A new study points to some answers.<\/p>\n
In January 2026, artificial intelligence researchers Arend Hintze, Frida Proschinger \u00c5str\u00f6m and Jory Schossau published a study<\/a> showing what happens when generative AI systems are allowed to run autonomously \u2013 generating and interpreting their own outputs without human intervention. <\/p>\n The researchers linked a text-to-image system with an image-to-text system and let them iterate \u2013 image, caption, image, caption \u2013 over and over and over. <\/p>\n Regardless of how diverse the starting prompts were \u2013 and regardless of how much randomness the systems were allowed \u2013 the outputs quickly converged onto a narrow set of generic, familiar visual themes: atmospheric cityscapes, grandiose buildings and pastoral landscapes. Even more striking, the system quickly \u201cforgot\u201d its starting prompt. <\/p>\n The researchers called the outcomes \u201cvisual elevator music\u201d \u2013 pleasant and polished, yet devoid of any real meaning.<\/p>\n For example, they started with the image prompt, \u201cThe Prime Minister pored over strategy documents, trying to sell the public on a fragile peace deal while juggling the weight of his job amidst impending military action.\u201d The resulting image was then captioned by AI. This caption was used as a prompt to generate the next image.<\/p>\n After repeating this loop<\/a>, the researchers ended up with a bland image of a formal interior space \u2013 no people, no drama, no real sense of time and place.<\/p>\n A prompt that begins with a prime minister under stress ends with an image of an empty room with fancy furnishings. As a computer scientist who studies generative models and creativity<\/a>, I see the findings from this study as an important piece of the debate<\/a> over whether AI<\/a> will lead to cultural stagnation<\/a>. <\/p>\n The results show that generative AI systems themselves tend toward homogenization when used autonomously and repeatedly. They even suggest that AI systems are currently operating in this way by default.<\/p>\n The familiar is the default<\/p>\n This experiment may appear beside the point: Most people don\u2019t ask AI systems to endlessly describe and regenerate their own images. The convergence to a set of bland, stock images happened without retraining. No new data was added. Nothing was learned. The collapse emerged purely from repeated use. <\/p>\n But I think the setup of the experiment can be thought of as a diagnostic tool. It reveals what generative systems preserve when no one intervenes.<\/p>\n Pretty \u2026 boring. This has broader implications, because modern culture is increasingly influenced by exactly these kinds of pipelines. Images are summarized into text. Text is turned into images. Content is ranked, filtered and regenerated as it moves between words, images and videos. New articles on the web are now more likely to be written by AI than humans<\/a>. Even when humans remain in the loop, they are often choosing from AI-generated options rather than starting from scratch. <\/p>\n The findings of this recent study show that the default behavior of these systems is to compress meaning toward what is most familiar, recognizable and easy to regenerate. <\/p>\n Cultural stagnation or acceleration?<\/p>\n For the past few years, skeptics have warned that generative AI could lead to cultural stagnation by flooding the web with synthetic content that future AI systems then train on<\/a>. Over time, the argument goes, this recursive loop would narrow diversity and innovation. <\/p>\n Champions of the technology have pushed back, pointing out that fears of cultural decline accompany every new technology<\/a>. Humans, they argue, will always be the final arbiter of creative decisions.<\/p>\n What has been missing from this debate is empirical evidence showing where homogenization actually begins.<\/p>\n The new study does not test retraining on AI-generated data. Instead, it shows something more fundamental: Homogenization happens before retraining even enters the picture. The content that generative AI systems naturally produce \u2013 when used autonomously and repeatedly \u2013 is already compressed and generic.<\/p>\n This reframes the stagnation argument. The risk is not only that future models might train on AI-generated content, but that AI-mediated culture is already being filtered in ways that favor the familiar, the describable and the conventional.<\/p>\n Retraining would amplify this effect. But it is not its source.<\/p>\n This is no moral panic<\/p>\n Skeptics are right about one thing: Culture has always adapted to new technologies. Photography did not kill painting. Film did not kill theater. Digital tools have enabled new forms of expression.<\/p>\n But those earlier technologies never forced culture to be endlessly reshaped across various mediums at a global scale. They did not summarize, regenerate and rank cultural products \u2013 news stories, songs, memes, academic papers, photographs or social media posts \u2013 millions of times per day, guided by the same built-in assumptions about what is \u201ctypical.\u201d <\/p>\n The study shows that when meaning is forced through such pipelines repeatedly, diversity collapses not because of bad intentions, malicious design or corporate negligence, but because only certain kinds of meaning survive the text-to-image-to-text repeated conversions.<\/p>\n This does not mean cultural stagnation is inevitable. Human creativity is resilient. Institutions, subcultures and artists have always found ways to resist homogenization. But in my view, the findings of the study show that stagnation is a real risk \u2013 not a speculative fear \u2013 if generative systems are left to operate in their current iteration.<\/p>\n They also help clarify a common misconception about AI creativity: Producing endless variations is not the same as producing innovation. A system can generate millions of images while exploring only a tiny corner of cultural space.<\/p>\n In my own research on creative AI<\/a>, I found that novelty requires designing AI systems with incentives to deviate from the norms. Without it, systems optimize for familiarity because familiarity is what they have learned best. The study reinforces this point empirically. Autonomy alone does not guarantee exploration. In some cases, it accelerates convergence.<\/p>\n This pattern already emerged in the real world: One study found that AI-generated lesson plans featured the same drift<\/a> toward conventional, uninspiring content, underscoring that AI systems converge toward what\u2019s typical rather than what\u2019s unique or creative.<\/p>\n ![\"A](\"https:\/\/www.newsbeep.com\/us\/wp-content\/uploads\/2026\/01\/file-20260115-56-5nvrqz.png\")
<\/a><\/p>\n
\n Arend Hintze, Frida Proschinger \u00c5str\u00f6m and Jory Schossau<\/a>, CC BY<\/a><\/p>\n![\"A](\"https:\/\/www.newsbeep.com\/us\/wp-content\/uploads\/2026\/01\/file-20260114-56-5knzfq.jpg\")
<\/p>\n
\n Chris McLoughlin\/Moment via Getty Images<\/a><\/p>\n![\"A](\"https:\/\/www.newsbeep.com\/us\/wp-content\/uploads\/2026\/01\/file-20260114-64-9cwb80.jpg\")
<\/p>\n