“The history of invention is not the history of a necessary future to which we must adapt or die, but rather of failed futures, and of futures firmly fixed in the past.”
― David Edgerton, The Shock of the Old: Technology and Global History since 1900
Acknowledgements: Matthew Blakemore for productive sparring and Rachael Ogunbowale for unwavering support.
Shifting metaphors: 📰 → 🔎 → ✨
In early internet days, in many ways parallel to our present hype around AI, everyone was thinking with portals.
The idea of how to capture the attention and loyalty of emerging cybernauts and web surfers was to cram as much of the internet into one end-all, be-all, catch-all start page with a neatly categorised selection of sundries, broad enough to appeal to everyone.
Now AOL and MSN are still around, but Alta Vista — once the undisputed “start page of the internet” — sure isn’t, and we all know why: the winner turned out to be not the one that carried over the notion of a curated topical overview from the then-dominant paper & ink newspaper paradigm to the nascent medium of the web “page” unchallenged — but the one that went for the bog standard, dirt simple UX native to the medium; the default one drilled into every computer user from the start: a blinking cursor on an empty command line.
Instead of a buffet of options, your blank page to fill, with an active invitation to press an expanding range of fingers against plastic squares, free from distraction.
And with the same allure:
It’s all here at your fingertips, at your command. Anything you can think of.
Under hood, out of mind
Now unless you work in online propaganda, you are likely blissfully unaware of the raging arms race between SEO spammers, bot farms and multidimensional filtration deliberations under the Google hood; you accept that empty search box as your convenient entryway to all the internet you could ever possibly need.
You can happily disregard not only the raging information wars, but also the massive sprawl of data centers, fiber optics and radio waves, their hardware and infrastructure in turn, the maintenance, the power grids that make it all tick.
Like the pumps and flares and mines and refineries all required to conjure those plastic squares and glass rectangles at your fingertips, it’s just there.
Massive, surely, in some vague way, but unseen and therefore largely unheeded.
And by that same standard, the information generated with every press of a button and swipe of a mouse cursor intuitively taken to be as ephemeral as any other mode of communication drilled into our genes and cultures: lost to the winds, like a gesture, a wink or a whistle.
Except it wasn’t.
It never was.
Which externality will you deny today?
When the costs became more apparent and the tech commoditised, a range of options sprang up: Ecosia, the search engine that acknowledges environmental cost and does something about it (one tree planted for every query — hopefully in some corner of the world that doesn’t burn), DuckDuckGo, the one that ostensibly doesn’t spy on you (not that you’d be able to tell), Bing, the one by the rivalling Pepsi-or-Coke empire, Yandex, the one by the rival superpower, and so on through a range of niche products.
The same wretched consumer choice as at your local supermarket: should you go for the one without child labor today, or the one that pollutes somewhat less? Be sure to bring your magnifying glass for the icons.
– I would like something without rat in it.
– That would be this one right here sir, the rat soufflé. It only has some rat in it.
There are search engines that let you straight up, you know, pay, for results rather than slice off parts of your eyeball with every page of results served. Just as you can pay for email, or online privacy, or watching user-made videos.
But who does that?
As Jaron Lanier has expounded on, free is expensive.
As Tristan Harris quipped, ad-funded social media becomes a race to the bottom of the brainstem.
And as Cory Doctorow has expounded on, “free” enshittifies platforms over time.
With AI generated content now eating the user generated content paradigm, “free” enshittifies the entire internet.
How to slay giants
When our new batch of tech oligopolist hopefuls set off to go from zero to one, they revisit those same first principles: how do we leverage online anonymity to harvest the abundance of digital content, leverage global labor arbitrage by outsourcing data cleanup to poor nations, leverage Moore’s law by processing it through large data centers, then serve it in as intuitive a way as possible to users (text prompts, then speech) with the lowest possible barrier to entry (freemium) to harness end-user ghost labor (prompting, image uploading and curation for RLHF) and make it maximally developer viral for mindshare (open weights via Hugging Face, OpenAI APIs).
A solid infrastructure play, in the words of former Stability CEO Emad Mostaque. The race to bottom of the brainstem in one end, to viral social feeds for leverage and to the bottom of the tech stack in the other end.
It is no coincidence that Midjourney, Stability Dream Studio and ChatGPT all started their end-user offers as a blinking cursor.
Only this time around, the value proposition is raised from serving a curated slice of internet, to a pre digested slice of internet.
And with the same allure:
It’s all here at your fingertips, at your command. Anything you can think of.
Having studied the playbook of past winners, today’s fine young cannibals keep the inner workings of the machine neatly blackboxed, armed guards patrolling remote data centers.
What is new here is the pivotal marketing spin:
You made this.
