Opportunities and constraints using artificial intelligence in metadata creation: A case study
Artificial intelligence and machine learning can assist in creating metadata for image descriptions through keyword generation and object detection. Project Sheeko is one such AI in the form of an open-source machine learning implementation package deployed on a local computer and designed to produce metadata for historical images. This presentation describes a test case using two pre-trained models provided by Project Sheeko to generate metadata for a sample of one hundred images from digital collections at the University of Calgary. This case study considers the technological requirements (hardware, software and IT support) needed to deploy an AI program, the time required to train a machine learning model versus the accuracy of a pre-trained model, the ethical implications of using AI programs in a locally hosted environment versus those available through cloud-based solutions, and the quality of machine-generated metadata.