In three years, generative AI has reached fifty-three percent of the world's population. Faster than the personal computer. Faster than the internet. The noise about what AI is grows at the same speed as its use, and the two get confused. This blog starts from a non-negotiable filter: this is not what it looks like. There's no hand-holding here, no reassurance. The question and the answer will be equally uncomfortable.
The figure nobody disputes, and everything else
Fifty-three percent is not a marketing estimate. Stanford HAI publishes it in its 2026 AI Index Report, and it sets the curve against the two obvious benchmarks of the previous century: the personal computer and the web. Both took longer. Both gave society time to get used to the tool before it turned invisible. Here there's no such margin.
What is open to dispute is everything else. And the disputing has been dreadful.
Open a newspaper, a TED talk, a thread on X, a YouTube video, a hallway conversation: you'll find two opposing stories told by the same people. Existential risk and infinite productivity. Skynet in the morning, copilot in the afternoon. The curious thing about it is that both stories share a sender: the same voices warning of the apocalypse are selling the salvation. When two opposite versions compete for the same listener, what happens is that the listener stops looking at what's right in front of them. They stare at the two doors and argue over which to open, never noticing there's a third corridor they're already walking down.
That third corridor is what interests me here.
The filter. This is not what it looks like
There is no agenda behind this blog. No closed thesis. No list of victims to defend or villains to denounce. What there is, is a single underlying suspicion that orders everything else: the AI revolution is not what gets told. Not what its sellers promise, not what its accusers fear. It's something more opaque and faster than either, imposing itself without anyone having chosen it while the noise busies itself with the two false versions.
That's the whole filter. Applied case by case, post by post.
Applied means: when a headline appears, before taking a side, ask what story they're telling and what's actually happening on the plane that headline doesn't name. When productivity comes up, look at what's being lost. When risk comes up, look at who's winning while the risk is being argued. When someone announces that AI is going to save education, look at what AI is doing in classrooms right now. And the other way around: when someone announces its imminent collapse, look at who's billing while the collapse is announced.
It's not professional skepticism. It's a habit of stepping off the frame you're handed.
Why Tears in rain
The title is on loan from a 1982 film: Blade Runner, directed by Ridley Scott. At the end, a replicant named Roy Batty —who has hunted humans throughout the plot— sits on a rooftop in the rain and speaks to the man who has come to kill him. In the original English monologue, written in part by Rutger Hauer himself, he says:
I've seen things you people wouldn't believe. Attack ships on fire off the shoulder of Orion. I watched C-beams glitter in the dark near the Tannhäuser Gate. All those moments will be lost in time, like tears in rain. Time to die.
What matters about the monologue isn't the nostalgia. It's the fact of it. Things that existed and stop existing because nobody looked at them in time. The memory of a being who knows it's about to switch off and understands that the only things it has seen go with it.
The same scene, another scale
That is, at another scale and another speed, what's happening with AI. Each new model drags the previous one off before the previous one has had time to settle into the culture. In 2023 there were months of public discussion about GPT-4. Today those discussions are footnotes; the models being debated are outdated toys; the questions that then seemed urgent, nobody remembers. The conversation never reaches the point where what happened gets understood: by the time it's understood, something else is already being argued.
This blog is the attempt to stop and look before that row of moments dissolves. Not to file them in a drawer. To see them while they're still there.
The pragmatic voice, not the philosophical one
There's a lot of philosophy of AI written from the armchair. Arguments about consciousness, will, the soul of the machine, emergent identity. Some are good —Searle, Dreyfus, Penrose have pages that still hold up— and many are pretty exercises that avoid touching the object.
Not here.
Here AI is looked at as what it is right now, in 2026: a set of concrete models, with concrete design decisions, concrete energy costs, concrete regulations, concrete consequences for how a person who uses them every day thinks. Philosophy shows up when it's needed to understand the object, not as a starting point to replace it.
Pragmatic means something else too. It means that what matters more is the effect of delegating a decision to a model than the debate over whether the model "thinks." What matters more is what a kid loses learning to write with an assistant in front of him than the debate over whether the assistant has intentionality. The abstract, when it works, helps you see the concrete better. When it doesn't, it gets in the way.
