𖠫 From bench to bot: Does AI really make you a more efficient writer? - Requarth - 2024
when it comes to more complex writing tasks, such as a “Specific Aims” page or a delicate email, will using AI actually save you time? Or will all of the requisite prompt-crafting, factchecking and manual editing only serve to slow you down? (From bench to bot Does AI really make you a more efficient writer - Requarth - 2024, p. 2)
In other words, are we rationalizing AI technology? Requarth goes on to review several studies that effectively assess time to write a document vs. time to write prompts and edit the results for the same document. The results are mixed: there’re benefits for undergrads and business applications, but not for MDs replying to clinical messages. However, the MDs seemed to find the AI-supported task easier, cognitively, even if it took the same amount of time — which could be a major gain of its own accord. Moreover the quality of the outputs was better than control for both students and business contexts, and while AI outputs contained hallucinations and inaccuracies for the MDs, the AI-aided messages tended to be kinder, a benefit appreciated by the recipients.
Requarth concludes with a caution: overreliance on AI, and particularly using it continuously instead of at discrete steps in your workflows, might make AI less of a tool and more of a coauthor. We are inevitably influenced by the ideas AI introduces into our writing, and this can subtly change how we think as well as how we write.
Be transparent about AI use, especially in academic writing, and follow any guidelines or policies set by your institution or publication venues. (From bench to bot Does AI really make you a more efficient writer - Requarth - 2024, p. 6)
Requarth includes the above as a strategy recommendation for AI use. Yet it’s not clear yet how to report on AI use in many media. The email-writing example Requarth provides is a great demonstration of this: if you use AI to make an email sound nicer, do you disclaim it in a footnote? Certainly a transparency statement like “ChatGPT was used to remove all my bitter vindictiveness from this message so that you would be more likely to cooperate with me in the future” would lessen the empathic benefits of the AI-edited email.
It’s even less clear how audiences will value this kind of admission over time. I cannot imagine anyone in the next few years valuing a piece of writing more if it has been AI-assisted. Meanwhile, even if it’s the right thing to do, writers will probably omit admission of AI use if audiences see it as dirty, leading to the bootleg future I described in How might generative text tools be useful in academia?