Context Window 88
It’s been a quiet week for AI in publishing, though the authenticity questions I wrote about last week are rumbling on. Thanks to Thad McIlroy for sharing his comprehensive analysis of the research behind it. But that aside, the freshest things this week were from different, adjacent industries such as journalism and film-making. Together they reveal something interesting: different creative industries are arriving at very different accommodations with AI.
You’ve probably seen the headlines this week about a new study which claims that most participants in a blind test preferred AI-generated to human-written stories. This has, naturally, fed into the end-of-days-for-creatives narrative. However, I think it’s important that anyone who cares about this looks a little deeper at the methodology.
The researchers asked readers to compare three short stories sourced from literary journals and anthologies, and three written by ChatGPT (if you’re as big a nerd as diligent as me you can read all of them here). On a basic level, those are different things. Good literature pushes at the edges. Competent prompting clusters around the mean. Should it really be so surprising that people would prefer the average? If you gave most people in a nationally representative sample a choice between Ella Langley and experimental jazz, they’d choose Texas.
Last week I wrote about differing attitudes to AI among fiction writers, where there is a more critical view, compared with non-fiction and scientific research where AI is used more openly for research and other tasks. Add journalism to the latter category: eight of this year’s Pulitzer Prize journalism awardees disclosed AI use to judges, and the whole thing looked like business as usual. As a thought experiment, imagine the reaction if a similar proportion of finalists for the literary categories had done so.
Of course, this doesn’t always map neatly to book publishing, particularly where it cannibalises a separate audiobook channel. But for journal articles, monographs and other areas where there simply isn’t an audiobook market, this looks like a compelling proposition: AI narration as a feature, especially for accessibility, not a substitute.
Independent publisher Fred Zimmerman has published a new research paper analysing coverage of AI in the international book trade press. It reaches some interesting conclusions: that coverage is framed around issues rather than capabilities, focuses more risk than opportunity, and lacks technical scrutiny. Fortunately there’s a newsletter for those things.
More than 10% of open jobs in Hollywood are connected to AI use, according to the LA Times. Job descriptions provide a great, leading indicator of where an industry is heading, and from a quick eyeball that proportion is not something I see in publishing job listings. One of the themes running through this week’s stories is that different creative industries are arriving at very different accommodations with AI. Journalism is using and disclosing it. Hollywood is hiring for it. Publishing is still debating authenticity and validity. The technology is broadly the same; the industry cultures surrounding it are not.
That’s not to dismiss authenticity. With a book coming out later this year and (at the time of writing) nearly 100,000 words of posts and newsletters on my website, I am getting asked more regularly about my own AI use. It seems to me that writers owe their readers clarity and transparency: obfuscation and lack of provenance trouble me more than AI use per se. Regular readers should not be surprised that I take an essentially pragmatic view on the usefulness of AI tools and the validity of AI-assisted content. So, with transparency in mind, I’ve drafted a policy setting out where I personally use AI in my work and writing. This is a snapshot: it is quite likely that the tools I use will evolve over time, though the underlying principles should be enduring. I’d welcome feedback from readers.
An editing gremlin last week led to the omission of a section on a new consultation by Publishers’ Licensing Services on introducing workplace AI usage rights into CLA Education licensing. An earlier, related change to Business and Public Sector licences has already resulted in revenue being shared with publishers, so this is a live commercial area as much as a policy discussion. If you are a PLS member publisher, you have until 11 September to respond.
Finally, something fun for the weekend: someone compiled a classification of AI companies on a 2x2 matrix based on high school cliques: preppy versus goth and jock versus nerd. I am reassured that most of the platforms I use are in the preppy-nerd quadrant.