "ghost in the machine" fear comes from. Then he goes somewhere most writers won't like: writing can be evaluated objectively, all the way down to the sentence. Drawing on Gödel, Church, Turing, Deutsch, and Shannon, he explains why a finished piece of writing is a fully computable artifact, walks Tim Grahl through the exact process he uses to give feedback on a scene, and lays out what he's actually been working toward for 35 years: formalizing what he knows well enough that a machine can analyze a writer's work the way he would, and get them feedback in seconds instead of weeks.

Episode Notes

Tim picks up the artifact-versus-artifice thread from last week and asks the practical question: what does it have to do with automation and AI?

Shawn's first answer is a distinction. An LLM takes in the corpus of human artifacts, finds the most popular patterns, and stitches them into something that looks original but is a mimicked artifact, not a new one. Artifice, the capacity to project something into the future that has never existed, stays with the human. From there the conversation turns to what AI is actually good for (speed and research), what prompting really is, and then into the deep end: whether a written sentence can be fully computed, and what that means for how editing works.

Topics covered:

• Original artifacts versus mimicked artifacts, and why an LLM output is an integration of what already exists rather than a projection of what could be
• Why AI behaves like a confirmation bias machine, and why it's hard to get it to tell you something you don't want to hear
• Jaron Lanier's "Wikipedia on steroids" framing, and why the people building AI present it as a savior
• Shawn's Library of Congress story: two weeks of research that would take a second today, and Tim's espresso machine example from a scene he was writing
• Prompting as programming: knowing exactly what you want, the same way you once had to learn how to search Google
• Why developmental editing is slow by nature: you can't evaluate the first sentence until you've read the whole artifact
• Content and context, show and tell, and everything packed into the two-word sentence "Tim screams"
• The Russian doll structure: sentence, scene, arc, quadrant, whole, and the mythic standard that applies at every level
• Gödel's incompleteness theorem, the Church-Turing thesis, David Deutsch, and Claude Shannon's information theory, and how Shawn uses all five to argue that a halted artifact can be fully computed
• Russell and Whitehead's Principia Mathematica and the failed attempt to prove that one is one
• Shawn's actual editing process for a scene: did anything change, find the sentence evidence for each of the five commandments, confirm each is shown rather than told
• Inciting incident, turning point, crisis, climax, resolution, and the A-P-A-P-A pattern of antagonism and protagonism across the five
• Why a progressive complication is a reaction, not a transformation, and why the antagonist always leads the inciting incident
• Why language is a form of math and math is a form of language, and why F=ma and "don't jump out the window" are the same message
• Why Shawn won't hoard what he knows, and what it would mean to formalize it so a machine can analyze a writer's work the way he does

Books and works referenced

• The Story Grid by Shawn Coyne
• Principia Mathematica by Alfred North Whitehead and Bertrand Russell
• The Odyssey, The Iliad, Pride and Prejudice, The Lord of the Rings, The Hobbit