The latest

Allison Stanger draws a firm boundary around AI-assisted writing: the writer owns the argument, structure, memories and ideas, while the machine serves as critic, research aid and mechanical assistant. Drawing on her use of AI since ChatGPT’s 2022 release, she argues that the technology can deepen thinking when it exposes weak evidence, neglected objections and flawed transitions without taking over the choices that give prose its meaning and individual voice.

Details

• Human territory: Stanger rejects outsourcing the most important elements of a piece because writing helps people discover who they are and what they value. She argues that experiences and personal judgments cannot be delegated without diminishing the exercise itself, comparing that loss to a father using a chatbot to draft his daughter’s wedding toast.
• Adversarial reader: Her preferred use of AI is as a fast, tireless and unflinching critic. She asks systems to attack drafts, defend opposing positions and identify weak transitions. She then decides which objections are valid and revises accordingly, treating difficult sentences as signs that the underlying thinking remains unfinished.
• Citation checks: Stanger treats factual claims supplied by AI with suspicion, particularly quotations that fit an argument perfectly but appear in rarely cited sources. For her forthcoming book, she used two systems to review 566 endnotes: one flagged doubtful citations, another assessed the alerts, and she adjudicated every case against archived sources and original documents.
• Revealing failures: The systems failed in different ways. When one invented author names, those names moved toward the demographic center of its training data, displacing female and nonwhite thinkers who had written the actual sources. On another occasion, both systems denied an event Stanger had personally experienced; her archive resolved the dispute.
• Mechanical delegation: Formatting, finding repetition, smoothing transitions and diagnosing a chapter that lacks coherence are tasks Stanger readily assigns to AI. Automating such work allows writers to remain immersed in composition for longer, while directing more time toward interpretation, judgment and other work that only the author can perform.
• Voice protection: Writers must also resist familiar machine habits, including excessive em dashes, flattery and repeated antithetical constructions. For Stanger, the deeper limitation is experiential: a machine has never lived anywhere or been anyone, so it cannot reproduce the trace of an individual consciousness making distinct choices.

Background

Stanger compares AI’s potential influence on writing with computers’ impact on chess. After machines surpassed grandmasters three decades ago, humans continued playing and began training with them. She describes the generation raised alongside those systems as the strongest field of human chess players in history.

What’s next

Stanger’s book, “Who Elected Big Tech? How Your Phone Hijacked the Constitution,” is due in March. She is a distinguished endowed professor at Middlebury College.