Rather than presenting the query-response dopamine loop as an act of iterative retrieval, it is framed as an act of creation:
Text-to-X
Now to be fair, there are sound technical reasons for that terminology. It captures one of the many end-to-end technical processes of generative AI, most exciting perhaps being multi modality — where the statistical model that results from large scale analysis of media corpuses (I take pains to not anthropomorphise by calling it “training”) enables mathematical translation of a data stream from one modality to another at near real time; your face puppeteering Trump’s, your whistle turned into an electric guitar; a video transcribed to text; a poem turned into music and so on.
However, it also comes with a linguistic trap frequently leveraged as a sleight-of-hand in marketing: to posit your query as the point of origin.
“Create images from just text prompts”
Or in emoji, the increasingly ubiquitous:
✨
Now one could charitably argue that content generation is “creation” depending on how one loads the term. The devil here rather lies in “from just”: two words that magically vanish the gigabytes of compressed media data from the myriad sources and terabytes of proprietary works harvested, to be replaced by the “creative” act of asking a computer system for an output.
And from there, a massive snowball of accumulated misconceptions thunders down the alp side in ensuing debate:
Gutenberg and the scribes.
Ford and the conveyor belt.
John Henry toiling to death in his mine, caught in a futile race against the machine.
The tragically romantic “doom and loom” fate of the luddites of Manchester.
And on to photo vs painting, Photoshop vs traditional painting, CG versus traditional animation and so on.
All of these techno determinist-defaitist fairy tales of labor vs automation that at best elicit a pitiful pat on the head of “sorry we democratized your skills — but perhaps your job shouldn’t have been there in the first place.”
Recentering property rights
By centering efficiency of process we obscure a key difference to all of the above examples:
Gutenberg never raided every monastery for handwritten manuscripts, nor did he clam tithe on every future manuscript of theirs (h/t Dragan Stiglic)
Ford motor carriages were not manufactured from horse parts (although to be fair, the occasional old mare might have been rendered down for glue).
John Henry did not pedal the machine he overexerted himself against.
The luddites of Manchester weren’t being shaved for wool.
Early photographers were not labeled painters, and were not given the right to depict any person and any artwork without their permission.
Silkscreen printers, and later Xerox operators, were not labeled photographers, and were not given the right to mass reproduce the work of photographers.
In short, all of the previous popular framings present the process devoid of its essential inputs and legal boundaries.
![](https://substackcdn.com/image/fetch/w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F9df467ef-9730-423d-a5e3-59dab582324c_1170x812.jpeg)
By labelling the querying of a database for synthetic derivatives of pre-existing works as “creation”, we have not only “empowered creatives” and “democratized creativity” (as eloquently torn down by Brian Merchant) in exchange for word-of-mouth marketing and advocacy. We have also created a very useful false equivalence:
We neatly diverted the entire debate from one of property rights to one of automation. From capital to labor. From past and ongoing exploitation to hypothetically future labor replacement.
We get to tap all those juicy tropes of progress, inevitability, efficiency, innovation.
It is so much sexier to think of Runway as the future of media production than as yet another stock remixer.
We have wrested the means of legitimate protest out of the hands of “luddites”.
It is so much more seductive to imagine image generators as the ubiquitous trope porcelain terminator at the easel than a rather unsexy content McFlurry machine:
![NVIDIA H100 PCIe 80GB | VideoCardz.net NVIDIA H100 PCIe 80GB | VideoCardz.net](https://substackcdn.com/image/fetch/w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F64310019-d66a-437b-a48b-6d526803d034_1200x900.jpeg)
And we can muse and wax lyrical about future UBI.
Instead of engaging in the much harder and more contentious issue of licensing.
We have effectively shut down the discussion around our fundamental right to informed opt-in consent as to how our proprietary, private and personal information is being used – just as it is being weaponised against our values and livelihoods.
Small is beautiful
Founder fantasies of scrounging together 7 trillion for some new nukes to bootstrap their trip to Mars and building The One Web Portal To Rule Them All aside — and with an eye on Big Tech incumbent’s race to control the hottest startups while dodging anti-trust — we’re likely to see a much more diverse and interesting field once the present generation of illegal datasets are taken offline, models get regulated and model developers are fined into line. One only gets away with an internet-wide rug-pull once. The publishers who fell for it scurry to shutter their windows, bolt their barns and run for the hills.
Plenty happening in the ethical second wave of models emerging over at Fairly Trained, for text, music and images — narrower in purpose and scope, built on clean data and yielding royalties to rights holders.
Regenerative systems that remunerate fairly and do take “no” for an answer.
As it should be.
Excellent piece, Johan. Thanks for writing it. The Luddites are a much-misinterpreted movement, having been reduced to simply smashing machines, as opposed to protesting against the transfer of wealth from the many to the few (sound familiar?), and against the reduction of worker protection. They started in Nottingham, where I live. In fact my office is in an early 19th C factory, which made lace, and who knows, may have borne witness to elements of the start the of the movement back in the day!
Great title! Fantastic read - especially the false equivalence analogies of Gutenberg, Henry Ford, Manchester textiles, etc. I liked how you underscored this bait-n-switch tactic of focusing on UBI as the solution instead of upholding private property by endorsing Licensing.