AI isn’t Driving to the Average. It is Provably Making us Dumber.
The behavior we are alarmed by in the classroom is the behavior we’re celebrating in the boardroom.
Source: Adobe Stock.
Yeah, last week mostly didn’t happen – I got most of the writing done, but didn’t get this posted. So this is sort of two weeks in one.
Going back to my important questions to know about AI:
What does AI really cost?
What does the ROI actually look like, once we pay what it really costs?
If AI-driven shopping hollows out the customer relationship, what can retailers do to fight back?
What do consumers really want out of all of this?
Will AI-driven shopping turn into something manipulative and coercive, rather than just “removing friction”? (In other words, even if consumers want it, will it turn out to be bad anyway?)
This week is going to be about “what does AI really cost” – not the money, but (not to get overly dramatic) in the cost to our intellect and capacity to learn.
This is not a philosophical debate. Alarm bells should be ringing.
Retail Economic Indicators
First, the usual pit stop at retail economic indicators. The latest one worth paying attention to is the early read on Prime Day results, as shared by Numerator. It’s not exactly an apples-to-apples comparison, year over year, because Amazon moved it up and expanded it from 3 days to four.
But the results are definitely being framed as “consumers are beginning to crack – for real this time!”. Many of us have said that before, so I’m still waiting for July/August to take a stand on this one, especially with the sales period shift – moving it to the end of the month to get ahead of 4th of July may have moved it to a time when consumers had less money to spend.
But anyway, the results:
Average household spend $105, down 17% YoY
Average order $47, down 18% YoY
Average item $23.23, down from $24.59 (in the face of inflation this definitely points to shoppers seeking lower prices, not that the discounts were that much steeper)
69% of items purchased were under $20, just 3% over $100
Deal satisfaction 59%, down from 68%
>50% compared prices across retailers first - 49% also shopped Walmart Deals, 32% Target Circle Deal Days
Top sellers were Premier Protein, Liquid I.V., Temptations cat treats (not exactly splurge-y type items)
If you’re being kind, you might call these results “lackluster.” Keep this in mind when we get to Dan Frommer’s (The New Consumer) further analysis (below) of the Adobe AI results that I covered last week.
The Great AI-Driven Dumbing Down
In another piece I reference below, Dan Parris said something to the effect of I was worried that our future was going to look like Terminator and now I’m worried that it will look much more like Wall-E.
Two new pieces – one based on a survey of educators and one based on an actual study of students – point heavily to a Wall-E future. Put on your onesie, clutch your liquid diet in your hand, lean back in your hoverchair, and let’s see what they had to say while we can all still read.
I’ll start with Prof G Media’s Mia Silverio, who wrote about the survey of educators, plus some very distressing data about math and reading scores. If you want to know why educators are suddenly very keen to outright ban all screens in schools, all you have to do is look at the cliff-diving that students’ scores have done since ChatGPT became available (I used to wonder if Zuckerburg ever felt any twinge of conscience knowing that his site made it possible to revive and scale things like the ivory trade and the destruction of whole species of animals. Now I wonder if Sam Altman ever lies awake at night with the conscious thought that maybe he is responsible for kicking off the path to us becoming either Eloi or Morlocks).
Against an environment where math and reading scores for 9- and 13-year olds have been bombed back from a peak in the early 2010’s (before screens – phones especially – became ubiquitous) to a level not seen in 50 years, educators had something to say:
Detection tools for AI use don’t work, and AI companies seem completely unmotivated to fix that
The hardest-hit skills are writing, critical thinking, and creativity
80% of hiring managers say graduates are less work-force ready than previous generations
Key quote: “Offloading the struggle to a chatbot does not ‘free students up for higher-order work.’ It deprives them of building the strength to do any substantial cognitive work at all.” (Tyler Jagt, cited in the article).
But here’s why I wanted to write this essay, even though it’s not 100% focused on retail impact: We’re gearing up to fight this AI-driven decline in schools. But we’re actively encouraging people at work to dive in head first. More on that below.
The second piece is research that breaks down how this happens, a process the author calls “creeping cognitive displacement syndrome.” In stage one, you use AI as a drafting tool that you heavily edit. In stage two, you start putting in “unethical” prompts – not ones that violate terms of service (ha!), but ones that start to offload your thinking to generate arguments that you don’t evaluate. Heavily edit moves to cosmetic edits. In stage three, you have slid entirely down that slippery slope to full surrender of “cognitive sovereignty” and you ship what is becoming known as AI slop, or botshitting.
