Week 10: Co-writing — authorship with a predictive collaborator

What happens to the idea of authorship once the sentence you're writing is being continued, word by word, by something that has read more than you have

3 May 2027

A writer using a predictive tool — accepting some suggestions, rejecting others, occasionally taking a suggestion and rewriting it into something else entirely — produces a text that doesn’t have a clean answer to “who wrote this.” Not “the human, obviously,” because some sentences are the model’s own words, accepted verbatim. Not “the model,” because the model never decided to write a story about this particular subject, in this particular voice, and would have produced something else entirely for a different prompt. The honest answer is somewhere in the editing history, which almost never gets published alongside the finished piece.

This week takes co-writing tools seriously as a genuinely new position for a predictive system to occupy — not correcting a typo, not finishing a search query, but sitting inside the compositional process of something meant to be read as a considered piece of writing. That raises questions the earlier weeks didn’t need to ask: does accepting a good suggestion count as writing, in the sense the word usually carries? Does it matter, if the finished piece is good? The lecture doesn’t resolve this — nobody has — but it gives you enough vocabulary to have the argument precisely instead of vaguely.

Outline#

• three co-writing tools and how differently each one inserts itself into the writing process
• acceptance rate as a measure of who’s actually doing the writing
• the disclosure question: does a reader have a right to know which sentences were suggested
• what the lab’s paired-exchange exercise asks you to notice about your own editing choices