In January, Microsoft announced the first change in 30 years to its Windows keyboard, adding a new button for its artificial intelligence (AI) chatbot, Copilot. It’s not surprising, but it is symptomatic. AI is everywhere.
Scholars Michael Barrett and Wanda Orlikowski note in a March 2021 paper that technologies deployed at scale have both constructive and problematic outcomes. As library decision makers position the library as a source of advice and expertise, as they determine the products and services to invest in, and as they consider the welfare of their own colleagues, both constructive and problematic elements will have to be understood and weighed.
As we undergo this process, we need to recognize that development is unpredictable. For comparison, think of how the invention of the cell phone has changed our behaviors in ways we could not have anticipated. Who imagined the importance of something as seemingly simple as the selfie and the role it would have in travel and one’s mental wellness?
“The major elections that will be held around the world in 2024 will be the first in which the synthetic powers of generative AI are widely available.”
While the effects of AI can seem magical, it is also important to remember that real people are making real product and service decisions. There are real questions about privacy, reuse of data, replication of copyrighted materials, and so on. These are influenceable practices, through purchasing or policy action and libraries can be purposeful about directions they would like to see.
Libraries have no option but to embrace this messy middle, working under conditions of uncertainty and change, advocating for the interests of both consumers and creators.
So, what does this mean in practice?
Experience matters. Experience informs understanding. Without using different AI services, it is difficult to understand discussion about the impact of different prompting approaches, retrieval-augmented generation (likely to be deployed in lots of informational and publishing products), hallucination (presenting false information as if it were true), and so on.
Transparency, awareness, and empathy. Given the accumulated stresses of the last few years, and uncertainty about social and technical developments, empathy, education and appropriate transparency about planning and direction are now even more critical in the workplace.
Research skills. The major elections that will be held around the world in 2024 will be the first in which the synthetic powers of generative AI are widely available. Fake reaches new levels with synthesized news, images, and videos circulating freely. Law cases about copyright and memorization are growing in number. At the same time, AI companies are making deals to feed high-quality materials into their tools’ training. AI is more fully integrating in workflow, writing, and entertainment apps. Every search box will be AI-enabled. The research skills library professionals need—and will teach—will continue to expand.
Ethics. The ability to collect and process data, make new connections, propagate misleading or harmful data, and use products developed in ways not compatible with a library’s values all create a need for careful and informed attention. Educators are considering what is aligned with best practice and established values in research and learning, and organizations are considering making ethics frameworks more explicit.
Individual libraries have limited influence in this environment. Organizations such as the American Library Association, Association of Research Libraries, International Federation of Library Associations and Institutions, National Information Standards Organization, and Urban Libraries Council provide a place for the development of policy, standards, and practices to aggregate and scale library influence. They should coordinate their work in key areas, including:
- Advocacy and policy. Libraries represent public, scholarly, and cultural interests. They are committed to principles of open learning and equity. Library organizations should advocate for those interests and principles in relevant government and industry policies and proposals.
- Purchasing and convening. New procurement questions arise as library vendors implement varieties of AI. Which large language model is in use? Are additional measures in place to mitigate undesirable behaviors? What data is being collected, and how is it used? It would be good to have guidelines, advice, and briefings that address these topics. Going beyond this, does it make sense to convene industry groups to develop shared views of technical approaches or policies?
- Open access. It is now possible to summarize and synthesize results from models trained on large bodies of scholarly literature, as well as on rich data from expertise profiles, research analytics, workflow services and so on. Several companies are well placed to offer services here, but to whom and on what terms? Will a new phase of open advocacy emerge, to explore equitable access to such resources?
- Library data and training. What is the role of structured library data, shaped by years of intellectual and professional investment, such as finding aids or LibGuides, that may be harvested by those building large language models. One might also imagine these materials being used to train community models, which could be used in custom services.
These and other questions are far-reaching enough to warrant concerted library attention and action. The organizations I mention above scale capacity and influence for the library community. It is time for them to confer and help determine what sort of future we want to create.
This piece draws on the LorcanDempsey.net blog post “Generative AI and Libraries: 7 contexts.”
