My mum grew up in Manchester and, as a result, had a whole treasure trove of sayings that she would apply to a wide range of situations. Some of them made little sense at all. For example when something was deemed to be nonsense, she would say ‘it’s all my eye and Betty Martin’ - a phrase that today, we’d probably flag as an AI hallucination. One of her favourites, when things were too much, over the top or unnecessarily spoiling an outcome or an event, was ‘well they’re just over-egging the pudding’.
And that is exactly what’s happened in Substack with the imposition of Pangram.
As those of you familiar with Substack may be aware, Pangram has been introduced to flag and highlight AI-generated text. If somebody is using AI to write their pieces for Substack, the idea is that Pangram will flag it, and people will know that it’s been generated using AI.
Whilst Pangram is constantly singing its own praises in terms of its capability, our lived reality and experience is that AI is not that accurate when it comes to flagging AI-generated pieces, partly because it’s entirely trained on human-generated content. So, of course, there are going to be linguistic similarities and patterns in the content just as there will be differentiators.
We also know, in the scheme of things, that many people use AI to generate material that, if it wasn’t for AI, they wouldn’t be able to produce. For some, that means they now have a voice they wouldn’t otherwise have had because they’re able to express their thoughts and their perspectives on what’s going on in the world with the support of the technology. They are able to tell their stories or share something they feel will be helpful to others but their writing ability has prevented them from doing so. In that instance, it’s a really helpful tool and nobody should be penalised for using AI in that way.
We also know that people generate any old nonsense, any old slop, to fill pages, get followers and monetise their content, just because they can. But again, it’s an opportunity for them to use a tool.
We have innumerable bad actors motivated by malintent, creating slopaganda that slurps and swishes its way around the web, causing all sorts of harm. And yes, that should be spotted and called out.
But to impose an artificial policing system inside a platform that has also sung its own praises in terms of creativity, freedom of thought, lack of algorithmic bias and so on, seems, at minimum, counterproductive and, at maximum, the fastest way to trash trust and engagement on said platform.
Now, there are lots of people — call them purists or diehards or old school — who feel that there’s no place anywhere for AI-generated content. But the truth of the matter is people have been using AI to generate content for many years. Newsrooms were using AI tools to generate copy at least a decade ago and we’ve had content automated across all channels for a very long time. That’s situation will only change when they switch the power off.
In the same way that tools similar to Grammarly have rewritten content for people in the past, so too the more sophisticated LLMs do it today. And the people programming their agents to write any old stuff, well, they’ve got the means and the platform to do it and so they will.
But the real problem with letting Pangram loose to judge, brand and mark everybody’s work with the sign of the robot is that human voices are the ones that will suffer. Human content will be penalised (by virtue of the fact that all of the models are trained on people’s words, experience and creativity) as it inevitably will have the traits and markers now highlighted as AI tells.
Rather than weed out machine-generated material, what will happen is that writers working without automation will be marginalised, marked down and made to stand in the Substack corner, where they won’t deserve to be.
I would be surprised if there’s a writer anywhere who doesn’t use a spellchecker or who doesn’t make use of the automation for correction within Substack itself. If you’re writing from the heart, if you’re speaking from the heart, if you are dictating from the heart, if you are feeding ideas into the machine and asking it to structure them for you so that others will understand, that is today’s way of doing things.
And humans can be relatively discerning, depending on the platform that they are engaging with. We can spot blather and we can spot automation - probably better than AI in some instances - and if we have doubts, we can ask or we can challenge.
Ghostwriters have been a feature of literature for centuries and what we have now at our fingertips, or within reach of our voice, is the ghostwriter in the machine. That’s the situation that we either accept or switch off.
What we don’t need, though, as humans, is a badge to say, “I’m a human.”
I know I’m a human, proudly so. And I use a variety of tools to help me do all sorts of tasks, be that an electric whisk when I’m baking or an oven to help me cook faster than an open fire.
There are countless ways that technology supports the work that we do. That doesn’t make our work any less valid or any less creative, nor does it make it appropriate for it to be dismissed simply because another (fallible) AI has passed judgment on it.
And that’s where I have a problem.
Substack has been a pleasure for me for quite some time. I had an early dabble when it was first launched, intermittent offerings as time went on and, in the last few years, I’ve used it for more regular thinking out loud. One of the things I’ve appreciated most about the platform has been the freedom that it gave me and the lack of interference.
But now it appears that those who run the platform are going the way of the other big tech networks and throttling human voices with tech, because (ostensibly) they want us to verify our very humanity. Whether it’s a tick on X, a check on Facebook or an authentication on LinkedIn - all of which can be fudged or fraudulent - none of them makes me any more or less human than I already am.
The algorithm shouldn’t be allowed to dictate what I say, how I say it, what I use or how I use it. And if humans reading or engaging with any work feel that it’s beneath them, worthy of critique or they simply don’t approve of robots, then that perspective is for them to hold dear - they can scroll by or directly engage with the creator.
AI shouldn’t be used to diminish the quality of a human work. If you are somebody who doesn’t use AI in any shape or form, good on you. That’s fantastic. But for those for whom the written word is a challenge, AI gives them an opportunity to engage. That’s just fine too.
And yes, I absolutely understand the problem of mis-, dis- and malinformation - probably more than most. It’s something I’ve written about, spoken about and advised on extensively for many years as part of my day job. And yes, AI is both helpful and harmful when it comes to that particular environment. This, however, isn’t that conversation. This one is for the writers choking on their word salads.
I also understand that the use of tools like Pangram within Substack is, potentially, an attempt to manage misinformation. But it’s the wrong tool in the wrong place.
I think Substack, as my mum would say, has over-egged the pudding, and, along with all the other networks, continues to do so.
The upshot is that instead of a smorgasbord of thoughts, opinions and ideas, when everything is being judged through automation, we end up not just with slop, but with curdled slop.
I’ve written about the challenge that we face as writers when it comes to AI, here and elsewhere. For years I’ve been running sessions including a couple of my favourites, Storytelling for Humans and Beat the Bot. Beat the Bot pits human writers against AI. Not so much to see whether we can tell the difference but to highlight the ways that sometimes AI presents a far more human story than the humans themselves.
Switch your attention to some of the corporate-speak nonsense that we hear, the business language bingo that’s thrown at us, the political parade of evasion and obstruction that peppers speeches every time politicians are on view - all examples of automated, often inhumane, responses created by human hands, not AI. There’s a balance to be found with all this and the insistence on marking our work with AI is not going to help anybody.
It will neither stem the flow of slop nor will it encourage those voices who would otherwise never attempt to share their stories with us.
And the imposition of automated silence is of far greater consequence than regurgitated revelations from a robot or two.
And the imposition of automated silence is of far greater consequence than regurgitated revelations from a robot or two.