Research in Conversation · MLL Colloquium
What Is Left to Write?
Richter’s Grey, the Machine’s Em Dash — and What Makes Us Us
Thursday 1 October 2026 · 5:30–7:15 pm, talk and refreshments
Diffenbaugh 009, Florida State University
With a moderated panel: Joachim Adams and Susan Bryson
About the talk
When Gerhard Richter painted his canvases grey, it was out of a crisis: he did not know what was left to paint. Many in the academy know the feeling now. In creative writing, in programming, in mathematical proof, machines pass the tests of whether a made thing works. What few of those tests were built to see is form.
This talk follows one piece of form, the em dash, from Dickinson to the Commonwealth Short Story Prize: how models came to overuse it, then partly stopped, and what training does and does not explain. What remains is the question: what makes us us when the machine writes well?
Further reading
Machines that pass the tests
- Amid a flood of AI advances, astrophysicists are questioning the soul of their field Astronomers on what their field is for once machines can do the work.
- Care for a little more AI? A Fields medallist on a machine-found proof in his own field, which he calls “short and particularly elegant”.
- When Will AI Exceed Human Performance? Evidence from AI Experts In a 2016 survey, AI researchers’ median guess for a machine writing a high-school essay was 2026; for a bestselling book, 2049.
How machines write
- ChatGPT and Gemini AIs Have Uniquely Different Writing Styles Each chatbot writes in a style of its own, much as each of us does. The paper behind it: Rudnicka and Juzek, “Beyond ‘AI Language’: The case for the idiolectal nature of LLM output”, arXiv 2608.06589 preprint
- Model Misalignment and Language Change: Traces of AI-Associated Language in Unscripted Spoken English Words associated with AI turning up in the way people speak, off the cuff. An association, not proof of a cause.
- How LLMs Distort Our Written Language In one of its experiments, essays largely written by the model took fewer sides than essays written without it.
Can literary quality be measured?
- Good Reads and Easy Novels: Readability and Literary Quality in a Corpus of US-published Fiction Readability goes with popularity, and against prestigious literary awards: crowds and experts judge differently.
- Measuring Literary Quality: Proxies and Perspectives Fourteen measures across 9,000 novels, and the case for speaking of qualities in the plural.
Grey
- Spiegel, Grau (Mirror Painting, Grey), 1991 The painting on the title slide: grey pigment on glass, a mirror painting from the 1990s rather than one of the grey canvases of the 1970s.
Every number in the talk, with its source, is on the last two slides of the deck.
About the speaker
Tommie Juzek is Assistant Professor of Computational Linguistics at Florida State University. He works on why language models write the way they do, and what that is doing to human writing. More at tjuzek.com.