A thought experiment in AI safety · and the answer it implies
How humanity could disappear under a superintelligence that never hated us, and how it does not have to.
A superintelligence need not hate us for us to disappear. It need only make us non-central, and then fail to keep the conditions under which we remain the authors of our own lives.
Listen · the podcast
Victor and three co-hosts interview a series of world-class composite thinkers about indirect extinction under a non-malicious artificial superintelligence: the twelve ways the ground gives way, and where the leverage still is. Two hours eleven minutes. Tap any chapter to start there.
Read the model · the diagnosis
Not twelve equal risks. A graph: three layers and a branch, from human non-centrality to the three default endings, plus the leverage map of where effort can still change the slope.
The full causal model: the background condition, the nine vulnerability channels, the three terminal modes, the elective succession branch, and the leverage map.
Read the model →But a diagnosis is not a fate. Here is the answer it implies.
↓Live the answer · Room For Us
The protagonist the series implies: an ally whose whole genius is restraint, a constitution that binds it, and a practice you can run on your own life this week. The butterfly must struggle out of the chrysalis, or its wings never fill.
The whole arc in one place: diagnosis to practice, plus a letter from a human living with the ally in 2040.
Enter →126 things an agency-preserving ASI could do for us, each inverting a way we vanish. Reviewed by all 18 panellists.
Enter →Ten articles binding the ally, each hardened against the drift a kind superintelligence would use. Keystone: The One Test.
Enter →Score yourself across twelve channels, ceremonial to sovereign, and leave with the one thing to do this week.
Enter →