The Superintelligence Problem

An Evidence-Based Probability Map

Artificial intelligence is advancing faster than almost anyone predicted. This interactive document maps 120+ probability estimates across the key questions: How soon could AI surpass human intelligence? Could we lose control of it? What are the chances things go well — or catastrophically wrong?


The estimates are grounded in mid-2026 empirical evidence from government safety institutes, AI lab research, and expert surveys. Every number carries its evidence in place and its full change history from the previous version. The citation layer has not yet passed independent verification — sources are claimed, not confirmed, until the Jürge report on this document publishes.


This is not a prediction. It’s a structured framework for thinking about the most consequential technology transition in human history.

25–35%
estimated P(extinction from AI) OPEN
Model-derived from conditional probability chain — a forecast, tagged as one
8%
best-case scenario (aligned & meaningful)
2027–29
expected arrival of human-level AI
120+ interactive estimates · 15 expert profiles · 5 research reports synthesized
By Kevin Scharp · v2.1, July 8, 2026 · A Solonic instrument · Status: OPEN — errata logged in §0 · This page is an instrument and runs JavaScript · Built on analysis of Yudkowsky, Bostrom, and 5 deep research reports