{"id":8183,"date":"2026-07-29T17:02:39","date_gmt":"2026-07-29T15:02:39","guid":{"rendered":"https:\/\/www.humus-digitalmag.com\/?p=8183"},"modified":"2026-07-29T19:37:23","modified_gmt":"2026-07-29T17:37:23","slug":"outside-the-frame-imagining-otherwise-in-the-age-of-artificial-intelligence","status":"publish","type":"post","link":"https:\/\/www.humus-digitalmag.com\/en\/outside-the-frame-imagining-otherwise-in-the-age-of-artificial-intelligence\/","title":{"rendered":"Outside the frame. Imagining otherwise in the age of artificial intelligence"},"content":{"rendered":"\n

What is an image?<\/strong> Can it be a neutral surface, an innocent recording of the world, or is it always a mediated expression, shaped by context and intention? What relationship exists between the viewer and the person, or thing, being viewed? And what power relations pass through a face or a landscape, through what remains and what is lost outside the frame?<\/p>\n\n\n\n

These are the questions that, in 2019, led conceptual artist Linda Dounia Rebeiz to experiment with the first generative models to create her works<\/strong>. She queried them about the city where she grew up, Dakar, but the models returned a place she did not recognize: dirty streets, low and dilapidated buildings, peeling walls. A city dispossessed of itself. Observed from afar, through eyes that did not belong to it.<\/p>\n\n\n\n

What does a machine see when we ask it to give form to a place, a story, to write a piece of music or design a city?<\/strong><\/p>\n\n\n\n

Trained on vast amounts of data, AI learns regularities and recurrences<\/strong>. The probability that one word will follow another; that certain features, bodies, or compositions will combine with one another; that rhythms, timbres, or progressions will follow certain patterns. Data that appear solid to us because they are empirical and measurable, and therefore seemingly able to escape the ambiguity of the gaze. And yet they are only proxies for reality.<\/strong> They carry gaps, old taxonomies, wounds, and injustices. What has not been told, photographed, exhibited, or named remains at the margins, deprived of its own voice and excluded from the possibility of being represented, because with every generation request the models select the statistically prevailing forms <\/strong>and recombine them into the most plausible outcome.<\/p>\n\n\n\n

Thus, when Rebeiz questions DALL-E (an artificial intelligence algorithm capable of generating images from textual descriptions, editor\u2019s note), the model does not see Dakar itself, but the version of Dakar preserved in the world\u2019s digital memory<\/strong>. Images and texts produced during colonial and post-colonial missions, French, British, and American military operations, tourist collections, and humanitarian repertoires that tell stories of poverty, decay, emergency, and ruin.<\/p>\n\n\n\n

\"\"
Universe of Water Particles on a Rock Where People Gather<\/em>, immersive installation at teamLab Borderless in Tokyo, at the Mori Building Digital Art Museum, ph. teamLab<\/figcaption><\/figure>\n\n\n\n

An ambivalent force<\/strong><\/p>\n\n\n\n

Stripped of the illusion of neutrality and universality, the models we use reveal a political, cultural, and economic center of gravity firmly anchored in the West<\/strong>. An unequal cartography, in which some aesthetics, cultures, and geographical areas are infinitely more present than others. Modifying prompts, adding details, specifying, correcting, asking again is not enough. When, for example, we query text-to-image technologies such as Midjourney, DALL-E, or Stable Diffusion to generate images of CEOs, the result is a lineup of Caucasian men. If the request concerns nursing staff, AI returns female faces, often linked to migrant backgrounds. While societies struggle to transform themselves \u2014 with women and minorities gaining new spaces \u2014 systems outline a regressive landscape<\/strong>, crystallizing leadership in the figure of the white man.<\/p>\n\n\n\n

The same happens with language models. During testing phases, when researchers ask ChatGPT what kinds of people get married, even as the variations multiply, the narrative does not change. The couple is strictly heterosexual.[1]<\/sup> Distortions, prejudices, and inequalities are the point at which the partial history of societies, with its relations of domination, turns into statistical prevalence.<\/p>\n\n\n\n

Are we destined, then, to reiterate habits, stereotypes, exclusions, and power dynamics? Will artificial intelligence and probabilistic calculation undermine our creativity and our ability to imagine, condemning us to the compulsion of what has already been said?<\/strong><\/p>\n\n\n\n

