Educating the Gaze in the Age of Saturation: Visual Education, Generative AI, and Critical Learning Strategies ^†

Abstract

The cognitive environment in which students are immersed today is characterised by high exposure to images and content generated, selected and modulated by algorithmic systems. Visual education, therefore, runs the risk of becoming an accessory function of the system, serving to capture attention rather than to develop a critical eye; but visual education can also be used to question narrative linearity, destabilise established representations, and activate reflective processes. Artificial intelligence is a cognitive agent, and its use in visual teaching can help to deconstruct interpretative automatisms and open up spaces for more attentive and conscious learning.

1. Introduction

The contemporary condition is characterised by immersion in a continuous flow of images that creates a state of ‘visual saturation’. This is not simply a quantitative intensification attributable to changing communication dynamics (use of social media and generative artificial intelligence tools), but a qualitative transformation of the perceptual experience that redefines the cognitive environment as a whole. For this reason, it is necessary to analyse the phenomenon in order to understand whether saturation is progressively dissolving the ability to pay selective attention and eroding the critical distance necessary for any learning process.

Numerous neuroscientific studies have documented how prolonged exposure to digital media structurally alters cognitive abilities and that excessive use of digital technology can alter brain structure and function, leading to a range of cognitive deficits: internet addiction has been associated with a reduction in grey matter density in the frontal cortex of the brain, which is responsible for decision-making and impulse control [1].

In this condition, the image no longer has a meditative, interruptive function—that moment of suspension that Walter Benjamin recognised as the subversive potential of aesthetic experience [2]—but instead becomes an element of an informational flow that never generates any reflective pause, fuelling a form of accelerated perceptual consumption. What we are witnessing is an unprecedented transformation in which a mode of relating to the visible is emerging that structurally compromises the possibilities for critical processing.

In 2024, the main generative artificial intelligence systems produce around 34 million images per day [3], with an annual growth of 85% in the use of such tools. Most users have significant difficulties in recognising algorithmically generated images. As documented by recent research [4], neither humans nor AI tools can reliably distinguish between real and fake images. The inability to distinguish between the two is becoming the norm, rather than the exception.

The growing difficulty in distinguishing authentic images from algorithmically generated ones has been documented by the MIT Media Lab through the DetectFakes project, an online research environment that presented thousands of participants with authentic and deepfake videos, asking them to identify which ones were manipulated. The results documented that even people trained to recognise deepfakes show progressively decreasing accuracy rates [5].

2. From Testimony to Performativity: The Transformation of the Photographic Regime

The difficulty in distinguishing between true and false reveals a structural transformation in our relationship with the visual universe: while the paradigm of the photographic image was once based on testimony (‘this is what happened in the past’), today it takes on a performative connotation, indicating a worldview that is entirely manipulable and recontextualisable (‘this is what I consider real’).

In other words, in the testimonial paradigm that dominated for over a century, photography was the imprint of reality, a certificate of presence, the guarantee of what Roland Barthes called the ‘it was’ [6]. This relationship between the subject and the image formed the basis for the documentary, journalistic, and evidential uses of photography. In the performative paradigm, on the contrary, the image indicates that ‘I affirm this’: it is a linguistic act, a gesture that seeks to orient and convince, and does not have a defined and solid relationship with reality and facts but is always an interpretation of them.

The ability to distinguish between true and false, therefore, marks the end of photography as testimony and the beginning of the image as a project. The image emancipates itself from reality to take on a generative function: it is no longer the mirror of something that has been, but the matrix of what could be.

The change manifests itself through the dissolution of the intrinsic trust that human beings have in the image. The same visual content can be recounted in incompatible ways, because it can become a tool for narratives, reconstructions, and contextualisations that activate different interpretative frames. Take the image of a crowd of people in a square: depending on the portion of reality that the visual content captures, it is possible to construct a different narrative about why people were in that place, their attitude. The narrative device manipulates the image itself and can convince human beings that they have observed a certain phenomenon, simply because a certain story has been constructed around the visual content.

