AI or not AI – That is the question? The impact of Artificial Intelligence on Oral History practices in the twenty-first century.
As an oral historian in 2026, discussions of my research with peers often result in insightful conversations and queries. One question is nearly always asked: will you be using artificial intelligence? Oral history is the practice of recording spoken word interviews discussing personal historical experiences to use as historical primary material. The process is rooted in human experience and how we as people who lived it understand our past. Some oral historians have embraced AI, and its use is changing our approaches to the field. AI impacts how we collect information, store it, and share it with the wider world. It is crucial for oral historians to understand how AI impacts our research. I find myself asking the question: should AI be seen as an academic tool or a detriment?
Every oral historian knows that before an interview can take place, informed consent is needed from participants. In my early days as a student, I discovered informed consent was only achieved through informing participants fully about the interview process, the implications of involvement, and any future consequences of involvement. Informed consent underpins the practice of oral history: if I were to act without it, I could be open to research-based and legal complications.
When using AI, achieving informed consent becomes extremely complicated – potentially impossible. AI programs are still developing, and it is often unclear where the information fed into these systems might end up and who might have access to it. I find myself unaware of key information that builds the foundation of informed consent. This impacts the future consequences for those taking part and the information they provide.
Scholars whose work has been made publicly accessible via the internet are experiencing similar issues. One researcher, Elinor Mazé, discusses how a research participant requested their interview be removed from all digital resources as the entire transcript was freely accessible to any who searched their name in Google. I view Mazé’s experience as a warning concerning the consequences of digital impacts on oral history. It also encourages me to argue that until such information about AI’s digital dissemination is clearly shared in full, the foundational pillar of informed consent is threatened.
When using AI, achieving informed consent becomes extremely complicated – potentially impossible.
In conversations with fellow oral historians, I have discovered a range of freely accessible AI programs that when used change how collected oral history data is analysed. Through a simple internet search I could sign up to Openpose designed to assess body language and a tool that could be applied to recorded video testimonies. Or I could easily access Whisper, created to quickly transcribe spoken word and a tool which could be used to reduce the amount of time taken by myself to transcribe interviews or bypass transcribing costs. I have also come across some scholars utilising AI to overcome challenges in oral history – previously thought insurmountable. Gerhard Dueck asserts that in using automatic speech recognition AI to train phonetic models, oral historians could document and make accessible languages with no written format.
These AI opportunities are fascinating, but my mind is drawn back to the significant issues. AI is by no means flawless: there are often inaccuracies in transcription services. These inaccuracies further distance participants’ original meanings from their transcribed recording. Additionally, the lack of information surrounding accessibility to information ingested by AI programs raises further questions about the security of private data. AI has the potential to shift and distort oral history, and ultimately changes the way scholars, researchers, and the public interpret the information oral historians provide.
Other issues I have experienced with generative AI programs include the acceptance that information provided is unquestionably true, and the alarming frequency of hallucinations. When AI insists misleading or outright false information is infallibly true, it causes confusion. I have not yet applied generative AI in my oral history research, but some academics have. Scholars searching for oral history interviews using AI have been given results which are either partially or wholly fictitious. Notably, a recent experiment was conducted by academics who fed biographical information of well-known historical figures into a generative AI to resurrect chatbot oral history sources. They experienced many instances of wrongful information, including one bot arguing Winston Churchill had never been an MP in Dundee. This is relatively amusing in a controlled setting. It is less funny when considering the value of integrity in oral history. Once again, AI raises issues surrounding consent and ethical application.
AI has the potential to shift and distort oral history, and ultimately changes the way scholars, researchers, and the public interpret the information oral historians provide.
After seeing how AI impacts oral history, my personal view remains unmoved. I would agree with any pro-AI oral historian that the beneficial opportunities are enticing. But problems surrounding its impact on informed consent, lack of clear information concerning data storage and accessibility, and the prevalence of hallucinations encourage me not to use it in my own research. These issues are further compounded by AI’s severely harmful environmental and social impacts beyond institutional walls, which further influence me to see AI as detrimental.
Puljak, Livia, ‘Letter to the editor: Artificial intelligence and large language models for interview transcription in qualitative research: competency, politeness, and ethical implications’, Journal of Clinical Epidemiology, 194:112208, (2026).
