AGI has eleven paths: mathematical (fractals, entropy), biological (evolutionary simulation), brain-structure, brain-thinking (empiricism, rationalism, analogy, Bayesian), plus two honestly marked unrealized: Einsteinian thought and emotion. But the paths are not parallel lanes — they are an elimination tournament. Every few years a path is proven dead or absorbed into another.
The tournament has one criterion: which path first builds "an improver that improves itself". Symbolism once led, surpassed by statistical learning; evolutionary methods had their moment, marginalized by gradient descent; today's Transformers are an engineering hybrid of empiricism and Bayes. Paths are merging, not competing.
Three identification methods. One, watch the money: the truly hopeful path absorbs 80% of resources. Two, watch citation networks: papers cited by all camps are the future's common foundation. Three, classify the bottleneck: compute bottlenecks are engineering (will resolve); data bottlenecks are structural (hard); paradigm bottlenecks are dead ends (will reroute).
The two "unrealized" paths are the best bets. Einsteinian thinking — thought experiments from minimal data — is exactly what large models lack and what reasoning models approach. Emotion — a compression algorithm for value judgment — if proven necessary, will turn the whole field.
Personal lesson: the tournament of paths is a tournament of practitioners. Those betting on a single paradigm expire with it; those betting on the north star — "an improver that improves itself" — survive paradigm shifts.
Full map of the eleven paths: psyverse.fun
Top comments (0)