AI as Combinational Engine
When Boden's taxonomy is applied to contemporary AI systems, a specific pattern emerges. Large language models excel at exploratory creativity within formally definable spaces (chess, code, mathematical proof), perform combinational creativity at extraordinary range (connecting concepts across their training corpus), and have not demonstrated transformational creativity in any convincing form. Of these, the combinational capacity is the one that most directly enables the human-AI creative partnership Edo Segal describes in The Orange Pill. The machine surfaces connections; the human evaluates which connections matter. The quality of the partnership depends on both — the machine's range supplying candidate combinations the human could not generate alone, the human's judgment selecting which are genuinely illuminating from the many that are merely plausible.
In the AI Story
The combinational engine framing cuts through much of the confused AI-creativity discourse. It does not claim machines are conscious or that they understand their outputs; it does not deny that they produce genuinely novel connections. It specifies what they do — generate unfamiliar connections across vast ranges — and what they do not do — evaluate which connections deserve attention.
The asymmetry matters practically. A builder working with Claude can generate ten plausible-looking analogies in thirty seconds. Without evaluative discipline, she will accept the first one that sounds right. With evaluative discipline — drawing on domain depth, cross-referencing to primary sources, testing the analogy against cases it should and should not explain — she can distinguish the one genuine insight from the nine superficial ones.
The Deleuze failure in The Orange Pill is the canonical example of combinational engine failure without evaluation. Claude produced a connection between Csikszentmihalyi's flow state and Deleuze's concept of smooth space that sounded like insight and collapsed on inspection. The connection was plausibly combinational; it was not actually accurate. The builder's subsequent disciplined verification — checking the reference, discovering the misattribution — is exactly the evaluative work the machine cannot perform.
The framing has implications for how organizations should structure AI-augmented work. If the machine is a combinational engine, the human role is evaluation. Training, mentoring, and quality assurance should focus not on learning to prompt more effectively (which the machine will soon do better) but on developing the domain depth, taste, and verification discipline that makes evaluation possible.
Origin
The combinational engine framing emerges directly from applying Boden's taxonomy to large language model behavior. Boden's earlier work had anticipated that computational systems would be strongest at exploration and combination, weakest at transformation. The empirical evidence of 2022-2026 AI systems has largely confirmed this prediction.
Key Ideas
Machine generates, human evaluates. The asymmetry is not a limitation to be overcome but a structural feature of the partnership that makes it work.
Range is the machine's contribution. Training corpus span enables connections across domains no individual mind can traverse.
Judgment is the human contribution. Distinguishing illuminating connections from merely plausible ones requires depth, taste, and verification discipline.
The Deleuze failure is diagnostic. Plausible-sounding combinations that collapse under examination reveal the cost of accepting combinational output without evaluation.
Organizational implication. AI-augmented work should cultivate evaluative capacity, not prompting skill — the machine will soon prompt itself.
Debates & Critiques
Critics argue the framing undersells what machines do — that sufficient combinational power, applied recursively, becomes functionally equivalent to transformation. Boden's response: the difference between combining existing elements and rewriting the rules of the domain is categorical, not a matter of degree. Until machines demonstrate the evaluative capacity to recognize framework inadequacy, the combinational engine framing holds.
Appears in the Orange Pill Cycle
Further reading
- Margaret Boden, The Creative Mind, Chapters 3-4
- Margaret Boden, Creativity and Art, Chapter 4