Generative vs. Evaluative Attention
Generative attention is the mode of creative production characterized by uncertainty, struggle, and dwelling—the writer facing the blank page, the artist before the empty canvas, the scientist formulating a question no one has asked. It is slow, often painful, and apparently unproductive for extended periods. Evaluative attention is the mode of selection and refinement—choosing among options, comparing alternatives, curating the best from a generated set. It is faster, less uncertain, and immediately productive. Both modes are essential to creative work. The crisis Citton diagnoses is that AI tools systematically cultivate evaluative attention while degrading generative attention's habitat. When the blank page can be instantly filled with five AI-generated options, the generative mode—with its discomfort, its slowness, its requirement for dwelling in not-knowing—loses its environmental support. The creator becomes an excellent curator and a diminished generator, not through any failure of talent but through thousands of micro-decisions to consult the tool rather than sit with the difficulty.
In the AI Story
The generative-evaluative distinction maps onto multiple existing frameworks: Kahneman's System 1 vs. System 2 (automatic vs. deliberate), Peter Elbow's first-order vs. second-order thinking (generating vs. editing), Donald Schön's reflection-in-action vs. reflection-on-action. Citton's contribution is the ecological reframing: these are not merely cognitive strategies but modes of attention shaped by environmental affordances. An environment that continuously offers generated options cultivates evaluative attention—trains it, rewards it, makes it the path of least resistance. An environment that enforces emptiness (the blank page with no AI assistance) cultivates generative attention—forces it, sustains it, makes it the only available path. Which mode a practitioner develops depends less on individual choice than on the structured affordances of the environment they work in daily.
The displacement of generative by evaluative attention produces a specific pathology Citton calls creative sterility through abundance. The AI-augmented creator has access to more options, more variations, more possibilities than any previous generation. Every imaginable style, every conceivable approach, every possible direction can be generated on request. And yet the creator's own imaginative capacity—the ability to envision something no prompt would request because the creator didn't know it was possible to want—atrophies. The abundance is real. The sterility is real. They are the same phenomenon viewed from different angles. Abundance eliminates the scarcity (of options, of alternatives) that forced the generative mode to operate. Sterility is the long-term cost of that elimination: a creative capacity that can select brilliantly among the generated but can no longer generate from its own resources the kind of vision that no algorithm would offer.
The cultivation of generative attention in AI-saturated environments requires what Citton calls protected generative time—deliberate periods when the AI tool is unavailable and the practitioner must work from internal resources alone. The practice is not romantic refusal of the tool—it is ecological maintenance. The farmer who understands soil health rotates crops not because the high-yield crop is bad but because continuous planting depletes the soil. The creator who understands attentional ecology rotates modes not because AI assistance is bad but because continuous evaluative processing depletes the generative capacity that evaluation depends on. One hour of unaided struggle at the start of each work session. One day per week without AI access. One project per quarter completed entirely without generative tools. Each practice is a fallow field, apparently unproductive, actually essential.
Origin
The generative-evaluative distinction is implicit in every model of the creative process that includes separate phases (divergence and convergence, ideation and editing, generation and selection). Citton's innovation is to frame these not as sequential phases of a single process but as distinct modes of attention—each requiring different environmental conditions, each vulnerable to displacement when the environment shifts. The framing transforms the question from 'how do I do both?' (individual skill problem) to 'how do I create conditions that support both?' (environmental design problem). The shift is crucial because it reveals that the creator struggling to maintain generative capacity in an AI environment is not failing individually—the environment is failing ecologically.
Key Ideas
Dwelling vs. scanning. Generative attention requires sustained dwelling with difficulty; evaluative attention requires rapid scanning of options—opposite temporal structures, opposite cognitive demands.
AI as option-generator. Every AI tool, by offering alternatives, shifts the cognitive task from generation to evaluation—an affordance that is individually productive and ecologically destructive.
Capacity atrophy. Generative attention is a muscle—it strengthens with use, atrophies without—and AI environments systematically eliminate the resistance exercise that keeps it strong.
Protected generative time. Sustaining generative capacity requires scheduled AI absence—not refusal of the tool but rotation between modes, allowing each the conditions it needs.
Appears in the Orange Pill Cycle
Further reading
- Yves Citton, The Ecology of Attention (Polity, 2017)
- Peter Elbow, Writing Without Teachers (Oxford, 1973)
- Donald Schön, The Reflective Practitioner (Basic Books, 1983)
- Mihaly Csikszentmihalyi, Creativity (Harper Perennial, 1996)