Adjacent Possible (Dawkins Framework)

In the AI Story

Kauffman, a complexity theorist at the Santa Fe Institute, developed the concept as part of his broader framework on self-organization and the origins of order. Dawkins, while skeptical of Kauffman's claim that self-organization rivals natural selection in explanatory power, absorbed the adjacent possible as a useful constraint on evolutionary possibility space. The Blind Watchmaker and Climbing Mount Improbable (1996) both emphasize that evolution cannot leap — it must climb gradually, and the landscape's topology determines what is climbable from any given position.

The language interface transition of 2024–2025 was adjacent-possible dynamics in technological evolution. The components existed independently: large language models trained on internet-scale corpora, transformer architectures, sufficient compute infrastructure, populations familiar with text-based interaction. The step from command-driven AI to conversation-driven AI combined existing elements into a new configuration — small architectural change, enormous functional consequence. Multiple organizations (OpenAI, Anthropic, Google) took the step near-simultaneously, not through coordination but because they stood at the same frontier.

The adjacent possible constrains individual builders as much as it constrains species. A builder's imagination is bounded by the landscape of configurations she can conceive, and that landscape is determined by her accumulated knowledge, experience, and exposure to diverse domains. Segal's Trivandrum engineers with deep expertise produced richer outputs than junior engineers using the same tool, because the senior engineers' adjacent possible was larger — they could specify configurations the juniors could not imagine. The tool expanded what each could build, but it did not expand what each could conceive of building. Imagination remains the binding constraint, and imagination is a function of biographical accumulation that no tool can shortcut.

Origin

Stuart Kauffman introduced the adjacent possible in Investigations (2000) and Reinventing the Sacred (2008), though the concept appears implicitly in his earlier work on autocatalytic sets and the origins of life. The core idea: the universe's possibility space is not static. Each actualized configuration opens access to new configurations that were previously unreachable. The biosphere expands not by filling a predetermined space but by generating the space it fills — each innovation is both an achievement and a platform for further innovation. Dawkins incorporated the concept without adopting Kauffman's broader complexity framework, using it to clarify how cumulative selection explores possibility space through incremental steps.

Key Ideas

One step from here. Evolution proceeds through configurations reachable in single mutations from the current genotype — distant possibilities remain inaccessible until intermediate steps are taken.

Landscape expands as explored. Each innovation opens territories that were not previously accessible — the printing press made mass literacy adjacent-possible, which made public opinion adjacent-possible, which made democracy adjacent-possible.

Simultaneous invention explained. When the adjacent possible is the same for multiple actors, they independently take the same step — not coincidence but convergence on the frontier everyone faces.

Individual imagination bounded. The builder's adjacent possible is determined by accumulated knowledge — the tool expands what can be built but not what can be conceived.

Small steps, large consequences. Some positions in possibility space are saddle points where incremental change produces phase transitions — the language interface was such a point.