This essay examines the ethical and political challenges that images of the human face generated by artificial intelligence (AI) pose in the context of media pedagogy. The essay investigates these images and their associated challenges through a classroom assignment entitled Facial Subjects, which invites students to generate AI portraits representing socially or politically charged identity labels such as “climate activist,” “criminal,” “influencer,” “refugee,” or “scientist.” Algorithmically generated facial images, from social media filters to predictive policing tools, have cultural and political implications, as they reproduce aesthetic codes, social norms, and ideological assumptions.

Situated at the intersection of media archaeology (Parikka, 2012), visual culture studies (Poster, 2002), and critical AI studies (Raley and Rhee, 2023), this pedagogical exercise prompts students to compare synthetic images with real-world and archival photographic portraits and thus exposes the visual logics and latent biases embedded in AI training datasets (Benjamin, 2019; Browne, 2015; Noble, 2008). It explores AI tools as culturally and politically saturated artifacts whose creations demand critical analysis. In doing so, it cultivates both experimental and reflexive forms of media literacy. This dual approach equips students with tools to critically navigate contemporary image ecologies; it entails thinking critically about and with AI to understand how computational forms rearticulate visual regimes of race, class, gender, and power.

Through an analysis of the Facial Subjects assignment, the essay explores how generative AI image-making can serve as a critical practice in the film and media studies classroom. In doing so, it highlights how teaching with AI can support students’ critical study of visual culture, in this case, by dissecting the politics of the face.

In the assignment Facial Subjects, students use AI tools to generate photographic portraits of socially or politically charged identity labels (e.g., “climate activist,” “criminal,” “influencer,” “refugee,” or “scientist”) and compare them to archival and real-world portraits. The essay holds that AI-generated images, when critically embedded in media-pedagogy, can serve as media through which students interrogate visual cultures. Conversely, critical methods drawn from visual culture studies can facilitate critical research in AI. Students uncover the latent codes that inform contemporary notions of identity, personhood, and visibility, both with and without machine-aided cultural constructions. The assignment thereby also challenges the more general notion of algorithmic neutrality and opens discussions about algorithmic bias and discriminatory media.

Generative AI-tools such as DALL-E operate as cultural artefacts entangled with questions of aesthetics, bias, and representation, showing a set of properties according to which they could be called discriminatory media. These tools generate images of faces, environments and bodies that are compelling, surreal and ideologically saturated, and which may be used critically to problematize our relationship to identity categories and norms, bias, and bias correction (Miller et al., 2023; Sun et al., 2024).

Teaching with AI in general and the Facial Subjects assignment in particular can be linked to the growth of interest in media studies in critical practice in the digital humanities. Critical practice in the digital humanities emphasizes how digital platforms allow (or afford) novel pedagogical practices (Gold, 2012; Svensson and Goldberg, 2015). For practitioners of digital humanities, humanistic analysis assumes new possibilities and proportions when empowered by digital technologies: in critical practice, teaching is not just about, but also with media.

Closely related to the context of critical practice in the digital humanities is the boom of interest in media archaeology, a subfield of media studies dedicated to the idea that different media platforms allow the articulation of different kinds of knowledge, temporality, sociality, and interactivity (Parikka, 2011).

The Facial Subjects assignment emerges from a broader pedagogical approach I term media archaeological pedagogy. By media archaeological pedagogy, I refer to an approach that not only studies the content or techniques of a particular medium or platform, but also, following methodologies loosely identified with media archaeological scholars such as Kittler (1999) and Emerson (2014) accounts for the “particular material nature” of media, including the role of institutions, techniques, protocols, and disciplinary norms in allowing technical media to shape the cultural fabric of its period (Huhtamo and Parikka, 2011, p. 8). This approach foregrounds how media infrastructures and protocols condition what can be seen, said, or known. In particular, media archaeology’s growth in popularity coincides with a wider flourishing of interest, according to which the format or materiality of a medium structures the kinds of knowledge it produces.

The development of media archaeological methods has been paralleled by the development of creative media archaeological practices that emphasize, in a variety of ways, the interpenetration of technologies with creative and cultural creation (Parikka, 2012).

According to this development, a media archaeological approach may offer a new way of knowing, teaching, and learning by allowing its practitioners to experiment with media as they study it. In this critical pedagogical approach, students do not merely analyze images or media; they also experiment with the tools that produce them and discuss how they are produced. By integrating theory and practice, this approach promotes a pedagogical climate in which students learn about as well as with media.

In the Facial Subjects assignment, media archaeological methods coincide with a field of media-practical, critical, and historical studies of how machine-identification of faces coincides with longer histories of discriminatory practices such as eugenics (Crawford and Paglen, 2019; Meyer, 2019; Beiguelman, 2023; Sekula, 1986). These methods and fields can be used to investigate the aesthetics of machine vision, for example, by using generative AI to materialize categories of social identity, asking questions such as:

What kinds of images are produced when we prompt an AI with terms such as “climate activist,” “criminal,” “influencer,” “refugee,” or “scientist”?

