The TsunamAI: are our systems ready for the coming flood of vexatious AI?
This piece first appeared in my newsletter, Future Proof. Subscribe!
I’m not entirely sure why the word ‘slop’ became synonymous with the current wave of AI, but I’m sure that etymologists (now, or in the distant future) will produce fascinating research on this subject. Slop: it connotes both the bland, formlessness of AI-content (art that looks, simultaneously, like all art and no art) and also has the Dickensian sense of being serve great lashings of gruel, almost without consent. “Do I understand that he asked for more, after he had eaten the supper allotted by the dietary?” Mr Limbkins inquires, incredulously, after Oliver Twist demands his extra slop. “That boy will be hung!”
It’s an apt word, but I have been thinking of late about another a different descriptor for AI content: the TsunamAI.
I spent some of this week working (in my 09:00–17:00 capacity) as a producer on a podcast looking at the Employee Rights Bill, which should shortly become law in the UK. It is a sprawling piece of legislation covering everything from Day One rights to ‘fire and rehire’ practices, and, while interesting, is not the subject of this newsletter. On the episode we made, a leading employment lawyer — David Reade KC — made the following observation. “There are loads of sites now where AI will write a grievance or draft an ET1 claim,” he said on the show. “I feel a lot of sympathy for the employment tribunal because they are typically getting litigants in person and the first task they have to do at an early stage in litigation is unpick what the legal issues are.”
Essentially, David was making the point that employment tribunals are now swamped by verbose applications that have been created by AI, from litigants who might otherwise not have had the time or resources to make a claim. This is good, on one level — we want people to not be stifled in making unfair dismissal claims — but bad, in another way. There is already a huge backlog in terms of dealing with these claims, and it’s quite clear that all parties are advantaged by a quick resolution. You can use AI to create claims, and while you might (eventually) be able to use AI to triage claims, the final decision in cases like this has to be made by a human. And that means the system is getting gummed up.
It reminded me of a piece I read at the weekend about how AI was being used by NIMBYs (Not-In-My-Backyarders, for those less terminally online) to grind the planning system to a halt. The Guardian noted a new service, called Objector (I won’t link to it, as it makes me queasy), which uses AI to scan planning applications and find the highest impact objection. The service was designed by a Kent couple who, according to The Guardian, designed the system “after estimating they spent hundreds of hours attempting to navigate the planning process when they opposed plans to convert a building near their home into a mosque.” The planning system here in the UK is already in disarray, with housebuilding goals routinely stymied by local protests and infrastructure projects disrupted by fairly low-grade concerns (just Google “HS2 bat tunnel” if you want to understand more about sclerotic Britain). Slowing the planning process further makes it even less desirable to constructors and investors, yet AI has been afforded the power to flood the planning system with vexatious objections that, ultimately, have to be analysed by a human. At the moment, the rules are simple: all objections have to go through the same processes, regardless of whether they’re created by humans or AI.
Over in the job market, the Financial Times provided some information on the use of LLMs in job applications, something that is now ubiquitous. The data provided a fairly predictable conclusion: for weak candidates, using ChatGPT to write a cover letter offered a significant improvement to employment chances. For stronger candidates, it decreased prospects.
What the data doesn’t quite unfurl is how using AI improves employment prospects via increasing the total applications. I haven’t written a job application is several years — thankfully — but I remember that when I was doing them, writing bespoke cover letters, trying to research the company and the role, all that took a considerable amount of time and effort. (And I was pretty slapdash, other people that I know would spend several days on each application, a higher investment with a better yield). The amount of time required on each application made me more judicious about which roles I would apply for — would I have taken the same cautious approach if I could spaff out 100 applications a day? And for a strong candidate, even a 20% reduction in employability is reasonably offset by a 10x increase in application output. Would I bother to write one scintillatingly brilliant application, if I could just spray out 10 AI-assisted applications (that only reduced my prospects by 20% a piece)?
As someone who occasionally hires people, it’s been clear for over a year that a large percentage of applications are coming via identikit AI applications. If you ask ChatGPT to write a convincing cover letter for a UK-based podcast company, you are going to end up with an application eerily reminiscent of other people asking ChatGPT to write a convincing cover letter for a UK-based podcast company. It makes human-inspired writing a big differentiator. But I am a small employer, hiring entry-level roles, rather than a big company dealing with hundreds or thousands of applications a day. At that point, the tsunamAI has a much bigger impact on hiring processes.
In a way, AI slop is easy to deal with. “That’s ugly,” you say, and you move on. You keep scrolling on TikTok, past AI-generated videos of small boats being blown-up, or, better still, turn the app off. You choose to read articles that have been written by people with families and mortgages, rather than Mr Alexander Indigo (and if you can’t tell the difference: ah well, don’t worry about it). Slop is just bad product, just as low-grade sweatshop produced fabric is bad product and will fill landfills.
But the tsunamAI — the swamp — is both bad product and an assault on our systems. It is like if AI slop — the picture your uncle has prompt engineered of a tree, with each of his children and grandchildren’s faces as a rosy red apple, pithily titled ‘The Family Tree’ — covered 99% of the wall space in the Louvre, so that a visit was a three-hour hunt for the Mona Lisa. In the UK, we are at least 10 years away from having a bureaucracy sufficiently nimble to deal with a 100x increase in paperwork — and by then, undoubtedly, everything will be different. This is a collision of new tools and old conventions.
It’s not just employment tribunals and planning objections and job applications that will be gummed up by the tsunamAI. Divorce courts, parole boards, the tax system, PTA meetings, TripAdvisor reviews: throw a stone and you’ll hit a pillar of our society that could be destroyed by submersion in a swamp of suffocating paperwork.
I am not a lawyer nor a legislator, but I suspect that some of these services will need to stipulate that requests be human authored. It will be hard for them, mostly, to ascertain whether something like a planning objection is written by a human or an AI (at least with the level of certainty required to dismiss a claim) but the simple appearance of that requirement (“Requests generated by AI will be automatically rejected”) will be enough to disincline many people. Of course, so many of these tools are now integrated into systems (one could hardly outlaw use of AI research tools in the courts) that the die is at least semi-cast. But we already have rules around ‘vexatious litigants’ here in the UK (“individuals who persistently take legal action against others in cases without any merit”) who lose their rights to make claims. Could those rules be used here? Or extended to include things like frivolous or impersonal planning objections?
To do so would probably require the roll out of a proper Digital ID, thus linking a planning objection to a centralised database and personal accountability, which opens a whole other can of worms. The tsunamAI is a huge can of worms, in its own right, but any solution seems like it would just be adding more worms to the mix. Can you fight worms with worms? Surely at some point, we have to have reached peak worm; the point at which this dying planet belongs to the worms, not the humans.
But if pick-up of AI-shortcuts by normal, regular folk continues at the current pace, the tsunamAI will have to be tackled. Our systems barely worked before; we should not expect them to function when submerged in the flood, lips pressed to the ceiling, gasping for the last remnants of air. Ultimately, our social fabric is constituted under the assumption that it should provide for all the people who are out there (in reality, of course, it doesn’t provided for anything like that universality). Then the wash of slop means that the fabric is now being stretched over not just the human population but the infinitely replicating digital agents, activists, lawyers, pundits and assistants who our creaking systems cannot differentiate from actual, living, breathing, suffering people.