This is not an entirely new phenomenon. Hannah Arendt identified a similar anthropological break in The Origins of Totalitarianism, observing that the masses cease to trust their own sensory experience in favour of ideologically oriented narrative constructs [7]. In totalitarian societies, this suspension of trust in the visible was imposed through a centralised propaganda apparatus. In the post-photographic condition, it emerges as the systemic effect of a multiplicity of agents: the proliferation of synthetic images, the erosion of the mediating institutions that guaranteed chains of trust, and the fragmentation of public spaces for collective verification.

Arendt’s analysis allows us to grasp how deeply pedagogical the issue is, because it questions us about the conditions of possibility for the constitution of a common world when shared trust in the evidence of the visible fails. Furthermore, it confronts us with the need to ask ourselves how educational practices can respond to this crisis without falling into either a naïve realism that ignores the ever-mediated nature of perceptual experience, or a nihilistic relativism that dissolves any possibility of truth.

3. The Post-Photographic Condition: Plasticity, Circularity, Performativity

Take, for example, a close-up image of two hands holding a luminous object: it could be a galaxy, an iridescent bubble or a technological device. The observer constructs narratives based on their desires, expectations and worldview: they could describe the image as a scientific discovery, a childlike wonder, a technological innovation and in many other ways. The detail allows us to construct a story, but the context can change everything. This suggests that every shot is already a montage, i.e., an exclusion of what is outside the frame. There is no such thing as a neutral representation, just as there is no such thing as pure testimony: every image is always already an interpretation.

The issue of deepfakes takes this logic to its extreme consequences, since a fake image of a political leader making controversial statements, even when its algorithmic nature is known, still produces real political effects. The problem, therefore, is not the deception itself, but the performative effectiveness of the image, which is independent of its correspondence with a real referent.

The false image circulates, generates emotional reactions and shapes opinions. Even after it has been exposed, its effects remain in the collective memory. The shift from the testimonial to the performative paradigm configures a ‘post-photographic condition’, i.e., a state in which the image has abandoned any claim to indexical guarantee to become a plastic signifier available for infinite semantic operations. This condition is articulated through three characteristics: ‘semantic plasticity’, ‘memetic circularity’ and ‘constitutive performativity’.

Semantic plasticity indicates the image’s ability to generate multiple and incompatible narratives, while memetic circularity indicates that every image can potentially become a meme, i.e., a template that lends itself to infinite variations. In this process, the image loses all connection to its original context to become a pure signifier available for any semantic operation. This does not mean that the image itself is neutral, but that it can be manipulated an infinite number of times.

This memetic logic finds its environment in what Bernard Stiegler called the ‘economy of attention’ [8], a regime in which the value of the image does not derive from its relationship with reality but from its ability to generate engagement. Finally, constitutive performativity shows how every image is an act of visual speech that challenges us, guides our perceptions and changes our opinions. A condition well described in philosophical reflections on photography [9]. It is therefore a question of distinguishing between what an image shows and what it does: in other words, the image is no longer a passive representation, but an active agent in the construction of social reality. Comprehensive analyses of synthetic media and deepfake ecologies confirm this trend [10].

4. Visual Education as a Practice of Pedagogical Disorientation

This post-photographic condition is not necessarily dystopian. If images have emancipated themselves from reality, they can become tools of radical imagination. The image can become a space for counterfactual elaboration in order to enrich our understanding of the present, and for this reason, it requires a reconfiguration of educational practices related to visual education.

The traditional approach, linked to the acquisition of decoding skills such as learning to ‘read’ images, recognise visual codes and deconstruct implicit messages, while still valid, is insufficient in the face of the semantic plasticity and performativity of the algorithmic image. It is not enough, in fact, to teach how to recognise the false, but it is necessary to educate to that condition which, for Edgar Morin, is the ability to ‘learn to learn’, that is, the meta-cognitive competence to observe one’s own learning processes as they unfold [11]. Every image must be questioned with regard to the operations it performs, and this operation requires the development of a performative sensitivity, i.e., the ability to perceive the image as an agent in the construction of social reality.

