People’s Graphic Design Archive
Briar Levit, Brockett Horne, Louise Sandhaus, Morgan Searcy
The People’s Graphic Design Archive (PGDA) is a project that makes graphic design history more inclusive and accessible. It is a crowd-sourced website that allows anyone to upload materials for the historical record, challenging other collections that are built only by those with academic training, large funding sources, or access to special resources. For the C/Change project, we explored ways AI can encourage more people to participate in shaping history.
Concept
After a preliminary phase of exploration, we focused on ways that AI could help tag items in The Archive’s collection so they can be more readily found by users. We explored the pros and cons of Optical Character Recognition (OCR) to recognize text in Graphic Design artifacts. We examined a few off-the-shelf tools for OCR and tested the possibilities for building custom tools that could learn from Graphic Design objects. Since the PGDA includes non-canonical objects and narratives that are overlooked in other graphic design histories, many items include hand-generated text, abbreviations, artful typography, languages other than English, and other expressions that challenge conventional OCR tools. Rather than be discouraged by the finding that few OCR tools could reliably “read” Graphic Design artifacts, we imagined working with “failed” OCR to recognize experimental and expressive arrangements of language, a feature of the best Graphic Design.
We explored other off-the-shelf AI models that could help identify historical graphic design objects, such as color extraction, style identification, and ways to support tagging the metadata of the People’s Graphic Design Archive collection.
We hope that these experiments will encourage conversations about potential uses of AI for Graphic Design objects that are inclusive in nature.
Throughout the process of working with others in our C/Change cohort, the staff, our advisor, and the large community of developers we consulted, we became more informed about the possibilities of Machine Learning for inclusive archiving. Even during the project, new tools emerged. Our original proposal, was speculative and too large in scope for this project, but was bolstered by research afforded by the grant.
Process
Our project currently exists as a digital prototype, ripe for identifying the correct ML developer to pursue further.
Before the C/Change project, our idea was too large and unrealistic in scope, but the process has been instrumental in testing what could be possible and more tightly defining a viable project.
Our original idea was a comprehensive “reverse-image” system that could offer text, color, style, or even insights about design history artifacts. As we honed the goals, we focused on OCR recognition tools to make tagging items easier. Better tags help researchers find better information.
Lessons
We had trouble matching the scope of the grant with available expertise. It was challenging to estimate labor for ML developers collaborating with our team of volunteer designers. We struggled to find a balance of specialized expertise with availability, interest in our project, and rate. Our grant was a set fee, yet most developers charged by the hour. Further, the process of dispersing funds to collaborators was more challenging than we accounted for, which set up confusing communications with possible developers in the beginning.
An unexpected part of this project was finding sustainable expertise.
An excellent part of the project was the discovery phase in which we met with over a dozen developers to describe the project and get feedback. This resulted in a knowledge dump of various ideas, a sprawling set of information to an already broadly scoped project. It was challenging to fit that with C/Change expectations and what would be most helpful for our project—to encourage more diversity in Archives. However, these meetings were instrumental to inform us.
Future
New tools exist for collecting art and design history, and new voices are honored. In the same way that Spotify makes anyone a DJ, GitHub makes anyone a developer, our hope is that PGDA makes anyone a historian.
Another hope is that Machine Learning and mobile technology can transform how archives and museums engage with their publics to collect and give value objects.