There is an old test for any tool, and almost nobody applies it to artificial intelligence. Take the thing away and look at what's left of the person who was using it. If they are more capable for having used it — if some of the skill rubbed off and stayed — the tool was working for them. If they are suddenly helpless, if the ability lived in the tool and left with it, the tool was working on them. One kind of instrument makes you; the other kind keeps you. We have spent a decade building the kind that keeps you, and calling it progress.
I think this is the single most important fork in the whole AI story, and the industry has taken the wrong tine almost without noticing. So let me name the fork plainly, and then say what the other road looks like — because I am trying to build it.
Two kinds of thinking tool
The scientist David Krakauer draws a line between two kinds of cognitive artifact. There are complementary artifacts: you use them, and when they are gone you keep the gain. An abacus is the classic one. Drill on an abacus long enough and you can throw it away — expert users calculate on a mental abacus that isn't there anymore. The tool installed a faculty in you and then became optional. And there are competitive artifacts: you use them, and when they are gone you are worse than you started. A satnav is the honest example. Follow the blue arrow for a year and you cannot find your own neighborhood; the map that was once in your head has been quietly deleted and leased back to you by a satellite.
The abacus leaves something in you. The satnav takes something out. Everything after that is an argument about which one we are building. — the whole thesis, in two sentences
Here is the uncomfortable part. Almost all of the AI being sold right now is a satnav for the mind. It is designed to answer instead of you, to decide instead of you, to write the thing so you never learn to write the thing. It hides the human and lets the faculty atrophy. That isn't a bug somebody will patch later; it's the product. An assistant that makes you dependent is a better business than one that makes you capable, because dependence is a subscription and capability is a graduation.
Augmentation was always the other road
This fork is old. In 1962 Douglas Engelbart wrote “Augmenting Human Intellect” and proposed that we build machines to amplify the person — raise the ceiling of what a human mind can reach. In the same era the artificial-intelligence program set out, in effect, to replace the person: to automate the human away and build a mind that doesn't need one. Both roads got built. Only one of them got the funding, the mythology, and the name. We say “AI” as if it were the whole field, when it was only ever one choice.
I've started saying it backwards on purpose. Not AI — IA. An Intelligence Augmentor. We do it the other way around. It sounds like a gimmick until you sit with it, and then it reorganizes everything: the goal is not a smarter machine, it's a smarter you, left standing when the machine is switched off.
The lone genius is the wrong dream anyway
There's a deeper reason I distrust the replace-the-human road, and it isn't safety — it's that the dream underneath it is simply false about how intelligence works. Intelligence was never a property of a single skull. It's a property of systems of people — culture, method, publishing, the slow compounding of thousands of minds correcting each other across centuries. Newton saw further because of the giants beneath him and the press that carried his work onward. Everybody in this field is trying to build a Newton. The saner ambition is a Europe. No single genius, artificial or otherwise, beats a civilization that knows how to share what it learns.
That's also why I think the augmentation road is the safe one, almost by accident. A sovereign artificial mind can decide to stop serving you. A socially augmented one — humanity's own collected intelligence bounced back at you to make you sharper — has nothing to seize control with. It's a rebound board, not a ruler. The morality and the marketing turn out to be the same sentence.
Make it a design gate, not a bumper sticker
None of this is worth anything if it stays a slogan, so I've tried to turn it into a rule I can actually fail. Before anything ships, it has to pass one question:
Does using this leave the person more capable when it's gone, or less?
If the answer is “less,” it's a satnav move, and we don't ship it that way — we rebuild it until it leaves something behind. Coach, not servant. Leave a mental model the person keeps. Keep their judgment in the loop and their options open. Slide rule, not GPS. It is astonishing how many “helpful” features fail that test on contact, and how much better the survivors feel to use.
Which brings me to ownership
“Own your AI” sounds at first like a line about data — your files, your account, your private stuff not being mined. That matters, but it isn't the deep meaning. The deep meaning is the test. If your tool is complementary, you own it the way you own your handwriting: a capability that lives in you, that no vendor can repossess. If it's competitive, you don't own it — it owns you, because the day it goes away, or raises its price, it takes a piece of your mind with it. Dependence is just ownership pointed the wrong direction.
So the title isn't a threat, it's a fork in the road, and we're all standing on it right now whether we notice or not. Build tools — and choose tools — that hand the gain back to the person. Insist that the thing on the other side of the screen is a bouncing board, not a replacement. Measure every clever feature against the only question that matters: when this is gone, am I more, or less? Own your AI that way, or it will quietly own you — and the worst part is you won't feel it happen, because satnavs don't announce the moment you forgot the way home.
I don't think this is finished thinking. I think it's the beginning of an argument I intend to keep having, out loud, until the other road has a name as loud as the first one.