This blog post is by Rosa Méthol, finalist for the DPC award for Most Distinguished Student Work in Digital Preservation
Having been shortlisted for the most distinguished student work prize category as part of the Digital Preservation Awards for 2026 is an honour! Thank you to those who have taken the time to engage with my work in this way.
Learning of the shortlisting has caused me to reflect on my dissertation, which I submitted as part of my Masters course in Archives and Records Management at the University of Liverpool in 2024. The dissertation was about the future ethical potentials of introducing Facial Recognition Technology, or FRT, into search for photographic or audiovisual collections. The aim was to combine theory with direct conversations with archivists in order to advance ethical debates as this technology becomes more prevalent. Although some archivists around the world have begun experimenting with FRT, little has been written about its ethical application in this context.
I chose the topic to have a better understanding of AI, an understanding which felt like a necessity in a technological world that sometimes feels like it could run away if we don’t hold on. I had been working in public libraries for years before training as an archivist, and had seen the cuts to funding which led to closure of services that had previously been considered essential, with solutions to the lack created sometimes including non-human replacements for library tasks that were deemed as no longer needing a face: swipe card systems to open up branch libraries that no member of staff was present at and machines for every stage of book reservation, retrieval and return. Sometimes no solution to a lack-of-person was offered and services were simply shaved a little bit more from every angle, a steady snipping which will eventually result in a shape that doesn’t look like a library at all. Automation in this case felt like a concession: even if we can’t have the people, at least we can keep the thing running. But it is a complicated concession to have to make.
A central problem for archives is that human time and labour will always fall short of the sheer size of collections, making it difficult for users to find what they seek. Machine learning excels at mass data processing and retrieval, meaning archives are exploring uses that could alleviate this constraint. Since training in archives, my personal orientation to the world of AI has come from a sense of collective responsibility to develop an approach to it in this field, however this task is complicated because of the fact that humans all feel quite differently about things. We are not the decisive, reasonable-sounding chatbots who would be able to figure out a plan in seconds if prompted, and so even though archivists can often be quite pragmatic, most of us also have quite mixed feelings about AI and its potentials, which is what I wanted to probe at a little more by interviewing archivists on this topic.
What I also wanted to ask was: in a world where archives are addressing their colonial histories with decolonisation initiatives and anticolonial projects, what place do technologies that have been trained in a way which makes them racially biased have in our systems? FRT in particular is a dual use technology, meaning that it can be used for good, for example as part of one of the projects I referenced in my dissertation From Numbers to Names which uses FRT to search image archives of Holocaust victims and survivors and return them to the care of surviving family members. Layers of non-textual searching could also open collections to non-speakers of the language of an archive, and to other groups who struggle with text-based access. However, the same technology can also be used by law enforcement in a way that disproportionately damages the lives of racialised people, and it is used in the automated weapons that are now being deployed as part of a genocide. Because of the way that it has been trained by humans, FRT reflects human biases, showing greater inaccuracies for women, Black people and people of colour. Its opaque, black-box decision making cannot be interrogated like human decision making. Because of this, FRT usage in archives risks perpetuating colonial violences, yet equally it could also support research into colonial histories by making collections more accessible.
My focus groups with archivists revealed an openness to technologies that broaden access, alongside a mixture of apprehensions. Some of these were focused around efficacy, for example the potential for paradoxical information loss through an algorithmic failure to match combined with automation bias (a human sense that that the machine can’t be wrong). Archivists also raised concerns about unethical big tech practices, the need for explainable AI, issues around privacy, consent, data protection, public trust, but also the potential for increased workloads and the unintended consequences of FRT experimentation: all the stuff that could happen which we haven’t thought of yet.
Both the literature and focus agreed that careful, gradual experimentation is preferable to rapid adoption. In my conclusions, I recommended controlled archival experimentation before introducing FRT, enabling co-designed systems tailored to practical and ethical needs, alongside archive-specific guidance (which is something that has since been developed further in UK archives via the Archives and Records Association’s AI Preparedness guidelines for archivists). The question of applying racially biased technology to archives built on colonial legacies is huge: in which ways could FRT help or harm decolonisation efforts? I feel that experimental projects should foreground this issue. Finally, because legislation lags behind AI development, archivist-led focus groups could help to shape future policy. Measured experimentation, thoughtful governance and explicit attention to anticolonial aims are essential as archives consider use of FRT in their collections.