Effective and Ethical

by Sarah Bender

In early 2023, a student researcher approached STEM Librarian Haoyong Lan with a problem: The student wanted to gather articles for a literature review and had asked ChatGPT for sources, but half of its suggested references didn’t exist.

Lan wasn’t surprised. He had been tracking the rise of generative AI tools like ChatGPT and knew they tended to “hallucinate," or produce false information with confident authority. “That moment made it clear,” he recalled. “Users don’t just need AI tools — they need the skills to question, verify, and refine AI-generated results.” 

Since then, Lan has prioritized AI literacy efforts, equipping students and researchers with the skills to navigate generative AI. He has worked with specialists around the world to increase his own expertise in the field and help others master AI literacy. At the same time, he has led practical AI initiatives to make the technology more accessible to the CMU community.

Assembling Worldwide Expertise

In order to better understand the evolution of the field, Lan has joined forces with several groups of academic librarians and other professionals from around the world to share knowledge and develop recommendations.

In 2023, Lan was appointed to the International Federation of Library Associations and Institutions Artificial Intelligence Special Interest Group (AI SIG). AI SIG members are tasked with exploring AI innovations and concerns from a libraries-focused perspective.

Ultimately, the group seeks to provide an international platform to increase AI awareness, education, best practices, and literacy. Shortly after convening, they published a working paper designed to guide libraries in developing a strategic response to AI, and they continue to share updated information with library professionals as new developments emerge.

Then, when Associate Dean for Academic Engagement Nicky Agate received an IMLS grant to support her Project on Open and Evolving Metaliteracies (POEM), she recruited Lan as a part of her team as well. POEM is a peer-reviewed collection of teaching resources used by high school and college instructors and librarians, focusing on three topics: understanding AI, working with data, and navigating media and disinformation. Lan serves as co-editor for the topic focusing on AI literacy.

In his role, Lan works with co-editor Allie Tatarian, research & instruction and data librarian at Tufts University, and 10 additional keyword editors. Their task is to curate the most relevant open educational resources about generative AI literacy concepts and frameworks — everything from in-class exercises or digital projects to executable notebooks, videos, or podcasts. Once assembled, others can then reuse the content in their own classrooms, and adapt or modify the resources for a more appropriate fit. The team aims to launch the collection in Spring 2026.

Finally, as principal investigator for a multi-year Ithaka S+R project, Lan is uncovering how AI might reshape automated life science lab research. He conducted structured interviews with researchers from the Mellon College of Science to examine what kind of role automation plays in their experiments. By exploring researchers’ perspectives about generative AI, his team will identify new ways to provide specialised AI support that elevates groundbreaking research and its impact.

Making AI Work for You

During a Spring 2024 workshop designed to help attendees construct a successful prompt, Lan posed a challenge to participants: Ask ChatGPT for the implications and applications of AI in higher education, then tweak your prompt to get a more accurate and detailed response. Students got to experience firsthand how small changes — like specifying a date range or asking for citations — improved the tool’s output.

This kind of prompt engineering is a key skill Lan says is crucial to enhancing the relevance of outputs that generative AI tools can produce. In order for users to successfully utilize a tool, they must be able to clearly articulate what they’re looking for — and the Libraries can help them learn how.

“Librarians have spent decades teaching students how to refine search queries,” he explained. “Now, we’re doing the same thing for AI prompts.” His expertise on this topic led to a peer-reviewed article in October 2024, detailing how academic librarians can train users in AI-assisted research.

Naturally, another topic Lan emphasizes for users looking to achieve AI literacy is AI ethics. Here, his advice shifts focus from “what AI can do” to “what it should do.”

“AI isn’t neutral,” Lan explains. “It reflects the biases of its training data. That’s why users need to question everything, even when the answers sound convincing.”

Driving Innovation at CMU

Lan is also leading practical AI initiatives to get the CMU community more engaged with AI.

One of his long-term projects is an agent-based, AI-powered chatbot for the Libraries to provide more comprehensive coverage for the Libraries’ reference services. Today, Lan is working with colleagues and the Libraries’ student employees to test the chatbot and see where improvements can be made.

The goal isn’t to replace human expertise, and the chatbot will have additional functions that can redirect users to a chat with staff or librarians if their question requires more help. But this way, the CMU community can receive immediate assistance overnight or during the holidays, when faculty and staff aren’t available to respond.

In addition, Lan introduced several generative AI options offered by the Libraries that the campus can use, including research tools Scopus AI and Scite. He also hopes to transform his workshop series into an official course, with a more comprehensive syllabus and additional time devoted to developing trends.

“As each discipline develops more specialized tools with both new benefits and new pitfalls, AI literacy skills will continue to be essential to successfully navigating this evolving landscape,” Lan said. “Whether you’re a casual user, a researcher, or somewhere in between — the difference between success and failure is not just using AI, but understanding it.”

Feature image John Schnobrich on Unsplash.