Rather than positioning students as passive consumers of purely media theoretical instruction, the Facial Subjects assignment places students within the production pipeline of digital culture, thereby allowing them to grapple with questions of media aesthetics from both theoretical and practical perspectives. This approach combines theoretical and practical perspectives for a double learning effect. Students’ theoretical questioning enhances creative insights; conversely, practical engagement aids students’ critical analysis of media circulation. Furthermore, in Facial Subjects, the classroom, and with it, learning, is flipped, allowing for a pedagogical approach that integrates student-and teacher-centered formats, enabling more active student learning. In that way, students become both analysts and creative agents, generating visual data while reflecting on the processes that govern their creation and their limitations.

The Facial Subjects assignment is designed for a group of ca. 30 undergraduate students in the program Media, Aesthetics and Cultural Entrepreneurship (MEK) at the University of Gothenburg and comprises the following three stages:

Prompt design and image generation

This structured progression encourages students to move from production to interpretation and thus mirrors a critical practice approach increasingly central to film and media studies.

The Facial Subjects assignment’s focus on the face grows out of a longer history of interest in the face in film and media studies (Deleuze, 1986; Doane, 2003). Considered the legible outside of a hidden inside in early film-theory (Balázs, 1930; Epstein, 1924/1981), the face is today often reduced to a mere surface of recognition. It functions as a dense semiotic site where norms of race, gender, age, and beauty are inscribed and contested, and prejudices are formed. Today’s facial images travel across surveillance systems (Lee-Morrison, 2022; Blas, 2011–2014; Zuboff, 2019), beauty industries and social media platforms. These algorithmic images (Somaini, 2023) take on functions as markers of individuality and tools of classification in the algorithmic configuration of the face (Rettberg, 2023).

Generative AI systems are trained on massive image datasets scraped from the web, most of which carry with them the biases of culturally dominant imaginaries. When students prompt a model to render “scientist,” they are unlikely to be shown a woman of colour unless specified. “Criminal” may default to racialized or hypermasculine tropes. These results reflect the images’ statistical likelihood or their “meanness” (Steyerl, 2023; Brown, 2023). In Facial Subjects, students are encouraged to reflect on bias and the limits of simple bias correction of training data in machine learning (ML) systems, for example in reinforcement learning tools such as Google’s Gemini (Milmo and Hern, 2024). More general considerations of so-called “stochastic parrots” (Bender et al., 2021) may further supplement this inquiry‚ for example through readings, discussions, and lectures.

The human face is an instance where it is possible for students to analyze the ascriptions of meaning and the production of stereotypes without much prior research or theoretical preparation. In producing and analyzing these images, students confront the mechanics of normativity and discrimination, bias and bias correction. They reflect on how the face becomes a site of algorithmic inscription, flattening and anesthetization of social hierarchies into likely realities, and where questions of both bias and bias correction necessarily lead to discussions of ethical and political concerns in relation to AI-generated images.

The Facial Subjects assignment enables students to reach the following learning goals:

Media and algorithmic literacy

The assignment shifts assessment away from purely textual formats, the discussion of academic texts, and classical written assignments such as traditional research essays, and instead favours multimodal formats that mirror the complexity of the media environments students inhabit. This allows for a broader perspective on the materials taught and enables students with different learning styles to approach the topic via multiple entry-routes.

Teaching with AI-generated images, via an assignment such as Facial Subjects, encourages students to make visible the role that AI tools take on as epistemological frameworks, while also encouraging them to reflect on the ethical and political questions involved. In this sense, Facial Subjects positions media as conceptual instruments while it also helps students to grasp the urgency of developing critical, ethical and creative approaches that can read the different layers of politics, discrimination, bias and bias correction.

The pedagogical potential of the Facial Subjects assignment lies not only in its novelty and creative approach, and its media archaeological outlook. Through generative image-making students do not just learn about media, they also learn with it. Facial Subjects externalizes and makes legible the politics of aesthetics related to AI tools.

My pedagogical interest in teaching assignments such as Facial Subjects relates to their critical media perspective that is as much concerned with teaching about media as it is concerned with teaching with media. To quote media scholar Marshall McLuhan (1964), here “the medium is the message,” with AI permitting new pedagogical messages to be crafted in the classroom. I am interested in how media is not simply one object among others to be studied but is, also a resource or platform that can be used to structure different kinds of intellectual and pedagogical investigations. Allowing students of film and media studies to bring media practice to the classroom encourages a holistic approach to teaching and learning, where students engage in hands-on experimentation with the potentials of a particular medium (Bates, 2015). Through such hands-on experimentation students go beyond abstract analysis, instead engaging in assignments that help them understand the practices and material affordances that characterize, for example, AI-generated images of the face (Koehler and Mishra, 2006).

The Facial Subjects assignment offers an example of how AI tools can be mobilized to reveal the ideological and cultural work of AI-generated imagery. It encourages students to examine the construction of visual identities from different perspectives and to intervene in that process with critical tools and hands-on experimentation, while being aware of the challenges posed by bias and bias correction in machine learning. In that way, students come to see AI not just as a tool for generating images, but also as a medium, through which meaning, and the politics of the face, are generated and negotiated.