AI Policy

Generative AI tools, or large language models (LLMs), can be used to generate “original” content – although they’re not actually thinking/writing, but instead just predicting what the next word should be (like the auto-complete options in your phone keyboard, but trained on much larger data sets of content).

In contrast, Traditional AI tools perform more limited, predetermined tasks, like spell checkers and search engine algorithms, but do not create “new” content.

Aside from moral/ethical and environmental reasons for preferring not to use Gen AI, many writers and editors have found that its capabilities are still quite limited and often problematic. While it might be helpful for writers to use for certain tasks, like brainstorming or organizing ideas “behind the scenes” of your writing, I do not recommend its use for any public-facing writing for several reasons, including:

• AI-generated writing flattens out all uniqueness and personality (“voice”).
  • A unique author “voice” is important, e.g., for maintaining reader interest, even in most formal academic or non-fiction writing.
• AI-generated writing relies on repetitive, simplified styles and sentence structures.
  • Readers are likely to experience this as boring or even distracting or irritating.
• AI-generated content often skips over logical steps, creating gaps or discontinuities.
  • This can create a mess for the writer (and editor or reader) without careful untangling, which means it’s often better to start from scratch and write an explanation without AI.
• AI regularly “hallucinates” and presents misinformation.
  • This means, if you use AI for research, everything it generates needs to be carefully checked: Not just “Does that source actually make that claim/finding?” but also: “Does that source even exist?”
    • Free chatbots are especially unreliable, but even with a paid chatbot, you would still need to check all sources yourself.
• AI can corrupt your documents when you delegate a task to it, introducing errors and causing major formatting issues in your documents.
  • This means that you should never allow AI access to the only version/copy of your draft.
• AI-generated text is likely ineligible for copyright in the US, and has been restricted or banned by some specific academic journals or publishers.
• AI bias (which of course reflects human bias, since Gen AI tools were trained on human-generated writing) can not only cause harm, but can also shift people’s attitudes – even people who are on the lookout for AI bias.

Last, the act of writing itself (including brainstorming, drafting, rewriting, revising, polishing, etc.) is a form of critical thinking (see, e.g., Bean & Melzer, Engaging Ideas, Chap. 2), and is a non-linear, messy process through which writers come to think through and discover what they want to say and how they want to say it. This process is fundamentally missing from AI (see “Generating vs. Writing” in this post), which is one thing that sometimes makes it easy for me to differentiate whether I’m reading AI- vs. human-generated writing.