Push Singh (2003), Computing and Informatics 22(6), 521–543 (revised manuscript received 4 November 2003) — journal article page https://www.cai.sk/ojs/index.php/cai/article/view/467, open access
Key takeaways
- History. The theory began in Minsky and Papert’s discussions at the MIT AI Lab in the early 1970s. Their hand-eye “copy-demo” robot showed them that “no single method ever worked well by itself”. Frames with attached procedures (1974) were “the ancestor of the concept of an agent”. Minsky and Papert drafted a book in 1976, the unpublished “Brazil” drafts, and abandoned it. The theory then appeared piecemeal in Minsky’s 1977 and 1980 papers.
- Mechanisms. Agents and agencies; K-lines, which turn on a remembered set of agents; nemes, which represent, and nomes, which control, with polynemes, micronemes, isonomes, pronomes and paranomes. Frames, frame-arrays and transframes. Difference-engines. Censors and suppressors. A- and B-brains. Protospecialists, predestined learning, and learning from attachment figures. Papert’s Principle and the Principle of Non-Compromise.
- Communication without a shared code. Agents are “too simple to understand the languages of most other agents”. They communicate by arousing other agents through K-lines. They use randomly connected “bus” wires whose meanings are learned, after Calvin Mooers’s superimposed coding. They use paranomes, which update parallel representations so that no message is sent. Ambiguity is tolerated, not removed.
- Since 1986. The theory resembles probabilistic graphical models (its ring-closing recognition is like belief propagation), case-based reasoning (derivational analogy “came directly from the K-line theory”), blackboard systems, Cyc’s microtheories, Soar (chunking resembles K-lines) and multiagent systems. Each lacks something the theory needs: nomes, managerial knowledge, or reflection. “The Society of Mind theory has not” seen a series of implementations.
Standing in the record
Filed on 2026-10-01 at the owner’s request for Minsky’s Society of Mind programme. The brief asked for it as a sympathetic exposition by Minsky’s student, and that is what it is. Singh says he was given the unpublished 1976 drafts and discussed the history with Minsky and Papert. The draft of The Emotion Machine (LIT-432) cites this paper for how the two books relate.
It is the record’s best reachable account of The Society of Mind (LIT-429), which is unread here. Read it for the map, and for the history it quotes. Do not read it for evidence that the theory works: it says the theory has never been built. No anthology topic holds it. Its survey of AI techniques is a comparison of architectures, not advice for machine-learning practice.
The journal page dates the issue only by year, and Crossref has no record
of the article. The PDF’s “revised manuscript received 4 November 2003”
means the issue, no. 6 and the last of the volume, appeared no earlier
than November 2003. published: is the year placeholder the record uses
when no month is given. The journal site’s “Published 2012-02-20” is the
date the archive was uploaded, not the first appearance.
The brief’s citation is right in every detail checked: author, title, journal, volume 22, issue 6, pages 521–543, 2003, and open access. There is no DOI to check.