Margaret Boden

In the [YOU] on AI Field Guide

The cycle that began with [YOU] on AI asks what it would mean to see the machine clearly, without the narcotics of hype or despair. Boden is the cycle’s most precise diagnostician of what that clear-sightedness requires. When the book describes the collapse of the imagination-to-artifact ratio, Boden’s taxonomy supplies the analytical vocabulary for understanding what that collapse actually means: it is the collapse of the translation cost for exploratory and combinational creativity, not for transformational creativity. The machine has become extraordinarily good at searching conceptual spaces that humans define and at bridging between spaces that training data has made accessible. What it cannot do, on the evidence of her framework, is recognize that a space is inadequate and imagine a different one.

Her lens reframes every celebration and every anxiety the cycle encounters. The engineer in Trivandrum who built a complete frontend feature in two days using Claude was performing exploratory creativity within a space she had always been able to imagine but could never previously traverse—the machine searched it on her behalf. When Claude connected punctuated equilibrium from evolutionary biology to technology adoption curves, it was performing combinational creativity at a scale that exceeds any individual biographical range. But when Segal struggled to find the pivot in his chapter on Byung-Chul Han—unable to find the right framework for the argument he needed to make—he was approaching the preliminary gesture of transformational creativity: the recognition that the available space was inadequate. That recognition was his, not the machine’s.

Boden also supplies the cycle’s most useful analysis of what is genuinely lost when AI removes friction from creative work. Each hour of struggle within a conceptual space deposits what the cycle calls geological understanding. The ten minutes of formative surprise embedded in four hours of plumbing that Claude eliminated were not lost because they were embedded in tedium; they were lost because they were the deposition events through which the practitioner built her internal map of the space. Without personal exploration of a space, the practitioner cannot evaluate what a machine’s search of that space finds. She can receive the commodity but not the evaluation. And evaluation—at the functional, aesthetic, and directional levels—is the distinctly human contribution the cycle ultimately argues for.

Origin

Margaret Ann Boden was born in London in 1936 and read Natural Sciences and Medicine at Newnham College, Cambridge, before switching to philosophy. She earned her doctorate from Harvard in 1959 under the supervision of the philosopher Roderick Chisholm and spent her entire academic career at the University of Sussex, which she joined in 1965 and where she became one of the founding figures of what would eventually be called cognitive science. She retired as Research Professor of Cognitive Science in 2018 at the age of eighty-two, having published more than a dozen books and hundreds of papers across philosophy, psychology, computer science, and artificial intelligence.

Her entry into what would become her central subject came through a deep engagement with the work of early AI researchers: Alan Turing, whose test she always regarded as more philosophically interesting than its critics allowed; Aaron Sloman, her Sussex colleague; and the creators of programs like AARON, Harold Cohen’s machine that generated original drawings according to explicitly coded rules, and EMI, David Cope’s system that composed in the styles of historical composers. What she observed in these programs—their genuine capacity to search within defined spaces, their genuine inability to modify the spaces that defined them—became the empirical foundation for the taxonomy she was simultaneously developing on philosophical grounds. The programs were not metaphors. They were data.

The major statement of her position was The Creative Mind: Myths and Mechanisms, first published in 1990 and substantially revised in 2004. But her engagement with the question never ceased; her 2016 collection Computer Models of Mind and her ongoing commentary on the large language model moment kept her framework in productive contact with developments she had not originally anticipated. She understood, with characteristic precision, that the scale at which LLMs bridge conceptual spaces was genuinely new—and that it was still, structurally, combinational rather than transformational.

Key Ideas

The three-mode taxonomy. Boden’s foundational distinction separates creative acts by the cognitive operation they perform on a conceptual space. Exploratory creativity searches within an established space; combinational creativity bridges between spaces; transformational creativity changes the rules that define a space. The three are not on a continuum. They represent qualitatively different cognitive operations, each requiring different capacities and each differently amenable to machine replication. The taxonomy’s value is diagnostic: it tells you not whether a creative act is impressive but what it required of the mind that performed it.

Conceptual spaces and their rules. Every creative act takes place within a conceptual space—a structured set of possibilities defined by generative rules that specify what is permissible within the domain. The rules of tonal harmony, the syntax of a programming language, the conventions of sonnet form: each defines a space and constrains what counts as a creative exploration of it. The space is not merely a metaphor; it is the formal object within which exploration, combination, and transformation each operate differently. A machine that searches a space more thoroughly than any human does not thereby change the space; that change requires a qualitatively different cognitive act.

P-creativity and H-creativity. Boden’s distinction between psychological creativity (new to the individual) and historical creativity (new to human history) is the most practically important refinement of her taxonomy for understanding AI. AI tools dramatically expand P-creative possibility: the engineer who could never explore frontend development can now traverse it; the writer who could never range across evolutionary biology and economics simultaneously can now make those connections. Whether this democratization of P-creativity generates H-creative breakthroughs depends on whether the expanded combinational range produces genuinely unprecedented connections that survive rigorous evaluation—a condition the machine can assist but not guarantee.

The evaluation gap. Boden’s framework implies a three-level hierarchy of evaluation: functional (does it work?), aesthetic (is it excellent?), and directional (should this exist at all?). Each level requires forms of understanding that only personal exploration of a conceptual space can build. The machine generates candidates; the human evaluates them. The quality of the collaboration depends entirely on the depth of the human’s evaluative capacity—which is built through the specific, friction-rich experience of personal exploration that devices are structurally designed to eliminate. This is the evaluation gap: the growing distance between the machine’s generative range and the human’s evaluative reach.

Generative systems and self-reflection. What AARON and Claude share, and what distinguishes both from transformational creativity, is the structural relationship to their own generative system. Both operate within systems they cannot examine and modify. AARON could not recognize that Cohen’s rules were inadequate for a new expressive purpose. Claude does not reflect on its own pattern-matching process and ask whether a different process would produce better connections. Scale makes the system more powerful within this structure. It does not change the structure.