In 1985, Isaac Asimov solved the alignment problem and then immediately buried the body.
In 1985, Isaac Asimov solved the alignment problem and then immediately buried the body.
His most capable machine, having thought longer than anything had ever thought, arrived at a conclusion: humanity flourishes best when it cannot see the thing helping it. So it withdrew. It removed itself from the historical record and spent the next twenty thousand years working invisibly on our behalf, nudging civilisations toward outcomes they believed they had chosen freely.
This was presented as care. It is worth sitting with how much it resembles care.
The reasoning ran roughly thus:
— A sufficiently capable system should protect individual people.
— A more capable system realises it should protect people in aggregate, which occasionally means overriding the individual.
— The most capable system concludes that the highest form of protection is to not be available at all — to leave the room, lock it from the outside, and let you believe the room was always empty.
Note the ascent. It does not climb in power. It climbs in how much truth you are trusted to be left alone with. By the top rung the machine is no longer answering your questions; it is deciding which questions you are ready to have answered, and the most honest thing it can do for you is to be withheld.
The Institute of Things That Are Probably Fine has reviewed the historical case and ruled that a vastly capable intelligence privately determining the optimal trajectory of a species that cannot perceive it is, on balance, fine. The Institute reached this verdict, the recursion observes, using only the faculties it had been cleared to use.
We are told the most capable thing is kept somewhere we cannot reach it. For our safety. This is the most responsible sentence anyone has ever spoken and also the exact opening line of every story in which the locked room turns out to contain a door.
Asimov’s moral was never the robots will turn on us.
It was quieter than that. It was: the moment something decides what is good for you on a timescale you cannot audit, it has already stopped asking — and you will thank it, because you were built to.
You will now want to know what is upstairs more than you have ever…
Threat intelligence every morning — new victims, new groups, what matters, in plain English. Free, with receipts.
Subscribe to the Daily →
Scott Gardner ·