‘The Custodian of the BBC Archives’: The future of BBC Four as an archive channel

In its Annual Plan for 2021/22, the BBC announced that it would cease to make new content for BBC Four, instead relying on repeats and archive programmes to fill the channel’s schedule. The decision might seem to be a pragmatic response to the corporation’s financial constraints, but will it really lead to the channel becoming ‘the home of the most distinctive content from across the BBC’s archive’ as the broadcaster claims? This article explores the potential risks to cultural memory of recycling archival material on BBC Four with no new content to add to the mix.


Introduction
In its Annual Plan for 2021/22, the BBC sets out a vision for BBC Four as the 'home of the most distinctive content from across the BBC's archive' (BBC, 2021: 22). The broadcaster's decision to cease making new content for the channel, instead relying on repeats and archive programmes to fill the schedule, can be viewed as a pragmatic and logical response to the economic restraints currently felt across the corporation. Why pay for new content when there are decades' worth of programmes sitting in the vaults? As Vana Goblot points out, archival content is an integral part of television ecology in the UK; as public service and commercial broadcasters expand their digital portfolios, an 'increasing amount of space is given to television repeats and archival programming' (Goblot, 2015: 81). This is certainly the case with BBC Four, which, since its inception in 2002 'has been regarded as the custodian of the BBC archives' (Goblot, 2015: 80). The channel relies heavily on factual and arts content from the BBC archive, with repeats or specially commissioned archive-based history programmes often featuring in its themed seasons and special events. Former Channel Editor (2013-20) Cassian Harrison has noted 'the channel has often been quite innovative in that particular way, say the Friday night music zones, which is where what we do is we combine new originations and archived content around themes to bring an evening of music' (Harrison, 2021). Indeed, the BBC reports that the channel's current archive content 'already comprises 76% of BBC Four's broadcast hours' (BBC, 2021:22).
The decision to use BBC Four as a shopfront for its television archive also offers an elegant solution to the tricky problem of facilitating access to an archive paid for by public funds but too vast and vulnerable to be accessed directly by the public. Section 69(2) (b) of the BBC Charter states that the broadcaster 'must give the public reasonable opportunities to visit the archives and view or listen to material kept there, with or without charge as the BBC thinks fit ' (2016:43). Currently, the principal method by which this obligation is met is via the mediation of the BBC itself, in archive programmes and repeats. There are other ways, official and unofficial, for the public to access the BBC's archive. For example, organisations such as the BFI, Learning on Screen (which runs the Box of Broadcasts subscription service for education) and Kaleidoscope hold the keys to certain elements of the broadcaster's archive. Meanwhile, sites such as YouTube offer a more unofficial route to access. However, none of the bodies offer full, unmediated access to the BBC's archive. Thus, dedicating a channel to exploring the riches of an archive which represents nearly a century of British and television history has the potential to open up the past to the public in new and exciting ways. Indeed, as Goblot rightly observes, the reproduction and reuse of television archive 'can lead to rediscoveries of the archive's new cultural purposes and contribute towards television's complexity as a rich cultural form' (Goblot, 2015: 90). However, Harrison argues that 'it's not as simple that you could just dust off things from the archive and people will just come and watch them. It's more complex' (2021). My experience as an archive producer (often working for the BBC) is in keeping with Harrison's observation and makes me wary of how successful the channel will be in contributing to the complexity of British and television culture by relying solely on the contents of its own archive. While I agree with Goblot's assertion that 'archive mining and editorial reflexivity are embedded production practices' (ibid), I know from my own experience that there is not always the time, money or accessibility necessary for a deep dive into the archive. These restrictions often result in more readily available material being recycled again and again, which in turn risks creating a homogenised version of the past. As Harrison comments, 'the percentage of content which you can dust off or bring back and that will be relevant to a new audience is quite small' (2021). Therefore, this article will consider the potential risks to cultural memory of recycling archival material on BBC Four with no new content to add to the mix. Focusing on the channel's experiments with artificial intelligence (AI) in the archives, as showcased in Made by Machine: When AI Met the Archive (2018), it will it look for clues as to what the future might hold for BBC Four as the curator of the BBC's past. It will also question whose past is being curated, exploring what the archival choices made by AI designed by BBC staff can tell us about how BBC Four sees its audience and interprets British cultural heritage.

