Rethinking Fear of Recursive AI

STABILITY ENGINEERING – Volume 9 www.manhattanproject20.com 1 117-45-D Rethinking the Fear of Recursive AI From models to cognitomes, from control to stabilization Introduction One of the most common fears in artificial intelligence is the fear of recursive self-improvement. • The argument is straightforward. • An intelligent system modifies itself. • The modified system becomes more capable. • The more capable system modifies itself again. • The cycle repeats.