About
    • AMMDI is an open-notebook hypertext writing experiment, authored by Mike Travers aka mtraven. It's a work in progress and some parts are more polished than others. Comments welcome! More.
Search
Page Contents
Full
Incoming links
from Marvin Minsky/Society of Mind

Examining the Society of Mind

30 Oct 2021 - 11 Apr 2026
Open in Logseq
    • Paper by the late Push Singh. I don't know that I ever read it, preferring to keep my own direct interpretations of Society of Mind, but suddenly realized that is dumb. Also SoM itself may be dumb. Obsoleted by new approaches.
    • Promising opening as he emphasizes the multiplicity of SoM. You need many different kinds of mechanisms, and each one presents fascinating sets of what are basically engineering problems.
      • You can see that this view is grounded in standard programming or engineering practice. Problems are carefully analyzed, solutions constructed from legible combinations of legible parts. This is GOFAI. Marvin's view was rather more far-reaching than others, but the approach was the same.
      • ML is utterly different. You don't have to understand anything about how vision or reason or motor actions work – you just throw things together, put in some concrete goals or examples, go nuts.
      • Marvin's thinking led in exactly the wrong direction here. Hard for me to admit that, and "wrong" may be too strong – but it wasn't the direction that led to the current success of ML-based AI.
    • Goes back to earlier Minsky paper A Framework for Representing Knowledge
    • Ref to Smalltalk and Actors, good
    • Quote from 1976 SoM draft, refers to "sub-persons"
    • Later use of corporate metaphor (probably unfortunate, the mind is not organized like a corporation I'm pretty sure)

    • Note

      • (belongs elsewhere if it's actually of interest) Obvious stance of Marvin and his followers: yes LLMs are wonderful but the interest is in the higher-order reasoning that they do (yes they do that I know)
      • Society of Mind – maybe a bad set of tools for building a mind, but what about analyzing one?
      • Something weirdly Platonic about Minsky theory. His machines have an ideal nature to them, like, it doesn't matter about neurons or complexity, the behavior at a symbolic (aka cognitive) layer is what is important.