The AI helps you on this slide. It’s not like it defends an argument if you challenge it. The worst part is the use cases that were the most insidious in handing over your brain were actually rated the most empowering by the study participants.
The author concludes that the slide can be reversed “only if recognized early and resisted deliberately,” and requires “the humility to produce less and the courage to think worse, at least initially.”
What can I say? AI rots your brain. And that slope is oh-so-slippery. I am hyper-vigilant, and yet I also have been guilty of abdicating intellectual responsibility. I would offer a variant on the three stages. The first time, you’re in a hurry, you don’t have time to give something proper thought, so you just see what the AI comes up with, this once. And then you throw it out to other people – “here’s a first cut, but it was made with AI, so quality-check it.” And if no one catches anything, then you have positive reinforcement to try it again. And even if someone catches something, you say, “I’ll prompt better next time.” Or “I swear I’ll read it before sending it out next time.”
Total abdication on my part, because in addition to just relying on what AI says, now I’m also outsourcing even the quality review of the results to other people. And every time it works, the temptation is there to use it again.
Tell me you have never done the same.
The worst part is the loop. Not the loops that the AI bros are talking about, not even the reinforcement loop that I got away with it. I’m talking about the much more insidious infection that comes from absorbing all that botshit out in the world – everything from cultural norms to our personal word choices.
Daniel Parris at Stat Significant dove into that (should I say “delved”?). Citing multiple studies, Parris argues that we’re already seeing a convergence toward semantic sameness on the web, driven by use of AI. He looks at the huge boost in usage of signature vocabulary like “delve” and “bolster” and the use of em-dashes. I have to add “Let that sit for a moment” and “This is X dressed as Y”.
It’s not just the word choices, though, it’s the style. Nothing is controversial or hard-hitting (you’d have to actively intervene to make it that way). Everything is empathetic and sympathetic, “smoothing disagreement in sedation and sameness,” as Parris says. Thus the Wall-E future, and not the one hunkering in caves to escape the machines determined to kill us all.
This is an infection that is much harder to resist than even the cognitive abdication. Parris cites a study that finds that half of all web pages already show signs of LLM-generated copy. Statistically, then, half of all the emails and Substacks and Mediums and beehiv’s that I read have some LLM-generated content in them (actually, probably less, because I have selection bias towards people who are not all rah-rah AI).
The line I use above – “But here’s why I wanted to write this essay” – I rewrote it three times just to make sure that it didn’t sound like AI. I use that turn of phrase a lot, “here’s why”. I’m mad that AI is co-opting it to the point where I question my own (original!) use of it. But this is where we are.
Ironically, We Don’t Need AI to Make Us Dumber
While there seems to be a growing body of evidence that AI actually does make us dumber, we weren’t exactly waiting around for AI to make that happen. But when AI combines with our natural tendencies, you can get a compounding result.
John Cutler highlights this in The Beautiful Mess, where he takes to account the meme-like rush into talking about bottlenecks and either removing them or watching them shift. I talked about this a little bit last time but more in the context of AI driving ROI. If you cut actual code development down to almost nothing, but don’t really improve anything around it, you’re not going to get a huge ROI because coding is only 20-30% of the total development process.
Cutler builds on that concept. He points out that if engineering is your only bottleneck, then AI removing that bottleneck should terrify you, because you are only competing on development speed. And he highlights where business consultants (I can take pot shots here, I was one) come in and make it worse by jumping to stupid conclusions: “Imagine you have a 10% margin business. You spend 25% of your revenue on R&D. In comes that Deloitte person who says … ‘cut by 30%, spend 5% on tokens, book it all as AI productivity, and your EBITDA will magically expand, margins will look incredible, and you’ll somehow steal market share from other companies using the exact same technology…’”
I’ve already argued that assuming token spend maxes out at 5% is foolhardy. But even if the cost stays the same (it won’t), what will happen is, companies will do this, hollow out their organization, book the short-term gain as a sugar-hit win, and end up losing the long game because there is no foundation for durable value underneath. You fire everyone who knows things, you replace them with AI that sucks out the will to think critically from everyone left, and then you, I don’t know, ship code that lets anyone take over an account just by asking nicely, even highly protected celebrity accounts.
