351
Carl T. Bergstrom
Carl T. Bergstrom

Imagine being the author of the worst take on AI yet published by Nature.

When AI writes the words, who is the author?

www.nature.com

Thought leaders are often full of shit. Self-described "thought leaders" are invariably so. I hope the AI is ashamed of itself.

This is also quite embarrassing for Nature, given that their author policy requires "Declare AI use transparently. If an AI tool helped generate text, data analysis, or content, acknowledge it in the Introduction, Preface, or Acknowledgements." Why did this guy get a pass instead of a hard reject?

While we're at it, Dean Fielding might wish to read his institution's own position statement on AI use, or at least get ChatGPT to summarize it for him.

1 / 3
This opinion piece was made in the same way. It began with me speaking an untidy account of what had happened and what it made me think. AI did almost all of the writing. I questioned it, redirected it, removed what was too smooth, restored what mattered and decided what I could stand behind. The ideas are mine. The expertise is mine. The experience is mine. The responsibility is mine. That does not make AI’s contribution invisible, and it does not make the work of professional writers less valuable. AI did almost all of the writing. I remain the expert and the thought leader behind the piece. Does that also make me its author? I do not expect everyone to agree. But that is the debate I think we now need.
Then the handling editor wrote back. Two AI-detection tools had classified the piece as 100% AI-generated. She invited me, very politely, to rewrite it in my own words or, if the result was wrong, simply to confirm that I had written it. She did nothing wrong. She was careful, fair and open to the possibility that the tools were mistaken. Still, the message landed hard. In this case, however, the result touched something real. AI had generated almost all of the final prose. For a few days, the distinction between writing with AI and having AI write for me suddenly felt rather thin.
Research: Transparency is important to ensure that trust in the research endeavour is not eroded. The use of AI in any form must be declared during the scientific and ethics review processes (where applicable) and clearly explained in any research activities (e.g. research design, processes, protocols, reports, dissertations, theses, publications, conference/workshop/meeting presentations) and outputs (whether it is a partial or derivative of an output generated by AI), similar to what would be expected with the use of other software, tools or methodologies. Specifics regarding the AI systems or tools used, the extent of their use, and the nature of the contribution to research must be formally declared. Note that individual publishing houses of journals or authored books may have more specific requirements in their AI policies

Share this Page