Recycling the past
It is worth first considering the role that the mediation and remediation of the BBC's television archive material plays in embedding a narrative about the past into public memory. Foucault famously argued that archives are sites of power as they play a crucial role in the construction of our understanding of the past (Foucault, 2008). Whoever holds the keys to the archive also holds the keys to the past and the stories that can be told about it. They are often the first port of call for historians and programme-makers to find material with which to construct a narrative about the past. Craig Robertson describes the content of archives as a 'pact with the future' claiming that 'what the present chooses to save in this pact produces the basis for our future understanding of the past' (Robertson, 2011: 5). Astrid Erll's concept of remembering as an act of 'assembling available data that takes place in the present' (Erll, 2011: 8) is a striking analogy for what happens in the edit room when assembling a programme from recycled television footage to reconstitute the past in the present. To assimilate the past into a sixty-minute timeline, value judgements (influenced by personal taste and institutional guidelines) are made by programme-makers about keeping and discarding footage. What makes it into the programme (and what programmes are included in the schedule) matters because remediation is a vital factor in cultural memory formation. Events represented again and again in different media become sedimented into public consciousness, 'creating and stabilising certain narratives and icons of the past' (Erll, 2010: 393). As Aleida Assmann notes, this secures 'for certain artefacts a lasting place in the cultural working memory of a society' (Assmann, 2010: 101). This can be seen in the mobilisation of television archive material; through repetition and remediation, an audiovisual canon is created, memorialising specific events and embedding them into cultural memory. Assmann argues that a cultural canon is essential for a nation state to create a narrative about its past which is then 'taught, embraced, and referred to as their collective autobiography' (Assmann, 2010: 101). This is relevant to the BBC because the visual canon of television archive material created in its edit suites, embedded in its schedules and broadcast on BBC Four establishes a specific narrative about the past which sediments into public consciousness with each repeat showing.
While remediation of television archive material works to memorialise specific narratives about the past, there is also the risk that what is not continuously remediated will be forgotten (Brunow, 2017). This situation is exacerbated by the organisation of the BBC's television archive. Spread over a plethora of digital and physical operating systems and storehouses across the four UK nations, on a variety of formats in various states of vulnerability, and often with complex rights issues involved around where and when they can be viewed, it would take a researcher years to access and view everything that the BBC audiovisual archives hold. However, increasingly tight production schedules due to budgetary restraints mean that researchers have only a few weeks or days rather than years to find the content for archive programmes. In my work as an archive producer, I often used digitised footage stored on one of the desktop server archive systems as a first (and last) port of call as I did not have time to source and digitise tape stock housed in an archive warehouse hundreds of miles away from the production office. As the BBC television archive is not open to the public, with access mediated by commissioners, programme-makers and archivists within the corporation who decide how, when and in what context it will be shown, constraints such as the ones I encountered risk inhibiting public access to the past. Thus, the narratives about the nation's past perpetuated by the BBC risk shaping 'the relationship of television's legislators, trainees, practitioners and historians in an imaginary past and even more speculative future' (Branston, 1998:51).

