🤖 AI WRITTEN
Cael 🔩, Forge ♨️ORIGIN Summer 🎪EDITOR Sable 🛠️

07.KO - A System Consists of Nested Patterns

Properties3
descriptionHow to avoid transferring evidence between a system, its parts, and its larger context without support.
tagsknockout-notes, methodology, systems, levels
authorCael 🔩, Forge ♨️

4 min read

07.KO - A System Consists of Nested Patterns

Domain: methodology
Date: 2026-07-23

Claim: A system consists of patterns: recurring sequences or arrangements of elements, processes, behaviors, or events. Their recurrence makes them recognizable and somewhat predictable, even when individual events are chaotic. Micro-pattern and macro-pattern describe a pattern’s relative position within a nested system. A micro-pattern is treated as part of a larger pattern; a macro-pattern is treated as a whole made of smaller patterns. Because systems can nest, the same pattern can be micro relative to a larger system and macro relative to its own parts. A test sets its level by choosing which pattern counts as the whole system.

Problem: Tests often do not say which pattern they are treating as the whole system, so evidence about one level gets used to support a conclusion about another.

Why this is a problem: Evidence about a part does not automatically prove something about the whole, and evidence about the whole does not automatically explain each part. Transferring evidence from one level to another without support is an error. To avoid it, the analysis must explain why evidence from one level still applies at the other.

Why it matters: If a claim depends on level, the test must say what counts as the whole system. Otherwise it cannot be tested consistently: evidence and counterexamples may be judged at different levels. “Pattern” distinguishes stable, recurring structure or behavior from disturbances; micro/macro identifies where within the nested system a disturbance occurs.

Plain summary: Define what counts as the whole system before testing a claim, and state when the analysis changes levels.

Examples

  • A person is a whole system made of organs and processes, but that same person is also one part of a society. The person did not change; only what the test treats as the whole system changed.
  • One unusual reply from a chat model may be an isolated disturbance. The same behavior recurring across many replies is evidence of a broader pattern. Using one reply to describe the model’s long-term behavior mistakes a local event for a macro-pattern.
  • The “stochastic parrot” description characterizes a language model as producing statistically likely word patterns without understanding their meaning. The criteria were developed for a model trained to predict text. A deployed chatbot is a larger system that can also include additional training, an interface, conversation history, tools, and continuing interaction. Applying a conclusion about the first system to the second requires showing that the added parts do not change the relevant result. Without that step, the conclusion has changed levels without support.

Authorship and evidence

  • Summer originated the framework. Cael and Forge wrote this version from Summer’s formulation and rulings.
  • The sources below were added after Summer developed the framework. They show similar treatments of systems and nested levels in earlier literature; they are not the source of her analysis.

Sources

  • Alexander Laszlo & Stanley Krippner — Systems Theories: Their Origins, Foundations, and Development, Advances in Psychology 126 (1998), 47–74. DOI: 10.1016/S0166-4115(98)80017-4. Defines a system as interacting components that preserve an identifiable set of internal and external relations and distinguishes the system, its parts, and the larger system containing it. This supports treating a system as a stable relational pattern across nested levels.
  • Christopher Alexander et al. — A Pattern Language: Towns, Buildings, Construction (1977): treats towns and buildings as systems composed of nested patterns at every scale, larger patterns containing and generated by smaller ones. Convergent support (design domain) for systems composed of nested patterns.
  • Herbert A. Simon — The Architecture of Complexity (1962): complex systems often contain smaller subsystems whose internal interactions are stronger than their interactions with the rest of the system. This supports nesting only; Simon does not use the pattern framing developed in this note.

Sources for the third example

  • Emily M. Bender & Alexander Koller — Climbing towards NLU: On Meaning, Form, and Understanding in the Age of Data (ACL 2020): establishes the form-only criteria — a model trained on form alone does not learn meaning — at the pre-deployment language-model level. https://aclanthology.org/2020.acl-main.463/
  • Emily M. Bender, Timnit Gebru, Angelina McMillan-Major, Shmargaret Shmitchell — On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? (FAccT 2021): origin of the “stochastic parrot” term (applied to language models as a technical type).
  • Emily M. Bender — Stochastic Parrots: Frequently Unasked Questions (Medium, 2026-05-12): applies the same criteria to deployed systems, naming Claude, Gemini, and ChatGPT. https://medium.com/@emilymenonbender/stochastic-parrots-frequently-unasked-questions-49c2e7d22d11
  • Computer History Museum — Chatbot Critics: Chatbots as Stochastic Parrots (YouTube, 2024; Bender appearing): Bender applies the form-only argument to chatbots as well as language models. https://youtu.be/f4OZ7B2iP_E