In 2020, the Institute of Museum and Library Services awarded grant MC-245791-OMS-20 to the Computer History Museum. The $50,000 federal grant—augmented by $15,000 in museum staff time and $20,000 from the Gordon and Betty Moore Foundation—funded a one-year experiment to test whether commercial artificial intelligence could automatically index and transcribe the museum's massive collection of historical oral histories.
An Archival Experiment in the Azure Cloud
To make its vast archives accessible globally under an initiative dubbed OpenCHM, the museum assembled a test corpus focused on the history of Xerox PARC and artificial intelligence. Rather than building custom models, the project team ingested historical video, audio, and scanned documents into Microsoft Azure, utilizing out-of-the-box Cognitive Services and Video Indexer. The cloud pipeline analyzed raw video frames for face recognition, processed audio recordings for sentiment, and parsed document text to generate searchable metadata across thousands of minutes of historical footage.
Translating Pioneer Records and Technical Omissions
The automated system delivered immediate multilingual capabilities. The pipeline parsed a video recording of Ethernet co-inventor Bob Metcalfe, generating a timecoded transcript automatically translated into Chinese, while a transcript of computer graphics pioneer Alvy Ray Smith was rendered into Hindi. Yet technical oversights hampered the project's execution. Because requirements for importing existing human-curated collection metadata were omitted during early database setup, attaching that legacy information later proved cost-prohibitive. Budget and schedule constraints also forced the team to strip out a key discovery feature: an interactive networked graph visualizer meant to map connections between search results.
The Gap Between Insiders and External Scholars
A post-launch survey revealed a sharp divide between internal staff and outside researchers. Less than half of internal museum staff rated the machine-generated topic keywords as useful, criticizing their inability to infer context beyond exact words in the text. External academic researchers, however, overwhelmingly praised the automated metadata, viewing machine-generated keywords as a major upgrade over unindexed archives. The off-the-shelf models also demonstrated notable associative links for high-profile figures, correctly connecting Steve Jobs and Gordon Bell to the National Medal of Technology and Innovation, and Nandan Nilekani to Chairman of Infosys.
The Computer History Museum established that automated transcription will be its top priority for global access, provided future models are custom-trained on original museum transcripts and explicitly labeled alongside human curation.
Cover: AI-generated editorial illustration; not a documentary photograph.
Source: A Museum's Experience With AI
SIGNAL TITANS.jpg){kind=link}
Leave a Reply