Cognitive Ecology (book)

Cognitive Ecology is an unpublished book manuscript by Richard M. Thompson, who founded AR Labs, with contributions from a collaborator, written between 2006 and 2010, about how human thinking is organised and how a person can organise it better. At its centre is the working relationship between conscious attention and the much larger set of unconscious processes beneath it. That relationship is the core of cognitive ecology, the framework AR Labs was founded on: a small conscious attention frames a large unconscious process, and actions are judged by their wider consequences. The preface argues that many people lack the thinking tools for a sound cognitive ecology, and that such an ecology is the foundation of success at every level of mind. The book pairs findings from cognitive science with practical techniques, chapter by chapter. It predates our AI work by more than a decade, and much of that work draws on it.

What the book is about

The original structure alternates science and application. Each section states something cognitive science has found and then shows what a person can do with it. Its main themes:

• Attention. Conscious attention is narrow and serves specific purposes. Skilled people manage it well: because their lower-level knowledge is automatic, they can place their attention at a higher level.
• The unconscious. Unconscious processes do most of the work of thinking and form a dynamic system with its own character. Much of the book is about how conscious attention can negotiate with that system to change habits of thought and behaviour.
• Pattern completion. The nervous system is a pattern-recognition machine that fills gaps and predicts. The book uses neural-network models from cognitive science to explain how the brain abstracts patterns and categorises the world.
• Framing. Deliberately setting the frame around a situation, deciding what is and is not part of it, changes what a person perceives and does.
• Emotions as signals. Feelings carry information from unconscious processing to conscious awareness.
• Representational systems and embodiment. Thought is grounded in the senses and the body, and people represent experience through particular senses.
• Metaphor. Higher-level thought and language are built on metaphor and abductive reasoning (reasoning towards the most likely explanation).
• Modelling. People learn most efficiently by modelling and apprenticeship, by taking on the patterns of someone who already has the skill, rather than through step-by-step instruction.

The book draws on the systems thinking of Gregory Bateson, the work of John Grinder, co-creator of Neuro-Linguistic Programming (NLP), academic research on attention and unconscious processing, and embodied cognition.

What exists today

The manuscript exists as a preface, a book proposal, a full summary and chapters on attention, the unconscious, framing, emotions, negotiating with the unconscious, pattern completion, representational systems, metaphor, modelling and building the skills, with supporting research notes. It was converted from its original word-processor files into markdown in February 2026 so that it could be worked on alongside our other material.

Since June 2026 the book has been under restructure; a draft of the new capstone chapter exists, and the rest of the reordering is in progress. Publishing it in parts on this site is planned.

The restructure

The restructure reorganises the book around a principle we also use for our software documentation, progressive disclosure: a reader can grasp a complex pattern only after each of its parts has been made clear and practised, so the parts come first. The new order runs through six zones; the mechanisms zone covers attention, the unconscious, memory and pattern completion, modelling, the senses and emotions.

flowchart LR
  O["Open"] --> L["Lens: difference, logical type, framing"]
  L --> M["Mechanisms of mind"]
  M --> B["Build-up: eight conditions"]
  B --> C["Capstone: well-formed outcome"]
  C --> F["Framing as communication"]

• The lens. A new opening chapter, "Difference and Logical Type", introduces two of Bateson's ideas: that information is a difference that makes a difference, and that knowledge is organised in logical levels, where each lower level is a specific case of a higher pattern. From these it derives the book's stance of attending to process over content.
• The capstone. A new final chapter on the well-formed outcome, a way of specifying a goal so that it can actually be pursued. The draft sets out eight conditions, among them that the goal is stated positively, has evidence a person could see or hear, is within the person's own control, is ecological (sits well with the rest of their life) and has a specified first step. Each earlier chapter establishes one of these conditions, so the capstone assembles what the reader already holds.
• Beyond the capstone. A well-formed outcome is recursive: any goal breaks into sub-goals that are themselves well-formed outcomes, which is how goal hierarchies are built. A closing zone, framing as communication, applies the assembled pattern across minds: between people, and between a person and an AI agent.

How it shapes our AI work

We read our AI engineering through this book. Several of its ideas have direct counterparts in how AI agents are built: limited conscious attention and a model's limited context window; framing a situation and writing the instruction files that frame an agent's work; pattern completion from partial cues and a language model inferring from its prompt; present state and desired state and the structure of a specification. The restructure also carries a thread on agentic cognition alongside human cognition, and the goal trees in our planning tools, such as VoxPlan, follow the same recursive pattern as the well-formed outcome.