AI Use in Our Archival Processes

Contextualizing, Defining, and Evaluating AI

In the last few years, artificial intelligence (AI) has become increasingly ubiquitous, dominating news cycles, conversations, and business strategies, and widely embedded into software applications and workflows.

Even though AI is widely discussed, it is also often underdefined, commonly misunderstood, and impulsively implemented. The term “AI” has become a buzzword and is now generally used to describe a broad and rapidly changing range of technologies including machine learning, chatbots, generative AI, and AI agents. In many organizations, these technologies have been implemented under the umbrella of AI adoption in a top-down way as a reflexive response to every challenge, bypassing the processes necessary to understand real use cases and develop practitioner-led solutions.

The Rockefeller Archive Center (RAC) has been evaluating AI for a wide range of use cases in our archival processes and our support of researchers. As with all emerging technologies and consistent with our past practices, we believe it is important to take an intentional approach by asking the right questions before we make any implementation decisions. How well do these technologies align with our organizational mission, vision, and values? How do they interact with existing legal, ethical, and regulatory frameworks? What are we likely to gain or lose because of their implementation?

This document outlines the reasons why we don’t currently use AI in archival processes – including archival appraisal, acquisition, arrangement and description (processing), digital preservation, and access – as well as how we support our user communities who want to employ AI or other machine-assisted research methodologies in the course of their work. It does not address any potential use of AI in our business processes, and it is not intended to be an organizational AI use policy.

By clearly articulating our position on AI, we hope to support holistic thinking and action around its implementation and use.

AI in RAC Archival Processes

The RAC does not currently implement AI in any of our internal or external facing archival processes because we believe that the legal, ethical, environmental, and human costs of using these technologies outweighs the potential benefits for us. Our considerations include:

Copyright: AI companies deploy web crawlers to identify and copy original content to train Large Language Models (LLMs). These crawlers flagrantly disregard copyright protections. Our collections contain some records to which access may be restricted because of copyright or embargo periods. We always respect the copyright and intellectual property rights of our records creators, donors, and depositors, as well as any other conditions on access and use articulated in agreements with them. We never sell or make deals with AI companies to use our archival holdings as training data for LLMs.
User privacy: AI-branded features are embedded in almost every commonly used application, making it difficult for users to meaningfully opt out or understand how their information is being used. This practice is inconsistent with our commitment to user privacy as articulated in the RAC’s Privacy Policy.
Environmental impact: AI technologies, in particular generative AI, consume a disproportionate amount of the natural resources – specifically energy and water – necessary to sustain life. To meet AI’s power requirements, governments and corporations are reopening highly polluting power plants and building new facilities that will have devastating long-term environmental impacts. As articulated in our organizational values, we are committed to responsible stewardship of the natural environment as an extension of our commitment to the preservation of the collections in our care.
Dignity of human labor and reasoning: The RAC is committed to ethical archival practices that support our mission to preserve and provide broad and equitable access to the authentic historical records in our care. However, by mimicking the appearance of reasoning, AI can undermine this independent professional judgment of archivists, devalue expertise, and call into question the integrity and authenticity of the historical record itself. When driven by austerity and perceived “efficiency”, the implementation of AI often reduces the role of human labor to cut costs in the short term. It also misleads users into thinking that responses generated by AI are authentic and true, despite the technology’s well-documented production of “hallucinations” and inaccuracies.

Supporting External Users

We acknowledge that AI is part of the toolbox of many of our users, including researchers and staff of organizations who have placed their records in our care. We continue to support their use of AI by providing open licensing, bulk download, and API access to archival description and the delivery of archival records in accordance with donor agreements and copyright law.

Looking Forward

We continue to actively monitor developments in AI, its implementation in archives and philanthropy, and emerging use cases at the RAC. Any potential implementation of AI in our archival processes will be done in alignment with our organizational mission, vision and values, and designed to serve well-defined user needs.