Not by AI

Some time ago I came across a Mastodon post by @stammy@stammy.design linking to a website proposing “Not by AI badges”. The idea is to offer a preset badge (or rather 3 such presets) to “certify that your work (writing, painting or audio/visual) has been produced without (or rather with “less than 10%”) AI (or rather, as proposed by Stefano Quintarelli, SALAMI).

The idea piqued my interest, but I didn't have time to look into it when I first came across the site, so I just bookmarked it, leaving it for a time where cold look into how to add such a badge here. When I finally had time to look into it, I was underwhelmed by what it turned out to be.

The first thing that surprised me was the 10% threshold. If it's “not by AI”, I would expect the threshold to be either 0% (no AI involved at all) or 50% (majority of the work done a human). 10% sounds pretty arbitrary, and it's weakly motivated on the website:

The 90% can include using AI for inspiration purposes or to look for grammatical errors and typos.

Second, there are some very stringent rules about the badge use beyond the 90% criterion (placement, size, no modifications). I can guess some of the reasons why they would want to exert such control on these parameters, but the fact that no licensing or copyright information seems to be available on the website or the download package makes the use of the badge troublesome.

Third, still on legal, the badge holds no actual value, and at the time of writing the site actually is still asking for legal expert to get in contact with the organizers (who are they?) to

explore the potential of formalizing and regulating the use of the Not By AI badge.

Fourth, as if it wasn't enough that nothing is said about who's behind the initiative, they seem to have no Fediverse presence. And yes, this is important: especially when combined with the restrictive terms of use of the badges, it's an indication of a substantial lack of interest (at best) or antagonism (at worst) towards free culture and humane tech, which —given the alleged mission behind the badge— is a significant failure.

So, this «Not by AI» badge has no legal, nor technical (as remarked in a comment to my original thread) merit. What's the use, then?

The most obvious reason to apply the badge seems to be for clout: «proudly show that you're still doing things the human way!»

But I think there's a much more sinister reason behind it, that emerges when you think about cui prodest, who benefits from this badge existing and being so clearly recognizable?

The answer is that the main objective is to tag human-generated content to help scrapers identify non-AI content. Why is this important? Because as the Internet gets flooded by low quality machine-generated content, it will be more and more difficult for scrapers to find actually useful content to train their AI on (see also this minithread by @jetjocko@mastodon.social).

And what's the best way to achieve that? Ask humans to tag the contents themselves! (Ironically, this need is highlighted on the “Not by AI” website itself, although it's spun under a pretense of “helping humanity move forward”, rather than as a way to feed sufficient noise into SALAMI models to make them produce more varied content.)

In other words, my suspicion is that by using the badge you'd just be setting yourself up for maximum targeting by SALAMI training bots: you'd be certifying to scrapers that your creative output can be used to train their models safely.

Don't use the badge

I'm not saying that it's not important to advertise when one's art has been produced without AI: there absolutely is merit to it, especially when looking at the implications of using a tool built on nonconsensual exploitation of the creativity of others. But if the intent is to communicate human-to-human, there is no benefit in automating the signaling of human effort and making it machine-processable —in fact, it's going to be counter-productive in the long term.

I don't dislike the idea of the badge. In fact, I may even be working on rolling my own. And it'll be a hand-rolled multi-lingual SVG, like I do. And it will be released on a CC-BY-SA license, and people will be encouraged to modify it, customizing it for their own purposes.

On the other hand, everybody using a derivative of my hypothetical badge would still be “detectable” by AI scrapers, so maybe that's not such a hot idea …

Meanwhile, consider looking into the brilliant alternative by @XanIndigo@writing.exchange, an “AI data usage statement” that reads:

By using any of my writing, including social media posts or other communications, as training data for any AI writing tool, you are hereby agreeing to be legally bound to pay me for any and all text used, at my special AI premium rate of €5.50/word.

All payments are required in full within seven days. Late payments or use of any of my writing without disclosure may incur an additional penalty fee of up to €18/word plus full payment of any necessary legal fees.

I doubt it would be legally enforceable —which is, to be honest, quite a pity— but it undoubtedly works both as testament of the humanity of the work it refers to, and to show a clear intent in the relationship between this and the possibility for SALAMI to train on it —something which is completely omitted from the “Not by AI” badges.

And it's really time we started to be clear and loud on something much more important than our work being “Not by AI”: the fact that our work is “Not for AI” either.

By humans, for humans.