AI, Archives, and the Remixed Flow
von Manuel Meschiari
Introduction
In the last few years, GLAM (Galleries, Libraries, Archives and Museums) institutions have started to extensively use Artificial Intelligence (AI) and Machine Learning (ML), mostly for the automation of categorization and cataloguing tasks, while also exploring innovative methods for capturing, retrieving and creating content. Particularly, in audiovisual archives AI offers numerous possibilities for creative re-use and engagement with the audience, such as Citizen DJ [1] which allows users to create hip-hop music using material from the Library of Congress archive. In this landscape, television archives present great opportunities for new methods, given their vast amount of material at their disposal. One notable and innovative example comes from the BBC, who used AI to retrieve content from its extensive archive and created a new television program called Made by Machine: When AI met the Archive. However, this is not the first case of reassembling together archival material on television: the Italian show Blob, which airs on the Italian public broadcasting network Rai since 1989, uses clips and rearranges them, but without the use of AI. These two examples, seen from a media studies lens, pose the question of how they fit in the television landscape. This essay will explore how AI-driven analysis and reassembly of archival footage, as exemplified by the BBC’s experimental project, redefines concepts of television “flow” and introduces a new form of “pseudoflow” and “remixed flow” in the current AI age.
The BBC Made by Machine Experiment: AI in Practice
The initial phase for creating the program Made by Machine: When AI met the Archive was to use ML to first analyze the content of the BBC archive and identify programs that had specific qualities of the BBC Four channel [2]. To find these characteristic the model was trained on descriptions of themes typical of the channel. More precisely, they used the computer vision task type DenseCap (dense captioning), which creates description of images in natural language and identifies relations between objects [3]. This resulted in a ranked list of 150 relevant programs that were later broken down into small chunks (15,000 in total). The AI then managed to reassemble these chunks trying to find meaning and common thematics through the analysis of subtitles and visual dynamism [4]. Basically, what was interpreted as similar got cut together in the final edit. The program aired on September 5th 2018, on BBC Four [5].
Rai’s Blob
In the Italian television landscape, the program Blob (1989) works on a similar premise of what the BBC tried to do with AI. Blob takes bits and pieces from the Italian television archives, usually what aired on the same day or week, and reassembles them to create political satire. The official tagline of the show is: “Twenty minutes of ‘surrogate TV’. TV disassembled, reassembled and laid bare through editing to reveal what video and its everyday protagonists actually tell us” [6]. What is particular about the show is that what they present is not a re-mash or parody dubs, the clips are left unedited, and it is their juxtaposition that creates meaning. Interestingly, Italian scholar Roberto Balestri managed to create what he called “Blob AI” [7], which uses AI and ML to create a new Blob episode taking clips from Blob itself. This suggest that it is already possible to create a program similar to Blob with AI.
Flow, Pseudoflow and Remixed Flow
Raymond Williams defines flow as “the defining character of broadcasting” [8], where the singular units of programs, together with advertisement, are perceived as one smooth flow. What Williams means is that a television scheduled programming is always “flowing”, like a river, and we as viewers jump into this flow when we turn on the television. But would this still be applicable to programs like Made by Machine: When AI met the Archive and Blob? Cecilia Valenti argues that Williams concept of flow cannot be used in a program like Blob, because it is more similar to the action of “zapping”, and suggests using a new term: “pseudoflow” [9]. I would additionally argue that what Made by Machine and Blob do is more similar to Lawrence Lessig’s concept of remix [10]. He argues that remix embodies a “Read/Write” culture, where audiences actively participate in shaping and reinterpreting cultural artifacts rather than passively consuming them [11]. This leads me to propose a new term: “remixed flow.” It expresses how the established broadcast flow is actively taken and rearranged to create something new, much like how BBC searched directly from the BBC Four archives or how Blob was created from the existing flow of content throughout a day or week. At the same time, they also fit in the televisual flow because they both air(ed) on broadcast television.
Broader implications
This might suggest a move beyond traditional linear programming towards a more dynamic and potentially personalized viewing experience. AI’s ability to break down vast archives into “small chunks” and reassemble them based on thematic connections, object recognition, and subtitle analysis could lead to new forms of content generation, moving beyond the established “flow” model to a “pseudoflow” and “remixed flow.” This redefinition of television in the AI and digital age means that content creation might increasingly involve algorithms sifting through existing material to generate novel narratives or specialized programming, potentially offering viewers highly tailored experiences. This raises questions about how “television” will be defined when its content is not solely produced through conventional means but also through the recombination of existing media.
Conclusion
In conclusion, television is still evolving, and thanks to the integration of AI and the creative re-use of archival material it is now in a new phase. This new phase fundamentally challenges the traditional notions of “flow.” As demonstrated by the BBC’s Made by Machine: When AI met the Archive and Rai’s Blob, the act of meticulously dissecting and reassembling existing content, whether through algorithmic analysis or human curation, introduces a new dynamic that can be defined “remixed flow.” This new paradigm not only redefines how television content is generated and consumed but also positions the audience as active participants in meaning-making. Ultimately, as AI continues to sift through vast digital archives, the very essence of “television” may shift from a linear, broadcast-driven model to a more personalized and algorithmically recombined experience, pushing for an ongoing critical inquiry into its evolving definition in the AI age.
References
[1] Randi Cecchine, “The Potential for AI in Audiovisual Archives”, Blog, 10.05.2021, https://www.beeldengeluid.nl/en/knowledge/blog/potential-ai-audiovisual-archives, 16.06.2025.
[2] Caroline Alton, “AI and the Archive – the Making of Made by Machine”, Blog, 5.09.2018, https://www.bbc.co.uk/rd/blog/2018-09-artificial-intelligence-archive-made-machine, 16.06.2025.
[3] Daniel Chàvez Heras, Cinema and Machine Vision, Edinburgh: Edinburgh University Press, 2024, p. 38–40.
[4] Alton, “AI and the Archive”.
[5] “BBC Made by machine When AI Met the Archive”, Sonar Bangla, 12.07.2022, https://www.youtube.com/watch?v=ZLqqwJ7VzJo&t=2667s, 16.06.2025.
[6] Blob, RaiPlay, https://www.raiplay.it/programmi/blob, my translation from Italian, 16.06.2025.
[7] “L’intelligenza Artificiale ci ruberà il lavoro?: an AI-generated TV pastiche created with AI Blob!”, Roberto Balestri Unibo, 31.03.2025, https://www.youtube.com/watch?v=0e9o-N33CzU&list=PLF3Gpk1ZBOk2Y334h8WcCAi4VBopu4lXf&index=2, 16.06.2025.
[8] Williams Raymond. “Programming as Sequence or Flow”, in Marris Paul, Media Studies: A Reader 2nd ed., 2004, Edinburgh: Edinburgh University Press, p. 192.
[9] Cecilia Valenti, “Blob als Pseudoflow”, in Das Amorphe im Medialen, 2019, Bielefeld: transcript, p. 77–81.
[10] Lawrence Lessig, Remix – Making Art and Commerce Thrive in the Hybrid Economy, 2008, London: Bloomsbury.
[11] Ibid.