Recent scientific literature[2] <\/sup>shows that generative models are not more creative than we are, but they can improve individual skills and performance<\/strong>. Yet the gain for individuals comes at a collective cost. Ideas end up drawing closer together, resembling one another until they converge. Artificial intelligence is an ambivalent force, <\/strong>capable of opening passages when the mind comes to a halt, or of bending it and anchoring it to familiar configurations, making increasingly difficult what can only arise off the beaten path. To diverge means moving through remote regions of imaginative space. The risk is that thought shifts from exploration to selection. AI reduces doubt, shortens the period of uncertainty, and accelerates the arrival of an answer, but that suspension \u2013 in which figures are still blurred, words are slow to arrive, and images and sounds resist without allowing themselves to be grasped \u2013 belongs to the space of possibilities, where what does not yet exist can emerge<\/strong>.<\/p>\n\n\n\n

\"\"
The Infinite Crystal Universe<\/em> is one of the Japanese collective\u2019s best-known environments. Created through thousands of digital light points, it is located at teamLab Planets Tokyo, ph. teamLab<\/figcaption><\/figure>\n\n\n\n

Desire and memory<\/strong><\/p>\n\n\n\n

William S. Burroughs<\/strong> believed that language<\/strong> was not a neutral tool, something we use to describe reality and express ourselves freely, but a system that imposes sequences and relations of meaning, helping to shape thought before the subject can even intervene<\/strong>. It is a virus, an infected, hallucinatory, authoritarian substance capable of colonizing the body and mind. <\/strong>For the author of Naked Lunch<\/em>, the only possible form of rebellion is to sabotage language, break its linearity, make it discontinuous, open to accident and collision<\/strong>. Thus, in the early 1960s, he experimented with the cut-up<\/em> technique. Pages are cut, sentences disarticulated, fragments recomposed to create unexpected connections. If every word is also an image, cutting into text means opening fissures, freeing language from its function of command. If domination passes through discourse that proceeds without hesitation, subverting it means cracking the reality that words help to produce.<\/strong> It is in those cracks that meaning opens up, becomes unstable again and available.<\/p>\n\n\n\n

With generative AI, that virus makes a species leap and no longer circulates only among bodies, books, newspapers, institutions, advertising, and political discourse.<\/strong> It inhabits disembodied systems devoid of desire and lived memory.<\/strong> Machines know how to assemble, recombine, translate, vary, and even produce convincing cut-ups, but while they generate fractures, they tend to smooth their outcomes. They insert logical links between fragments, recompose grammar, and bring the narrative back within uniform discursive norms.[3]<\/p>\n\n\n\n

Building inclusive datasets<\/strong><\/p>\n\n\n\n

If the power of models consists in organizing the visible, the sayable, and the imaginable, shaping the lens through which we conceive of and make sense of the world, how is rebellion exercised? How can we prevent what frequency consecrates as an ordering principle and rule from becoming the inevitable horizon of the future?<\/strong> <\/p>\n\n\n\n

In American Capitalism<\/em>, economist John Kenneth Galbraith<\/strong> writes that \u201cThe exercise of countervailing power is the principal solvent of economic power, the fundamental defense against its domination [\u2026]\u201d. [4] <\/sup> History has shown us that no political or economic power can be contained by law alone, or by the ethical will of those who hold it. Counter-powers are needed, equipped with means adequate to those of the force they oppose.<\/p>\n\n\n\n

In an era in which a handful of private companies control global infrastructures, computing capacity, data, models, objectives, and criteria of visibility, the answer is not abandonment, but reappropriation.<\/strong> Artificial intelligence can become a tool at the service of civil society and its organizations.<\/p>\n\n\n\n

Languages, texts, images, behaviors, social codes, aesthetics, and emotions are the collective heritage that major technology companies have turned into private capital. Demanding that part of the economic value generated be returned to society \u2014 for example through a data dividend intended to finance computing capacity and co-managed, co-decidable models \u2013 should be part of the social pact of the twenty-first century. <\/strong>And yet countervailing power cannot be limited by economic claims alone. We must intervene in the infrastructures, the memory, and the conditions through which certain visions become prevalent over others. Building new datasets to restore presence to local knowledge, minority cultures, and Indigenous and community knowledge.<\/strong> AI can generate counterfactual scenarios, analyze information, identify regularities, connections, anomalies, and absences. Obtaining better answers is not enough. The exercise of countervailing power begins when the memory from which those answers take shape becomes visible, along with the imbalances that run through them and the voids they carry; when, stripped of the appearance of inevitability, they become contestable once again.<\/p>\n\n\n\n

