Why I feel icky about the Substack/Pangram partnership

And why you should too.

If I asked you to write like AI…do you think you would be able to do it?

Absolutely! Let’s dive in…it’s the three S’: saccharine, sycophantic and samey. In fact, it’s got those too - rules of three. And it’s not just boring; it manages to take all of the good techniques we know from rhetoric and turns them into blancmange, just like that contrastive construction I just dropped.

Humans are incredible artists of mimicry; we learn to imitate birdsong, we construct entire comedy shows around impressionists, and guess what? We can, if we want to, even sound like AI - and we might even be doing it without realising it.

It is one of many reasons why the Pangram partnership that has just been launched on Substack is daft at best, damaging at worst.

Some people might sound like AI without even meaning to

Earlier this year, researchers at the Max-Planck Institute for Human Development in Germany analysed 280,000 English language videos of presentations, talks and speeches from more than 20,000 YouTube channels from academic institutions. From their findings, they claimed the first empirical evidence that humans are increasingly imitating LLMs in spoken language. Not written language - where we’ve seen LLM-speak long seep into everything from news copy to academic literature and social media content. Spoken language.

I think there’s an obvious caveat here that perhaps there are lecturers who are reading off of notes that were written language generated from ChatGPT or Claude, but the researchers do point out that the word “delve” for example occurred both in the reading of prepared manuscripts and “in spontaneous conversations”.

Can you guarantee that your mind palace, and all the language it withholds, will go untouched by all of the Reels and LinkedIn posts being relentlessly hurled at you that are full of AI language? Including posts where creators have not been transparent about their own AI usage? You can’t - but Substack seems to think you can. And what is the science that it’s based off? Because…

Pangram is still generating false positives

It’s widely touted as having a “near zero” false positive rate. Yes, it does seem to be one of the best AI detection companies out there. But is any false positivity rate, however low, good enough for a platform that relies on authors - most of them writing for free - to harness? Has Substack effectively decided it’s worth sacrificing the small percentage of innocent people it’ll blame for using AI for the greater good of thwarting the AI slop it worries will alienate users? There are additionally legitimate worries about the use of AI detection tools wrongly accusing non-native English speakers as well as people with neurodivergence.

As CNET reported, Substack authors who feel like they’ve been wrongly accused can report the error to Substack and have the false scan remove. But what damage may be wrought in the meantime, and who in their right mind wants to scan their own hard-written copy every single time they publish a newsletter for whether they’ve used AI, when they haven’t used AI? Not me. Which brings me onto my final point…

Is AI the problem - or growth incentives?

Discoverability on all social media platforms is increasingly predicated on recommendation algorithms, content and social graph networks, and the need to show up as regularly as possible on said algorithms and networks to get your content seen in the first place.

Users are additionally expected to do all of that for nothing (translation: provide Big Tech with free content for Big Tech owners to profit from) until they qualify for monetisation, whether that’s from platform ad revenue or from subscriptions. None of that is sustainable, and the massive adoption of AI to write social media posts is born from the very monster of engagement-bait-mageddon social media platforms have generated in the first place.

Employing Pangram’s tool is supposed to look like care. For me, it feels like a distraction from the real problem.

---

Anyway, I’ve written this not only to have a rant with my morning coffee this morning, but to publish a newsletter I can actually test the new Pangram tool on. You can do it too. And then we’ll all head to Notes, where you can tell me what you think, and we’ll all hopefully move on with our algorithmically powered, and inevitably AI surrounded, day.