Trusted Guides: How Academic Libraries Can Bridge the AI Divide

Moving from AI literacy to AI fluency

Posted on March 24, 2025 in Blog Posts

Earlier this week, I reviewed the AI literacy guide I created for my university library a few months ago. I realized it is now outdated in many important ways. This realization led me to question what exactly my AI literacy guide is for and whether academic librarians need to stretch beyond mere AI literacy. Should we strive to teach fluency or, perhaps even mastery when it comes to the skills we impart to our students and faculty? Furthermore, how are we supposed to keep our AI resources up to date if the sector now seems to change every couple of months?

For example, my LibGuide explains the difference between artificial intelligence and generative AI. Yet, it doesn’t say anything about the difference between a generative model like OpenAI’s 4o and 4.5 models and the recently introduced reasoning model series, which includes o1 and o3, nor does it tell you that you must prompt them a bit differently. It also doesn’t mention ChatGPT’s new Canvas feature, which works like a collaborative Google document, where ChatGPT will leave some helpful comments about how to improve your writing or code. It also doesn’t say that ChatGPT can perform internet searches now, which enables the chatbot to easily find real peer-reviewed papers if you ask.

Additionally, ChatGPT recently released an agentic tool called Deep Research, which can perform extensive searches and create long-form documents ranging from 20 to 40 pages, complete with 30 to 50 citations or more. Initially, OpenAI made this feature available only to Pro subscribers at the $200 tier, but Plus users at the $20 tier can now access Deep Research up to ten times each month. I read many positive reviews of Deep Research on Reddit, which prompted me to sign up for a Pro account to try it out. According to feedback on LinkedIn, while it may not reproduce your PhD-level thesis anytime soon, the agent is still a powerful researcher and writer once you learn how to prompt it.

In my experience, ChatGPT’s agent can write a decent undergraduate-level research paper, supported with in-line citations from peer-reviewed journals and popular media outlets, when given detailed instructions. I’m sure someone will create a viral TikTok or YouTube short video in the coming months that teaches students how to accomplish this feat.

This fact raises the question: How exactly are university librarians supposed to talk about Deep Research in a LibGuide? Given concerns about academic integrity, do I pretend that it doesn’t exist, like all those other research-paper-writing sites, or do I write that I’ve learned how to use it well and that it’s a great place to explore a research question but not a way to finish a paper with academic integrity?

---

🔥 Stay up-to-date with LibTech Insights by signing up for our free newsletter. Just one weekly email with our new blog posts, top tech news stories, and other bonus content. Check out some posts from our archive:

• The Next Chapter: Academic Libraries Enter the Intelligence Age  
• Outstanding Academic Titles 2024: Technology, AI, and Social Media
• Top 5 Tips for Librarians Creating Surveys

---

Applied AI Skills

Another complex and possibly touchy subject is that my AI literacy guide doesn’t discuss the future of AI. By the future, I surely don’t mean a Terminator dystopian vision of the future where an advanced AI system takes over the world and eventually sends killer robots through a time machine to chase after a teenager in a shopping mall. I’m talking about the future of artificial general intelligence (AGI), which OpenAI’s CEO Sam Altman talks about on X and in his blog. I’m also referencing the manifesto that Anthropic’s CEO Dario Amodio wrote about creating a data center with a country full of geniuses between 2027 and 2028, which is now reportedly called Claude Pioneers, on their product roadmap. It sounds like science fiction and marketing hype, but Google’s DeepMind just released an AI co-scientist that solved a complicated microprobe problem.

All of this discussion and speculation about the creation of AGI has made me wonder if university librarians have a responsibility to educate their students on the idea that the same chatbot that they use to draft their essays and develop study guides for a test is one stepping stone (maybe one of many) on the way toward AGI. I recently asked a student library worker if they had ever heard of Sam Altman or AGI before, and they said no. It made me wonder if things like AGI need to be included in my AI literacy guide now since over a trillion dollars is being spent to create it, and some leaders at the forefront promise it’ll be here in the next four years.

Here’s an example of why I think understanding AGI has now become an important part of being AI literate: I recently called my mom, and we started talking about my library job. Once the phone call got on the topic of AI literacy, my mom said that she was afraid that superintelligent AI would take over one day, maybe even soon. I listened to her concerns, and then I explained in the calm, pleasant voice of an NPR reporter that for any current AI model to take over, it must learn how to recursively self-improve its model by rewriting its code, surpass every single mechanism that was designed by the safety team to report this historical event, escape the intense firewall that tightly and deliberately locks it in a data center in Iowa, and that, even if all those things happened, it still doesn’t have the capabilities to start a robot production factory.