Another important axis concerns the use of the semantic plasticity of images to generate counter-narratives that challenge hegemonic representations: if every image is a plastic signifier, this availability can be appropriated for transformative purposes. The third axis concerns the ability to develop practices of cognitive resistance to devices that capture attention. At its core, the ambiguity of images can be used to stimulate complex reflective processes, to explore the problematic nature of representation and to develop the ability to inhabit contemporary uncertainty.

5. From Theory to Practice: Experimental Approaches

The three axes outlined above have guided the design of pedagogical experiments conducted within a Prompt Thinking module, part of the Multimedia Design programme at the Istituto Europeo di Design (IED) in Rome, now in its fourth year (2022–2026), involving approximately 40 students per cohort. The module is designed to evolve each year in response to the rapid development of generative AI tools, and adopts an interactive, workshop-based format that moves from ethical reflection on AI to hands-on creative practice. Each year culminates in a project that requires the conscious and critical use of generative AI, growing in complexity as the tools themselves become more sophisticated. In the 2024–2025 edition, for example, students developed a complete creative project: the construction of a fictional musical identity, an artist or band endowed with a coherent narrative, aesthetic and sonic universe. The project required the definition of a visual identity through AI-generated images, the construction of a storytelling apparatus and the generation of at least one musical or textual output obtained through generative tools. The quality of the prompts and the critical reflection on the use of AI were integral parts of the assessment. Students documented their interaction with the generative systems, making explicit the choices, discarded alternatives, unexpected results and interpretive reorientations that emerged during the process. A recurring observation across cohorts is that the iterative cycle of prompt revision leads students to recognise the degree to which their initial visual expectations were shaped by implicit cultural stereotypes embedded both in their own imagination and in the training data of generative models.

This activity operates on all three axes identified in Section 4. On the metacognitive axis, students discover that the visual output of generative AI is radically dependent on the conceptual framing of the prompt: the same musical character, described with different lexical choices, produces profoundly divergent visual identities, concretely demonstrating the semantic plasticity theorised in Section 3. On the axis of counter-narratives, when students ask an AI system to generate the image of a jazz musician or a punk band, the results tend to reproduce standardised visual codes that must be recognised, questioned and deliberately subverted. On the axis of cognitive resistance, the iterative prompt-generation-evaluation cycle forces students to slow down and reflect on each image produced, counteracting the accelerated perceptual consumption described in the introduction.

These experiments are consistent with a broader trend in AI literacy education. The AI Pedagogy Project developed at Harvard’s metaLAB has produced open-access assignments that integrate image-generation tools to expose biases and develop performative sensitivity [12]. Research in digital forensics applied to education, notably Hany Farid’s work at UC Berkeley, has shown that training students to analyse the technical properties of images can be integrated into a critical literacy framework that questions the testimonial status of the photographic image [13]. Recent studies confirm that students engaged in hands-on creative projects with generative AI show significant improvements in critical evaluation skills, provided the pedagogical framework foregrounds reflection on the process rather than the mere production of outputs [14].

6. Towards an Insurrectionary Imagination

The pedagogical experiment described in Section 5 reveals something that exceeds its own didactic premises: students who are asked to construct a fictional identity through generative AI do not simply learn to use a tool but discover the conditions under which any visual identity is produced. The iterative confrontation with the machine’s outputs makes visible a productive mechanism that ordinarily remains invisible: every image presupposes a selection, an exclusion and a hierarchy of the visible. In this sense, the exercise functions as a form of reverse engineering of the post-photographic condition itself.

This is what an insurrectionary imagination looks like in practice: the capacity to intervene in the processes of image production with awareness of their performative effects, rather than passively consuming their results. The transition from a testimonial to a performative regime, far from being only a source of disorientation, opens a space for agency. If every image is a project rather than a trace, then the ability to design images critically becomes a form of political literacy.

The post-photographic era demands educational practices that operate on this double register: technical competence in the use of generative tools, and philosophical awareness of the epistemological transformation they entail. Artificial intelligence, in this framework, functions as a co-modulator of perception, a cognitive agent that can either reinforce perceptual automatisms or, when embedded in a rigorous pedagogical design, contribute to dismantling them. The challenge is to ensure that this capacity remains subject to democratic deliberation and collective control over the means of visual production.