What this blog will not do
It will not simplify. AI isn't complicated out of whim or technical showing-off: it's complicated because it mixes statistics, energy infrastructure, legal decisions, labor markets, cognitive psychology, intellectual property and geopolitics in a single piece. Simplifying that is lying. The fight here is to explain with precision, not to explain in thirty seconds so it fits in a short.
It will not reassure. If what you're after is someone to tell you everything's going to be fine, or someone to tell you everything's going to be terrible, there are thousands of accounts a click away doing exactly that. Both, besides, generate similar "engagement" and are interchangeable depending on the day. Having your own blog serves precisely so you don't have to dumb the message down to one of those two templates.
It will not default to defending the harmed party. The easy narrative says the worker is the victim, the artist is the victim, the user is the victim, the minor is the victim, the global south is the victim. Sometimes they are. Sometimes they aren't. Sometimes the picture is far stranger than that —the supposedly harmed party benefits, the supposedly benefiting party loses, both lose on different planes. The blog looks case by case. Empathy doesn't work as an argumentative shortcut.
It will not draw conclusions. Paragraphs close, posts close, but the final wrapper, that kindly paragraph that gathers it all up and hands it back to you in one digestible sentence, is not going to appear. The conclusion is the reader's to draw. If after reading a post you feel discomfort, the discomfort stays with you. It doesn't get neutralized with an epilogue.
It will not use rhetorical tricks to retain the reader. None of discover, we'll tell you, the answer will surprise you, don't miss it. That language has its place, and its place is not this one. What you see is what it is; if it hooks you, it hooks you for the subject; if not, it doesn't.
The relationship with whoever reads
Almost all the traffic to a blog like this arrives via Google to a single loose post. Nobody lands on the front page. Nobody follows a series in order. Nobody has read the previous post when they reach the next one. That forces a practical thing: each text stands alone. It opens its whole question inside its own space and closes it inside that same space, without leaning on "as we saw in the previous post" or "we'll come back to this."
The editorial consequence is severe. It means no series or narrative arcs get built. It means the same nuance can show up treated twice from different angles in separate posts, because each one has its own single reader, its own different entry, its own different exit. The apparent repetition isn't repetition: it's each nuance claiming its own spot.
You are not an audience. You're someone who has stumbled onto a text. The text addresses you, it doesn't accompany you. If you're expecting the warm voice of the explainer who walks you through it slowly, pausing so you can breathe, this is going to sound dry to you. It is dry. It's not a flaw that will be fixed later.
The editorial pact
It fits in one sentence: the question and the answer have to be equally uncomfortable.
If a post opens with a hard question and closes with relief, the post is wrong. If after reading you feel calmer than you did when you started, the blog has failed you. Not because calm is bad —sometimes it's the right call— but because that kind of ending is usually a small lie: the feeling that the matter is settled, when the matter is not settled. The content industry lives off manufacturing that feeling at scale. Not here.
The opposite commitment doesn't work either. The point isn't to always close on a dramatic note, or to add a gratuitous shiver in the last paragraph to seem profound. That's a trick too: the professional gloom-monger who confuses hardness with wallowing. Honest hardness is the kind that looks at the problem and leaves without dressing it up, neither downward nor upward.
The two readers
There's a type of reader this is going to drive away. The one who needs a spiritual guide to artificial intelligence, the one who wants their fears confirmed, the one who wants their hopes confirmed, the one looking for a warm travel companion. Fine. There are other places for that, plenty of them. What you can't do is serve those readers and make this blog at the same time: the two are incompatible.
There's another type of reader I am talking to. The one who has already noticed on their own that the noise doesn't add up, that the two official versions fall short, that something in the picture isn't named yet. To that reader the blog won't hand finished answers. It'll hand them ways of looking. Ways of looking made with care, different in each post, all with the same filter behind them.
References
Scott, Ridley (director). Blade Runner, 1982. The blog's title and central quote come from the closing monologue of the character Roy Batty, played by Rutger Hauer, partly improvised by the actor himself during filming. Text verified at Tears in rain monologue — Wikipedia.
Stanford HAI — AI Index Report 2026. Source of the figure on generative AI adoption by fifty-three percent of the world's population in three years, with the comparison against the historical adoption pace of the personal computer and the internet. Available at hai.stanford.edu/ai-index/2026-ai-index-report.
También te interesa
- How public conversation about AI broke
- Why AI criticism isn't written by real experts
- The problem of defining intelligence
No comments yet
No comments yet. Be the first.
Leave a comment