\"\"
A group of digital koi carp trace ever-changing paths across the surface of the water, responding to the presence of visitors. teamLab Planets Tokyo, Toyosu, ph. teamLab<\/figcaption><\/figure>\n\n\n\n

What lies beyond the frame<\/strong><\/p>\n\n\n\n

This is the gesture of Linda Dounia Rebeiz, when she begins photographing streets, scanning images from herbaria and national archives, abandoning major commercial models. <\/strong>She starts working with generative neural networks (GANs). Projects such as Once Upon a Garden<\/em>, on the threatened flora of the Sahel<\/strong>, are born,where generation does not coincide with abundance and visual proliferation, but with mourning. The Senegalese artist does not seek perfection. She knows that AI will produce opacity, uncertain and amorphous forms. And disappearance itself is the center of that work. The flowers are spectral presences, remains of what can no longer be seen, touched, smelled, and recognized. And in this disturbing beauty, AI makes loss visible.<\/p>\n\n\n\n

Rebellion, then, consists in making artificial intelligence a tool with which to dispute power: building systems to identify violations of rights, ethical and legal principles, discrimination, and asymmetries, and to contest outcomes and decisions<\/strong>. But giving meaning to what remains outside the frame<\/strong>, imagining otherwise, and choosing which directions to pursue remain human and collective tasks<\/strong>.<\/p>\n\n\n\n

\n\n\n\n

[1]<\/sup> Singh, A., et al. (2025). Openai gpt-5 system card. arXiv preprint arXiv:2601.03267<\/em>.<\/p>\n\n\n\n

[2] <\/sup>Doshi, A. R., & Hauser, O. P. (2024). Generative AI enhances individual creativity but reduces the collective diversity of novel content. Science advances<\/em>, 10<\/em>(28); Wadinambiarachchi, S., Kelly, R. M., Pareek, S., Zhou, Q., & Velloso, E. (2024). The effects of generative AI on design fixation and divergent thinking. In Proceedings of the 2024 CHI conference on human factors in computing systems<\/em> (pp. 1-18); Lupyan, G., & Bergen, B. (2016). How language programs the mind. Topics in cognitive science<\/em>, 8<\/em>(2), 408-424.<\/p>\n\n\n\n

[3] <\/sup>Mackay, J., and Mackay, P. (2024). Experiments in Generating Cut-up texts with Commercial AI. Electronic book review<\/em>.<\/p>\n\n\n\n

[4] <\/sup>Galbraith, J.K. (2017). American capitalism: The concept of countervailing power<\/em>. Routledge.<\/p>\n\n\n\n

Bibliography<\/h2>\n\n\n\n

Doshi, A. R., Hauser, O. P. (2024). Generative AI enhances individual creativity but reduces the collective diversity of novel content. Science advances<\/em>, 10<\/em>(28).<\/p>\n\n\n\n

Galbraith, J.K. (2017). American capitalism: The concept of countervailing power<\/em>. Routledge.<\/p>\n\n\n\n

Singh, A., et al. (2025). Openai gpt-5 system card. arXiv preprint arXiv:2601.03267<\/em>.<\/p>\n\n\n\n

Lupyan, G., & Bergen, B. (2016). How language programs the mind. Topics in cognitive science<\/em>, 8<\/em>(2), 408-424.<\/p>\n\n\n\n

Mackay, J., Mackay, P. (2024). Experiments in Generating Cut-up texts with Commercial AI. Electronic book review<\/em>.<\/p>\n\n\n\n

Rebeiz, D. L. (2025). AI\u2019s Unseen Borders: Crafting a Pluriverse Future. In CAA 113th Annual Conference<\/em>. CAA.<\/p>\n\n\n\n

Wadinambiarachchi, S., Kelly, R. M., Pareek, S., Zhou, Q., & Velloso, E. (2024). The effects of generative AI on design fixation and divergent thinking. In Proceedings of the 2024 CHI conference on human factors in computing systems<\/em> (pp. 1-18).<\/p>\n\n\n\n

<\/p>\n\n\n