202409220019
Status: #idea
Tags: #learning_theory
# Expert knowledge is vector-retrieval + graph search
We can think of knowledge as being represented as a graph, with related concepts linked together by edges. Concepts can be related by "rational" thought (e.g. using logical operators to transform one concept into another) or "intuition" (e.g. recognizing that a one situation is similar to another purely at a subconscious level).
Theoretically you can use the rational mode to get to the correct answer, but the mental strain for any non-trivial example makes this unwieldy.
In practice, experts use the equivalent of a vector search over their previous experience to find the nodes that are closest to the answer (i.e. they use the "intuition" edges). This way they only need to make a couple reasoning hops (i.e. moving along "rational" edges) to arrive at the answer for this particular situation.
---
# References