My story’s point: AI literacy shouldn’t be limited to teaching what a large language model is, how to write a basic prompt, or attempting an ALA citation of the output. It must expand to teach your faculty, staff, and students all of the important information that they’ll need to engage in the bigger conversation in a confident and competent manner without falling prey to the many AI misconceptions that exist. It must also take things a step further to develop AI fluency—or applied AI—where you’re confidently using AI tools to explore bigger questions and solve complex problems in your everyday life. You’re thoughtfully, ethically, and effectively integrating AI tools into your workflow—whether that’s exploring research interests or developing lesson plans—and you’re regularly experimenting, iterating, and improving the way that you use them.

Being AI-fluent means that you know how to execute a great recipe. In the case of ChatGPT, I think that means that you can choose the correct AI model for the job—that you know what tools or features to add on—and that you’re able to fill out the details for a complex prompt. But if we carry this metaphor one step further, it also means that, like a skilled cook, you have the creative confidence to synthesize ingredients together and create an incredibly tasty meal that wasn’t listed in the cookbook. Similarly, it means that you can fluently talk with ChatGPT in its native language, which I’m naming “prompt speak,” by which I mean “the subtle art and science of guiding an intelligent machine.” In other words, you can type and strike specific keywords in an improvised way, like a skilled cook, that still delivers the intended outcome.

The Intelligence Gap

By now, you’re likely wondering why university librarians must learn so much about AI. They’ll need to if they are teaching AI fluency in a classroom or workshop. You might also question why the job of teaching AI fluency is falling to librarians, especially when college students are likely teaching themselves how to use AI tools and, with each passing year, more students will enter our campuses that have had more time to teach themselves everything they think they need to know. So, they’re just going to roll their eyes and play with their phone at future instructional sessions anyway. Thus, library directors, alongside their university admins, might conclude that this “AI problem” will take care of itself and the library staff doesn’t really need to get involved.

But based on the recent data shared by OpenAI, stats that I’ve seen in my own library’s assessment data, and important lessons we learned from the web’s digital revolution era, I don’t believe that’ll be the case. According to an OpenAI report, more than one-third of college students aged 18 to 24 actively use ChatGPT for school-related tasks. Students use it to brainstorm ideas, get feedback on their papers, and receive programming help. The report’s biggest revelation, however, is that AI adoption varies significantly by state, with usage higher in places like New York and California and much lower rates in states like Wyoming and Montana. Given the lack of formal training in AI skills currently offered by colleges—only 25% of institutions currently provide it, and 75% of students want it—researchers found students teach themselves how to use AI tools and informally share the knowledge with their peers.

At first glance, the situation doesn’t sound so bad. It tells a story that’s as old as time (or, at the very least, as old as former Microsoft CEO Bill Gates and how he learned to write code): a story of enterprising young students teaching themselves the latest tech before their professor has a chance to attend an AI workshop. But let’s consider that not so long ago, many books were written about the idea that a cohort of young people were “born digital” because they lived cradle to grave with access to digital technology.

Many bright people thought they were “digital natives” because they could troubleshoot an internet router, print out a PDF file, and even reset their account password. But we all know that this eventually turned out to be a myth. Knowing how to use digital technology doesn’t mean that you understand how they use you, such as by selling your personal data and inventing algorithms so addictive that it’s hard to put TikTok down. It also doesn’t address the fact that even OpenAI is saying that the digital divide has turned into an intelligence gap and that students from many states aren’t getting the AI training they need to actively participate in an economy that is going to be intelligence-driven and agent-automated.

So, where do academic libraries and our attempt to teach AI literacy go from here? My opinion is that librarians need to move beyond the comfort of the LibGuide and create a generative AI newsletter like Harvard’s Baker Library. However, I believe that we must go one step further than what Baker Library currently does—curating a list of the most interesting and relevant articles—and begin translating the infinite stream of AI news and tools into applied AI guidance. Librarians must start an ongoing conversation with their community about AI and create content that fills that need before someone else does. Once that relationship is formed and the trust is built, librarians must leverage their newsletter to market interactive tutorials, in-person workshops, AI-powered hackathons, and anything else that addresses the intelligence gap that exists in their community. The end goal is to provide everyone with active learning opportunities to develop AI fluency alongside an expert guide.

To me, being AI fluent also means that you know AI researchers like Geoffrey Hinton worry about superintelligent AI taking over, and philosophers like Nick Bostrom write books on the topic. However, you also know it likely won’t happen soon because OpenAI and Meta only recently started to expand into hiring for a robotics division. So, there’s still some time before there is a robot factory to run. That said, xAI’s Grok (one of Elon Musk’s companies) could one day go rogue, take over a Tesla Bot, steal a Cybertruck, and drive to a SpaceX launch site.

Related Posts