We're Drawing the Robot
Every generation imagined the future and got the gadgets weirdly right and the world weirdly wrong. The pattern of their mistakes is the most useful thing we have for spotting our own.
Unsafe Intelligence
Here is the strange thing about old visions of the future: the errors are not random. They rhyme. Star Trek nailed the flip phone and the tablet, then handed them to a disciplined naval crew taking orders down a chain of command. The Jetsons gave us a flying car and a 1950s housewife. We were promised moon bases and got group chats. If the misses were arbitrary you could shrug them off — but they fall into a handful of repeating shapes, and once you can see the shapes, you can aim them at your own predictions and ask the uncomfortable question: which of the things I'm certain about is the housewife?
How futurism reliably goes wrong
Six patterns do most of the damage.
They get the direction right and the texture wrong. Star Trek's communicator, PADD, and talking computer all arrived — as a surveilled, ad-funded, addictive consumer product owned by a few corporations, not as a ship's utility. Right device, wrong political economy. The vector was correct; every detail of the texture — who owns it, what it costs, what it's for — was off.
They keep present-day society and only upgrade the gadgets. People can imagine the machine changing. They cannot imagine their own institutions dissolving, so they bolt tomorrow's tech onto today's org chart. The flying car still drives Dad to a 9-to-5. The starship runs on a navy.
They over-predict the spectacular, under-predict the mundane. Flying cars are legible, cinematic, and physically hard. Texting is illegible, boring, and world-changing. The imagination reaches for the dramatic object and walks straight past the cumulative, invisible one.
The killer app is always a surprise, and usually social. Bell pitched the telephone for broadcasting concerts. Nobody guessed it would reorganize teen life or collapse the wall between work and home. The second-order use — what people actually do with a thing — defeats every forecaster, because it's about human want, not technical capability.
The thing that matters is orthogonal to the thing everyone's watching. The rocket-age futurists stared at propulsion and missed the microchip sitting next to them. The real revolution was in the boring adjacent column.
And the whole picture is a self-portrait of its moment. 1950s sci-fi runs on automation and conformity fears; 1980s cyberpunk on Reagan-era megacorp dread. The future is a mirror. It tells you more about the year it was dreamed than the year it depicts.
So what are we getting wrong right now?
Point those six at our pictures of AI and the singularity, and the suspects line up fast.
We're drawing the robot; we'll get the ecology.
Every image of "AGI" is a single unified mind — HAL, Data, the company-in-a-box that wakes up one morning. The likelier reality is plural, disembodied, unbranded, ambient: a thousand narrow-ish services woven into everything, with no face, no single self, no moment of awakening. Asimov gave us positronic androids; we got a text API in the cloud. The "it" you're picturing may never exist as a thing. It'll feel like the weather.
We're watching for the arrival; the story is the boring middle.
Everyone is staring at the threshold — the day AGI is declared, the Singularity. But transformative periods are long, ambiguous, and un-dateable, and we'll argue about whether it "happened" for years. The economically decisive work is mundane: claims adjudication, code review, customer service, logistics. Not a glowing oracle. The revolution will be hard to date precisely because it has no premiere.
We're imagining superintelligence inside 2026's institutions.
We picture AGI with jobs, firms, the 40-hour week, universities, copyright, and borders all intact. But the printing press didn't just speed up scribes — it produced mass literacy, the Reformation, modern science, the nation-state. The deepest AI changes will be to the institutions themselves, in ways we literally can't render because they're the water we swim in. The wage, the credential, the firm — those are the naval hierarchy on the starship, the thing our grandchildren will find quaint.
We're afraid of Skynet and blind to the boring catastrophe.
The cinematic risk — rogue superintelligence — is vivid and probably not the one that gets us. The likelier bad ending is mundane: a world that quietly routes more and more decisions through systems nobody understands, humans deskilling and deferring, information filling with cheap synthetic slop, agency leaking away not in a bang but through a thousand conveniences. It's undramatic, which is exactly why it's underrated.
We modeled the productivity tool; we'll get the relationship.
We assume rational adoption for work. But the mass-market killer apps will probably be emotional and status-driven: companions, parasocial intimacy, grief-tech, personalized entertainment and propaganda, self-deception as a service. The biggest social effect of intelligence too cheap to meter might be loneliness, not GDP. We predicted Spock; we'll get Her.
The axis that matters is probably price, not genius.
Everyone's watching for smarter. The bigger lever may be the price collapse — intelligence becoming a near-free utility, reshaping the world the way electrification did: ambient, infrastructural, boring. Judging the AI age by how smart the models get may be like judging the computer age by how good chess engines got.
The one durable lesson
Retrofuturism teaches humility about texture, not direction. Past futurists usually got the vector roughly right — more computing, more connectivity, more automation — and the texture totally wrong: centralized became distributed, expensive became free, embodied became ambient, spectacular became mundane, utility became addictive product. So the safe bet for us is "more capable AI, big changes." But nearly every specific image we're holding — the humanoid mind, the clean before-and-after, the dramatic awakening, the work-but-with-robots world — is probably wrong in exactly the texture-dimension we can't yet see.
Which is the whole argument for how this lab thinks. Stop trying to draw the destination. The destination is the part you will get wrong. Pay attention to the crossing — because the crossing is the only part you can actually watch while it happens.