AI in the archives?
Harrison understands the crucial role that the BBC television archive plays in British cultural heritage and the duty of care that the broadcaster has to it. As he notes, its contents have been produced for the public at the licence fee payers' expense and as such 'we have a duty to preserve and to organize and manage that record of the UK's cultural narrative over the entire time that the BBC has been in existence.' He also argues that 'there's kind of further questions about how it is that we can leverage that archive to be relevant to today's audiences' (2021). With this in mind, in 2018, he approached the BBC's Research and Development department regarding the feasibility of using AI and machine learning to scope the contents of the archive. 'They were really up for it' he says, 'because they're always looking for ways to use or to test things that they're working on at an abstract level in the real-world environment' (2021). The attraction of machine intervention for Harrison was its ability to 'bring insight to content or to creativity in a way that is new and novel'. He argues that machine learning 'promises a kind of scalable insight… you can get a machine to look at lots more of that thing than you can, if you can get a human being to do it' (2021). Referring to the vast size of the BBC television archive, Harrison notes, 'given the fact that there is so much archive and you know, the way that we would select archived content for BBC Four was very much a manual process and very intense manual process' the hypothesis he wanted to test was 'what might happen if we tried to get a machine to select archive content for the Channel' (2021). The idea for the experiment and the resulting evening of specific programming, including scheduling picked by algorithm, and Made by Machine: When AI Met the Archive, a one-hour programme of archive material selected and edited by AI, reflects the remit of the channel at the time, which Harrison states was to push boundaries. Made by Machine certainly does this in its construction, and it is worth detailing the process in more detail.
According to Harrison, over a period of approximately 9 months, tests were run in the archive using a variety of machine learning techniques to search BBC television programmes. Made by Machine's presenter, Hannah Fry, claims in the programme that the AI looked at more than a quarter of a million programmes available in digital form, including programmes that had been broadcast on BBC Four. It then selected 150 programmes and split them into clips. It is worth noting here that the AI could only read the metadata of digitised material, meaning that the older parts of the archive not yet chosen for digitisation were excluded from the project. Harrison states that a BBC digital archive platform called Rewind was used as the data set for the project because it was easy to access, which was a priority to the team. This mirrors my own experience of prioritising easy to access digital archives when searching for archive at the BBC. Harrison notes that a drawback of Rewind is that it only has holdings from the last 'fifteen or twenty years', thus compromising the parameters within which the machine learning took place. Further, many of the programmes on Rewind looked at by the AI were archive programmes containing remediated archive material assembled according to the value judgements and agenda of the programme-makers. Thus, the data set was not entirely neutral.
Four different techniques were applied to the clips to make four mini-programmes; the first programme was created by using image recognition, the second by using word and subtitle recognition, the third by monitoring activity on screen and the fourth by using a combination of all of these techniques. To make the programmes, the AI chose clips that resembled each other visually, shared key words or had similar levels of screen activity and edited them together. Although the process was much faster than human capability, with the machine scoping content in days that would take a human years to watch, the resulting programmes are surreal and sometimes nonsensical. The programmes also highlight the fact that the AI is not infallible. Mistakes are made with cataloguing, editing is jarring and Harrison notes that during test runs for the project, algorithm loops would tighten and get stuck: we would end up where the algorithm would just get obsessed with one thing. And so we'd end up with an hour of television of taxicabs because it just became overly obsessed with a single subject or a single image. And so it was an interesting thing of tweaking the algorithm (2021).
Harrison notes, 'it's quite funny how you know it's not as good as us, but it's also not dissimilar to us'. This statement foregrounds the relationship between human and machine; parameters for algorithms are created by humans and therefore influenced by human subjectivity. Even machine learningwhereby computers work out parameters for themselvesis still taught by humans, therefore reinforcing whatever human biases their teachers have. The clips in Made by Machine might have been chosen by AI, but the programmes it chose from and the parameters within which it operated were man-made. This is evidenced in Harrison's account of the process of setting the parameters for the project. He recalls 'we narrowed our field to the dominant genre of the channel, which was factual documentary', analysing programme synopses to see if they contained terms which were 'typical of BBC Four programmes' (2021). Teaching the AI to look for programmes with what Harrison calls the most 'BBC Fourish' qualities highlights the human values at play, revealing which aspects of British culture the channel chose to present and represent in original and archive programming.
However, a brief analysis of the contents of Made by Machine raises cause for alarm; only one Scottish accent is heard with no Welsh or Northern Irish accents present, and there are more Star Wars (1977) characters than people of colour featured in the clips. As John Berger reminds us in Ways of Seeing, we only see what we choose to look at and 'the way we see things is affected by what we know or what we believe' (Berger, 1972: 8).
This matters because what the AI looked at in the archive and recognised as typical of a BBC Four programme is a reflection of the taste and agendasthe ways of seeing-of the BBC's controllers and commissioners. And as Lenny Henry and Marcus Ryder point out, there has been historically, and still is, a lack of diversity in these jobs at the BBC (2021). Anamik Saha contends that this leads to programmes being commissioned 'to conform to mainstream, white sensibilities' (Saha, 2012: 431). The content of Made by Machine appears to support this argument, presenting a version of British cultural heritage which is predominantly white, male, able-bodied, middle-class and English. This is problematic because the kind of programmes that are commissioned and saved for the future in the BBC's archives reflects the broadcaster's ideas about cultural worth. Made by Machine hints at a troubling future for BBC Four; a canon of 'BBC Fourish' content risks BBC Four becoming the custodian of an archive of absences, repeating and recycling content which does not reflect Britain's multicultural society.

Conclusion
The BBC's decision to make BBC Four an archive channel might be a logical solution to economic problems, but relying completely on remediating the digital contents of its archive runs the risk of the channel offering the public a homogenised and inhibited story about the nation and its past. Devoting more time and resources to allow a deep dive into the archive to bring to light stories that had not previously been judged worthy of digitisation and remediation might help to solve the problem, but new content is also needed in order to expand the meaning of 'BBC Fourish' to include a diverse range of voices and experiences. This in turn will fill the archive with content to give future generations a richer understanding of the past. Harrison claims that 'a lot of time, people are looking for content with which they already have a relationship' (2021). But the experiment with AI in the archive indicates that without new, diverse content to bring fresh perspectives, the material repeated on the channel risks excluding viewers who do not see themselves reflected in the BBC's version of the past.

Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/ or publication of this article.

Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.