Anthropic turned on invisible watermarks for Claude this week. This global rollout has been triggered by Article 50 of the EU AI Act. As you’d expect, it’s a hefty old document, but the upshot is pretty simple – just tell people when they’re looking at AI.
My sense is that reactions are broadly split into two camps. On one side, we’ve got prolific users who might be feeling a little exposed, while critics are feeling quite smug that AI-content is now carrying a figurative red flag.
We’ve had assistive tools for decades without an eyelid being batted
Spell-checkers and autocorrection tools keep our grammar in check. Satellite navigation has killed the skill of route planning. The calculator and the spreadsheet have been transformative for arithmetic. Google Translate helps us navigate the social aspect of trips to far flung destinations. Photoshop enables creatives to finesse their imagery. Autotune makes utterly dreadful vocalists only a little bit dreadful. None of these innovations have elicited the same moral panic as when someone has the temerity to use Claude in the process of writing an article.
I’ve tried to understand why the double standard exists, and think it comes down to what part of the task gets contracted out, and to what extent. I think AI usage for writing is a spectrum. Clearly, publishing AI slop with the complete absence of human editorial judgement and then passing it off as your own work is a no-no. But what’s wrong with writing your own draft and asking an LLM to check the flow and logic, while maintaining your tone of voice and (beautifully flawed) writing style?
The double standard really comes down to credit
There are already a shedload of places where AI use is normalised with no shaming in sight. I use it daily for research synthesis, meeting transcription and fleshing out ideas that initially only exist as an illegible scribble on the envelope of an important letter I still haven’t opened. I think these are accepted use cases because they fall into the instrumental bucket.
By contrast, human anxiety tends to spike when AI is used for anything where a byline exists. The not-so-quiet accusation associated with “gotcha, you used AI” is really that “you’re claiming authorship for something you didn’t earn.” It’s all a bit unseemly and unnecessary in my opinion.
For as long as I can remember, across every type of medium, there’s always been good and crap content. You only have to pay a visit to X to see that. What’s the point in getting agitated, even if it is proliferating? You just need to use your judgement when it comes to filtering what’s good for you.
This is where I argue with myself
I was tempted to call it snobbery, but then I thought of impacted friends, especially within the creative community. Illustrators, designers, copywriters and translators are being squeezed with the rise of AI. We’re talking about livelihoods here, and I don’t take that lightly.
There are also very reasonable arguments to be heard when it comes to copyright grievances in respect to how models have been trained, while a feed full of synthetic 'thought pieces' is a bit rubbish.
Individuals can decide whether they use AI and to what degree. Some will point to the acceleration it provides, while also explaining that independently formulating their thinking cogently isn’t a strong suit. Others may counter and choose to develop content with no robotic intervention, because there’s an art to doing so.
Irrespective, I think the key here is disclosure. If you’ve used AI, make sure you label it accordingly. The audience can then make a judgement call on whether it resonated with their circumstances and was both engaging and persuasive.
The secret is the only shameful bit
This is really important to remember in amongst the hullaballoo. If you read the statement carefully, Anthropic’s watermark isn’t a verdict. It doesn’t mean that the user shamelessly lifted an entire white paper from its platform and passed it off as original work. It’s just a signal that, somewhere along the way, AI was used.
There’s also been some speculation that watermarks from big tech companies could impact search outcomes. I’d encourage people to wait and see on this. GEO is a movable feast, so there’s no way of definitively knowing what the impact will be, although close monitoring in the months ahead is advisable.
In the spirit of transparency, I used Claude to help me with this article. I put a pretty detailed stream of consciousness into the system. I asked it to provide me with a plan for structuring the piece and then went away and wrote it based on that. I then asked Artlist to produce a tongue-in-cheek piece of artwork to accompany it.
What’s wrong with that?
