Recently, I had the opportunity of attending the fifth Fantastic Futures Conference hosted by the National Film and Sound Archive in Canberra. This year’s event was the first of this conference series held outside Europe and North America, nonetheless, the representation from institutions around the world was remarkable. Representatives from the British Library (next year's host), the Smithsonian, the Getty Research Institute, Stanford, Archives of New Zealand, and the National Archives of Australia gathered to share learnings, ask questions, and discuss the challenges encountered - and particularly on the conference's theme of AI in collecting organisations.
As a first-time attendee I was impressed by the quality of papers/speakers, and the organisation of the conference! While I have been aware of AI4LAM, the organising body, for some time, I have really gone down the rabbit hole and as such, I found Fantastic Futures 2024 an inspiring intersection of the tech sector, educational institutions, and the GLAM industry. Participants and attendees alike were looking forward and approaching various business problems with innovative methods. These challenges ranged from improvements in OCR, transcription of audio and visual material, using AI to assist with archival appraisal, to the imaginative repair and preservation of digitised films using AI and user illustrations.
There were three primary veins of talks:
- Practical applications of AI to advance workflows and record accessibility.
- Quandaries over the ethical implications of using AI and the data a model is trained on.
- Really cool stuff.
Many speakers discussed their use of Whisper - OpenAI’s speech-to-text AI model. It’s no secret that AI has biases and hallucinations, but it was fascinating to hear specific instances called out. Grant Heinrich of the National Film and Sound Archive discussed NFSA’s efforts to teach a tool to better recognise the Australian accent. While there were many learnings within this project, ultimately the tool was no better at recognising Australian accents at the end of the project than it was at the beginning. Kathy Reid, PhD Candidate from Australian National University apologised to all Scots in the room as she noted the Scottish accent has the lowest accuracy rate of any for AI speech to text translators. Other speakers noted that when using a speech-to-text AI, phrases such as “Like and subscribe” found their way into transcripts demonstrating that even AI has become infested with social media tropes. While AI will make many jobs easier, it was a common statement that at this point, AI will always require a degree of human review and correction.
A continual undercurrent of many talks was examining the source of the data used to train AI models and what the subsequent implications were and could be. The session, Peter-Lucas Jones & Kathy Reid – In conversation, was particularly thought-provoking. Peter-Lucas Jones, CEO of Te Hiku Media, has asked if AI is yet another tool of invasion. Further, through its use, who are the forgotten ones and who is left behind? These are questions that we must remember to ask as we proceed in this field.
One of the talks that was a pleasant surprise was Reanimating and Reinterpreting the Archive with AI: Unifying Scholarship and Practice, presented by Benjamin Lee (University of Washington), Kath Bode (Australian National University), and Andrew Dean (Deakin University). While the title did not give it away, did you know that author J.M. Coetzee used generative computing in an attempt to conquer writer's block in the 1960s? In the mid-1960s, J.M. Coetzee was a computer programmer. He built a lexicon of 800 words and wrote an algorithm that would compose poetry - essentially an SLM - Small Language Model. Facts like this disrupt the concept that AI is a new and freshly emerging field.
While this was my first time engaging with AI4LAM, it certainly won’t be my last. I look forward to staying engaged, continuing to learn, and contributing to innovative solutions for the GLAM sector.
If you have the time, I highly suggest checking out the Fantastic Futures 2024 abstracts here. AI4LAM has some great resources and you can watch past webinars on their YouTube channel.
If you want to start a conversation with us about anything in this blog, then feel free to contact me or follow us on our social media platforms LinkedIn, Facebook or Instagram
Sarah