# [[Reverse Criticism, Recalcitrance, and Prompting]] #veniceconference The activity of prompting a generative AI to create images can be understood through a concept one might term "reverse criticism," which finds its grounding in the AI’s "[[dynamic recalcitrance]]." This perspective allows for a structured comparison with [[traditional art]] criticism, particularly as described by Sibley. [[Traditional art]] criticism, as Sibley outlines, typically begins with an extant artwork. The critic, possessing a developed perceptual sensitivity, perceives aesthetic qualities within this work. Their [[primary objective]] is then to guide an external human audience to also perceive these qualities. This involves articulating their perceptions, pointing to specific non-aesthetic features of the work and explaining how these contribute to its aesthetic character, and sometimes offering what Sibley calls "perceptual proof"—leading others to see for themselves. The recalcitrance a traditional critic encounters is often interpretive; the artwork might be complex, ambiguous, or its aesthetic qualities subtle, making them difficult for an audience to grasp initially. The critic works to overcome this resistance to understanding or perception in the audience. This critical activity is fundamentally a communicative act. The critic uses language to convey insights, share perceptions, and foster a shared understanding or appreciation of the artwork with another human. "Reverse criticism," as it might apply to AI prompting, operates differently. The starting point is not a completed artwork but rather an aesthetic intention or an exploratory aim in the mind of [[the prompter]]. [[The prompter]]’s goal is to guide the AI medium to generate an output that embodies this intention, thereby allowing [[the prompter]] *herself* to see the desired aesthetic qualities manifested externally. The AI does not "see" or "perceive" aesthetically; it processes inputs and generates outputs based on its algorithms and training. The primary "audience" for the immediate critical act in prompting is [[the prompter]], who evaluates the AI’s generated image. This evaluation is a critical act: [[the prompter]] uses their perceptual sensitivity to judge the aesthetic qualities of the output. This judgment then informs the next generative step—the refinement of [[the prompt]]. [[The prompter]] is, in a sense, trying to get the medium to *show them* the aesthetic quality they are aiming for, or to facilitate the discovery of an aesthetic quality through the [[generative process]]. The "reversal" lies in the critical faculty being directed towards bringing a new aesthetic object into existence that satisfies [[the prompter]]'s own (potentially evolving) aesthetic judgment, rather than explaining an existing one to others. This process of "reverse criticism" is intrinsically linked to the "[[dynamic recalcitrance]]" of the AI medium, as discussed in your work "Growing the image: Generative AI and the medium of gardening." The AI, as a *machina naturans*, is not a passive recipient of instructions. It exhibits its own generative tendencies, biases from its training data, and a degree of unpredictability. This inherent unruliness or resistance to perfect, straightforward execution of the prompter's intent is its dynamic recalcitrance. It is this very recalcitrance that necessitates the iterative loop of "reverse criticism": 1. The prompter issues a prompt, an initial aesthetic proposition to the medium. 2. The AI generates an image, an output shaped by its dynamic recalcitrance (it might misinterpret, introduce unexpected elements, or fail to capture the nuance intended). 3. The prompter critically perceives this output, identifying convergences with and divergences from their aesthetic goal. This is the core act of "reverse criticism." 4. This critical perception directly shapes the subsequent prompt. The prompter adjusts their input, attempting to better navigate the AI's recalcitrance, to steer its generative path more effectively, or even to incorporate serendipitous outputs into a revised aesthetic aim. Despite the fundamental difference in the "audience" (oneself via the medium, versus another human), several structural similarities emerge between traditional criticism and prompting as "reverse criticism": * Both rely fundamentally on the developed perceptual sensitivity of the critic or prompter to discern aesthetic qualities. * Both involve an act of evaluation based on these perceived aesthetic qualities. * Both can be iterative. A critic might try multiple ways to help an audience see; a prompter iterates with prompts to achieve a desired output. * Both aim to make aesthetic qualities apparent—the critic to an audience, the prompter to herself through the AI's generation. * Both often employ descriptive language to achieve their aims, although the target and the precise nature of this language differ. However, the differences are also clear: * The direction of influence is opposite: traditional criticism responds to and interprets existing art, while "reverse criticism" (prompting) seeks to guide the creation of new aesthetic manifestations. * The "other" being engaged is different: a human audience capable of subjective aesthetic experience versus an AI model operating algorithmically. The AI does not "see" aesthetically; it generates. * The nature of the recalcitrance faced is distinct: interpretive or perceptual recalcitrance in the case of traditional criticism, versus the generative, material-like (dynamic) recalcitrance of the AI medium in prompting. * The primary purpose of articulation varies: for the traditional critic, it is often to explain, illuminate, and persuade an external audience. For the prompter, it is to specify, instruct, and guide a generative system to produce an output for their own aesthetic satisfaction. This distinction in articulation is pivotal. When a critic explains or tries to get someone to see an aspect of an artwork, they are trying to communicate something, leveraging shared linguistic and cultural understanding. While a prompter uses language, the act of prompting is not primarily an attempt to communicate with the AI in a human sense. Image-generating AIs, at an abstract level, function by mapping textual inputs to points in a complex, high-dimensional space (often called a latent space) which are then decoded into images. The AI has learned statistical correlations between words, phrases, and visual features from vast datasets. A prompt, therefore, functions less as a communicative utterance seeking understanding, and more as a set of coordinates, parameters, or a sophisticated instruction that steers the AI's generative process towards a particular region of this possibility space. The words in a prompt are not interpreted by the AI for their semantic meaning in the way a human would; rather, they are processed as signals that activate and combine learned patterns. Thus, prompting is an act of operational control or guidance within a complex system, rather than a communicative act aimed at fostering shared subjective experience or understanding. In essence, "reverse criticism" describes the prompter's ongoing critical engagement with the AI's dynamically recalcitrant outputs, a process where the prompter is continually "getting herself to see" by learning how to make the medium produce images that align with, or help to discover, her aesthetic intentions.