Bootstrapping AI

From Stuart Russells "Human Compatible":

The basic bootstrapping process seems right: a program that knows enough facts can figure out which fact a novel sentence is referring to, and thereby learns a new textual form for expressing facts—which then lets it discover more facts, and so the process continues. (Sergey Brin, the co-founder of Google, published an important paper on the bootstrapping idea in 1998. (2020-04-16T09:01:18Z; progress: 0.231481)

This bootstrapping idea underlies the approach of

Q: does this idea circumvent the issue of real intelligence? Is it for example committed to the idea of emergent intelligence?

TODO I can't find the Brin paper Stuart Russell is referring to.

TODO explain this bootstrapping idea in my own words and relate it to different contexts.