A few weeks ago, I got a private message from someone I didn’t recognize on Discord. They’re a member of the Western Mass Writer’s Guild Discord server. I don’t normally chat with random people on the Internet but I thought I’d see if they were a local writer or what. We exchanged a few messages. They said they were an artist and asked if I wanted to look at their portfolio. They had a dozen or more pieces of digital art where you could see both the original line drawings and the finished, full-color artworks.
I’m not sure why they messaged me. I think they may have seen the posts about starting Elegant Lich and hoped that maybe we would be in the market to purchase artwork. They never really said.
We exchanged a few dozen messages during the following weeks. I made it clear early on that I wasn’t in the market to purchase artwork. But, during the conversation, I also was trying to suss out if they were a human being or a bot.
I asked how they were. “Sad” because they didn’t have any commissions. I asked what they were going, “Nothing much. Trying to get some work.” I made a point of sharing public aspects of my life: the same things I was posting to social media: hanging out with Tanuki, getting a flat tire fixed, getting vaccinated, etc. They said nothing to suggest they had any kind of human life.
Finally, I admitted that I was trying to tell if they were a human being. They haven’t contacted me since.
This was all before SFWA ran their panel discussion today: I Swear I’m Not a Robot — How Publishers Are Handling the Pressure to Confirm Human Authorship. It was very helpful and I took careful notes. The panel included a number of editors that are struggling with the problem of identifying AI-generated manuscripts.
They pointed out that the problem is not just manuscripts. The entire ecosystem is under attack. It includes reviews and artwork and correspondence. “In every space there might be writing, people are using these tools to conduct scams” — M.L. Clark.
All of the editors reported a significant amount of trauma over the experience of having accepted, nearly published, or published AI generated manuscripts — and only learning they’d been scammed through reports from readers. It’s a struggle to extend and maintain trust — especially with new authors who most need to find a receptive, supportive community. What is the community’s role? What is the responsibility of the staff? Where does the buck stop? What do we do? “Dealing with not being able to know is the challenge” — Michael J. Deluca.
There is a substantial debate about the potential role of using automated tools to identify AI-generated manuscripts. Some of the tools seem quite good right now. But it’s an arms race between the AIs and the tools. When do you use them? As a first line of defense, to protect your slush readers? Do you really trust the machine to make these decisions? There’s not a clear answer.
One editor said they might engage the author in a series of questions about their writing process, e.g. “Why did you use this PoV?” I thought, “Oh, no. I’m fucked.” I don’t know why I write the way I do. I’ve never taken any writing classes so I don’t even have the language to talk about why I write the way I do. But other people were careful to point out that the burden of proving they were human should not be on the author but on the editor. Still, retaining artifacts of the writing process, like outlines, rough drafts, revision histories, could all be useful to help demonstrate human authorship.
The moderator commented that Yog’s Law should still hold: Authors should certainly not have to pay to prove that they’re human.
There were a few encouraging points. First, Gen-AI is not good at generating long manuscripts. It can write good flash fiction but, the longer the manuscript, the less likely the AI is to be able to develop and maintain a complex narrative. Many of the people using AI are fundamentally lazy: simple checks can catch many of them. Also, current models are being funded by venture capital. As the true costs shift to the users, it may well not be economically viable to generate AI for the meagre pittance that short fiction writers earn, and discourage scammers from pursuing the “market.”
In the end, the recommendations were for editors to educate themselves, talk to each other, and build stronger networks of human connection. All good recommendations.
