SUBSCRIBE NOW!

Vol. 44 No. 8 — October 2024

TECHNOLOGY & POWER

Those AI use cases piloted by various libraries and archives are still valid. Various AI models, services, and platforms from commercial providers, such as AWS, Azure, and Google Cloud, can make AI experiments even more approachable. However, the recent AI boom didn’t necessarily increase the number of AI/ML pilots and experiments in libraries that continue to test and explore those use cases. Instead, it turned libraries’ attention toward library patrons and their increasing use of AI products designed for personal use.

This shift is understandable given that the greatest recent sea change in AI was driven by the rise of the large language model (LLM), a type of AI model that performs various NLP tasks, such as recognizing, summarizing, translating, and generating text; answering questions; and carrying on a conversation. As with other fields, libraries weren’t anticipating LLMs to advance so quickly to produce general-purpose chatbots, such as ChatGPT, Copilot, and Gemini. When these new generative AI (gen AI) tools were adopted as consumer products, library professionals found themselves struggling with advising library patrons on their best use.

Naturally, many librarians started examining the benefits and drawbacks of those tools and promoting AI literacy at the same time. The advice and tips that libraries provide related to the use of these AI tools cover various topics, such as how to improve an AI chatbot’s performance by creating more effective prompts, how to properly cite AI tools, the ethical use of AI tools in academic writing, and AI-related copyright and publishing policies. Those who are interested in further details can explore many AI guides created by libraries. A list has been also complied by the Florida International University (FIU) Libraries as part of its own AI guide, “Artificial Intelligence Now: ChatGPT + AI Literacy Toolbox” (library.fiu.edu/ai/libguides).

EMERGING APPROACHES TO GEN AI IN HIGHER EDUCATION

While libraries are working hard to determine and offer useful and timely information and guidance related to AI tools for library patrons, the parent institutions of these libraries—universities, schools, and corporations—are focusing their efforts on the issue of how to provide AI tools for their communities’ use to their organizational advantage. At the same time, they are also trying to address some thorny questions, such as how institutions and individual users should balance the benefits and the risks associated with using AI tools and what kind of precautions and safeguards institutions should provide regarding the data and information that their users will enter into the AI tools.

An increasing number of colleges and universities—such as University of Tennessee, Knoxville (UT; oit.utk.edu/ai) and Indiana University (IU; kb.iu.edu/d/bing)—are licensing gen AI tools. Some of these organizations have a certain level of data protection as part of their licensing agreement, while others may not. For example, IU informs its students, faculty, and staff that the Copilot licensed by IU is approved to interact with data classified up to and including a certain level of IU-internal data for their users, if they are logged in with their IU account credentials when using it (kb.iu.edu/d/bing#precautions). UT, Knoxville explicitly discourages its users from having conversations with UT Verse, which is UT’s gen AI chatbot geared toward internal audiences, university business, and research, with regulated information such as PII (personal identifiable information), PHI (protected health information), and FERPA (Family Educational Rights and Privacy Act; oit.utk.edu/ai/ut-verse).

EARLY ADOPTION AT UNIVERSITY OF MICHIGAN

Naturally, not all institutions are in the same place in their adoption of AI. Even though it has been less than 2 years since gen AI captured everyone’s attention, some institutions have already moved farther ahead than others. The University of Michigan (U-M), where I work, is one of the early AI adopters among higher-education institutions. It has built its own custom AI platform and launched its three distinct AI services—U-M GPT, U-M Maizey, and U-M GPT Toolkit—in September 2023.

U-M GPT is the first and most basic service that the U-M Gen AI platform provides. U-M offers this AI chatbot service for free to everyone in its community, making it equally accessible. U-M GPT works very much like commercial AI chatbots. Equipped with robust data protection mechanisms, however, the U-M GPT environment is secure and private for users and keeps all user inputs within the platform, which is not the case in commercial AI chatbots.

U-M Maizey and U-M GPT Toolkit are the next-tier AI services that the U-M’s gen AI platform offers. U-M Maizey allows users to build custom AI chatbots with their own data in a secure and private environment in a GUI (graphical user interface) environment. U-M GPT Toolkit provides API access to AI models, so that U-M students, faculty, and staff can build their own AI applications and power them with various AI models. U-M Maizey and U-M GPT are currently free but are expected to become fee-based services to U-M users in the near future. Equitable and secure/private access to AI is what sets apart U-M’s gen AI platform and services from those of other higher-education institutions (or from commercial vendors) to date.

U-M’s generative AI website (genai.umich.edu) includes information about these three U-M AI services along with many rich, AI-related resources that are relevant and useful to all users of AI tools, particularly at colleges and universities. While these resources were developed for the U-M community as the primary audience, they are broadly applicable to anyone interested in AI and are publicly available. The topics addressed in the Generative AI Resources section of the U-M Generative Artificial Intelligence website (genai.umich.edu/resources) range from AI guidance customized for three different groups (students, faculty/instructors, and staff), commercial AI tools for various purposes, and AI resources for academic researchers to include AI prompt literacy.

Among these, I would like to highlight resources on prompt literacy. Prompt literacy refers to the ability to effectively engage with language models, comprehend the generated output, and accurately predict or influence the responses. It is a highly relevant topic for libraries whose mission is helping people to become better equipped at navigating a complex information environment. You can browse the U-M GenAI Prompt Library (genai.umich.edu/resources/prompt-library) and take the freely available Generative AI Prompt Literacy course that U-M developed (umflintpd.pdx.catalog.canvaslms.com/browse/ode/courses/generative-ai-prompt-literacy).

LOOKING AHEAD

In time, more universities may develop a similar AI platform to U-M’s. But not all colleges and universities will be able to do so due to the resource-intensive nature of such an endeavor. Regardless, many are likely to share the same goal that U-M’s AI platform is trying to achieve: to provide equitable, secure, and private access to AI tools and services for its community. In this context, libraries can work with their parent institutions to advocate for facilitation and find ways to enable such access to the extent possible in their own environments. Since many institutions are currently working on developing their AI strategies, this is an opportune time for libraries to be actively engaged with that process.

Gen AI technologies are evolving at a rapid pace, and we are seeing that some products come and go within even less than a year’s time. Given this, the best approach for libraries at this point is likely to be staying open-minded, focusing on the fundamental aspects of the library’s mission, and paying attention to how fast-changing technologies may be aligned to advance that mission in the longer term, rather than trying to judge prematurely how they can immediately serve their interests.

Librarianship is a profession that keeps a sharp eye on the societal issue of equal access to information and knowledge. It is not surprising then to see many library professionals raising questions regarding the ethical and social justice-related concerns in the use of AI. They are also voicing strong concerns about the negative environmental impact of the high level of energy consumption required for training LLMs. Although these are not necessarily new questions, the fact that they are now gaining more attention with the rise of gen AI is a good thing, and a deeper understanding of these issues in the library profession is highly desirable.

Libraries can look for ways to address and mitigate these concerns in their engagement with AI while continuing to work to serve their communities without completely rejecting AI. For example, not all AI applications use language models. And not all language models need to be large to be useful. There are and will be more possibilities opening up as AI and gen AI technologies further develop. The direction of such development will be set by those who actively engage with AI rather than by those who decide to distance from or reject it.