History Factory has partnered with Pixel Acuity for our digitization needs, and we are impressed with some of the potential uses of artificial intelligence. We have co-authored the following article with Pixel Acuity to explore this emerging technology.
Artificial intelligence and machine learning (ML)—the automated learning of AI that does not require programming—possess a certain allure. While AI is widely used across social media, manufacturing and security, to name a few applications, its potential has not yet been realized in scientific and cultural heritage institutions, or in archival science. As we strive to broaden its applications, we realize that understanding what AI is may not be as important as understanding what it can do.
For example, a hypothetical museum or archive has 500,000 records related to the provenance and ownership of a collection. It is planning to digitize the records—letters, receipts, lists, photographs and a number of other types of documents—not only for internal use but also for public research. Instead of committing an inordinate number of hours to organizing the data, the organization could use a properly trained AI model to quickly and efficiently sort through the records. A subject specialist would review any anomalies and continue to train the AI model by confirming or rejecting its work. Additionally, other metadata could be organized into relevant groups, and any time an anomaly is found, it could be flagged for review. Dates, names, object types, keywords, points of interest and other visual elements could be further defined to train the AI model/database.
As long as the documents have a reference such as an accession number, the process can be automated and tracked back to any related record. Harnessing this technology and process could save thousands of hours, allowing employees to devote time to tasks that can only be done by humans.
Training AI is not limited to basic metadata and document type. Other examples of trainable identification include artistic style, time period, artist, medium, content, color, brush strokes, differences between graphite and ink, coffee stains, mold, and any other visual characteristic of an object.
You might ask additional questions such as “What else can it do?” and “How much does it cost?” We are just starting to fully grasp what is possible, and how we can best implement AI in academia and the sciences.
Balancing the novelty and the value of AI has a number of challenges. We must define client requirements, budget, collection type and intended use. Additionally, we must define models and other potential use cases as we continue to develop and expand our services for future clients.
If you are interested in contributing to the scientific research on machine learning and artificial intelligence and want to explore their capabilities, Pixel Acuity has recently announced the Pixel Acuity Artificial Intelligence in Cultural Heritage Exploratory Research Grant (PA ArCHER Grant for short). The grant is available to all private, commercial, nonprofit, individual and government institutions with digitization programs in North America and associated territories.