Sugar hit leads to sugar crash.
What Does This Mean For Retail? Two Possible Futures
Let’s start with the worst case scenario: brain rot.
You have to keep in mind that brain rot can impact anyone using AI – whether inside the retail company, or in the mind of the consumer. So you could get double-brain rot happening at the same time.
The scenario plays out like this. Retailers addict themselves to platform LLM’s. They generate sameness in content and tone, they relinquish their capacity to differentiate themselves, even if they’re brands, and they give up on any content that would be used as context because all it does is make the LLM smarter and better able to steal customer relationships.
In turn, consumers outsource their ability to choose products for themselves, and thus lose the shopping skill – the one that makes the hunt fun. All intent is mediated by LLM answer engines. So consumers rely on the AI to tell them what to buy, and retailers turn their spend from SEO to paying an AI for intent (this is the red-alert future that Google feared at the outset).
Maybe there’s an intermediate step in there where AI platforms pretend this is not inevitable and spend some time selling retailers “intent insights” so that everyone can feel good about the slow numbing death of their critical faculties.
As retailers stop investing in early-funnel content, AI steps in to generate it for them, and consumer demand collapses into “what AI told me I need to buy.” Consumers may or may not insert their own AI agents in there, to help them save money and not spend so much. Or they may rely on retailer-driven agents that are only there to drive spend no matter what.
This is a possible future – retail’s own numbers suggest it may already be happening, as Dan Frommer unpacks in The New Consumer. AI driven traffic is up, conversion is up, consumers coming in from AI end up on the product detail page and likely have discussed your product with an LLM “several times.”
On the other hand, retailers could find a way out of this mess, by building valuable IRL experiences – ones that are both on brand and also localized.
Kind of like Starbucks has done with their TikTok creator initiative. I love highlighting these kinds of stories, not least because I strongly believe that locally sourced content, with a face and a name that you could actually meet by walking into a store, is an important way to build local connections under a big brand umbrella. Even better, Starbucks seems willing to let their employees reap the results of their organic and perhaps someday viral creativity.
The one thing they should be careful about is making sure their associates don’t put up AI slop, which would undermine the whole initiative. But otherwise, a genuine person with genuine excitement about the company and its brands they represent, hustling not just to get the job done but to also generate a lot of traffic on the side? What’s not to love about that? And about how genuine and impossible that would be for AI to duplicate?
What Have We Learned This Week?
I said last time that I go to many lengths to ensure that I inject some personality into my writing. This is both a rebellion against decades of writing research and white papers with bland corporate-speak personality, and also proof of life.
I thought, when I first started reading his articles, that Ed Zitron was over the top and needlessly abusive with his language. I now see it as a way he ensures that he comes across as 100% pure, old-fashioned, homegrown human, as they say in The Matrix.
The overall point this week, and I unfortunately have to give Claude credit for this line (I ask it to identify any through-lines that I missed or didn’t fully develop, for the record): the behavior we are alarmed by in the classroom is the behavior we’re celebrating in the boardroom.
I don’t think we’re alarmed enough about what’s happening in schools. And we absolutely need to be more alarmed about any company rush to push people to act more like the students we’re trying to snatch back from the brink of total brain rot.
In between, there are shoppers. It’s still overall a small percentage of the total number, and yet again another week has gone by where no one has come forward to definitively state that AI drives higher conversion rates versus just brings people in farther down the funnel.
So we’re looking at high potential for brain rot on both sides of the coin – shoppers who no longer really know how to shop for themselves or how to critically evaluate whether a product is the right one for them, and retailers who no longer know how to uniquely position their products to stand out to shoppers (or their AI shopping agents, for that matter).
Do you want the red shake or the blue shake in your hoverchair tonight?
Program Note: I am off to Atlanta this week and not sure how much I’ll be able to stay on top of the reading list. So no promises, but I’ll try.
Until next time!
Nikki
I write this weekly article as much to force myself to make sense of what I read on a daily basis as to say anything about it publicly. I will continue to do so - for free - for as long as I can sustain it.
If you like what you’ve read and you haven’t subscribed, please do! It helps me know I’m saying interesting things.
If you’re already subscribed but you can think of someone who would enjoy reading this, please share the article.
Or better yet, share the whole publication and let them get in on the weekly fun!
Thanks for reading, and thanks for your feedback and support!