Artificial Intelligence for Libraries

Brady Lund – University of North Texas

An Outline of Contents

• 1 – What is AI?
• 2 – How AI Works
• 3 – AI Ethics
• 4 – A History of AI and Libraries
• 5 – Library Systems and AI
• 6 – AI Literacy
• 7 – AI for Teaching and Learning
• 8 – AI for Scholarly Publishing and Research
• 9 – Building AI Solutions for Libraries
• 10 – Our Next Generation of AI Libraries

​

What is AI?

• AI = computer technologies using vast data and algorithms to perform tasks historically requiring human intelligence.
• Built on four foundations:
• Computers — billions of on/off switches storing and processing information
• Data — raw inputs (text, images, numbers) that reveal patterns at scale
• Algorithms — rule sets that transform data into meaningful outputs
• Intelligence — the ability to solve complex problems and make autonomous decisions
• Types of AI:
• Narrow AI — performs limited tasks (e.g., Siri, voice assistants)
• Generative AI — learns from data to create new content (e.g., ChatGPT)
• Predictive AI — forecasts future events using historical data
• General AI — human-level performance across all tasks (does not yet exist)
• Superintelligent AI — surpasses human ability (theoretical)

ChatGPT

• Powered by a Generative Pre-Trained Transformer (GPT) model
• Trained on massive text datasets to learn grammar, facts, and reasoning
• Uses self-attention to understand context and word relationships
• Generates original responses by on predictive patterns — does not retrieve pre-written answers

How AI Works: Under the Hood

• Computers process everything as binary (0s and 1s) — voltage signals controlled by microscopic transistors
• Words and text are converted into tokens, then vectors that map relationships between concepts
• Neural networks assign weights to patterns, learning which words and ideas commonly occur together
• Models don't "understand" language — they predict the most likely next word based on training data
• Training involves two stages: pre-training (learning patterns from massive datasets) and fine-tuning (human-guided refinement for specific tasks)
• GPT-1 (2018): 117M parameters, trained on ~1B words; GPT-4o (2024): 1+ trillion parameters, trained on 40+ terabytes — and adds multimodal input/output (text, images, audio)

Core Ethical Threats in AI

• Transparency — AI is a "black box": users rarely know what data trained it or how it makes decisions
• Bias & Fairness — human biases baked into training data produce discriminatory outputs (e.g., biased hiring models)
• Privacy — prompts and user data are routinely harvested; companies like Meta openly use social media data for AI training
• Security — vast data stores make AI companies high-value targets for breaches
• Environmental Cost — training a single large language model can consume more energy than 100 homes use in a year

Hidden Human Cost of AI

• AI models depend on human-labeled training data, yet annotators are often invisible and exploited
• Work is outsourced via platforms like Amazon Mechanical Turk to low-income workers in India, Bangladesh, the Philippines, and similar countries
• Workers annotate thousands of images daily — often earning below minimum wage with no benefits or job security
• Psychological harm is widespread; annotators are frequently exposed to disturbing content with no employer support

History of AI and Libraries – Pre-2000

• 1936–1955: Turing's computing theory and the coining of "artificial intelligence" at Dartmouth lay the conceptual groundwork
• 1965: Licklider's Libraries of the Future first envisions computers as mediators between humans and knowledge
• 1970s–80s: "Expert systems" emerge as practical AI tools; librarians debate how to stay relevant amid rapid technological change
• 1992: The concept of "technostress" introduced — library professionals struggle to keep pace with accelerating innovation
• 1999: AI-enhanced information retrieval gains traction, but raises early concerns about loss of context and depth

History of AI and Libraries – The Road to ChatGPT

• 2015: First operational library AI chatbot ("Xiaotu") debuts at Tsinghua University in China
• 2019–2022: Academic librarians show enthusiasm for AI but report low awareness and limited adoption
• Late 2022: ChatGPT launches, transforming public discourse and pushing libraries to act — not just plan
• 2023+: Libraries begin building custom ChatGPT-powered reference chatbots; "prompt engineering librarian" emerges as a proposed new role

AI Applications in Library Systems

• Acquisitions — algorithms analyze circulation data to optimize budget allocation across print, digital, and audio formats
• Automated cataloging — computer vision captures metadata and assigns call numbers, verified by human catalogers
• Chatbots — handle directional and basic reference queries, freeing librarians for complex patron needs
• Smart search — conversational AI helps patrons articulate information needs and formulate better queries
• RAG (Retrieval-Augmented Generation) — combines real-time retrieval with generative AI to deliver synthesized, personalized research responses
• Student support — early detection systems identify at-risk students and connect them with library resources proactively

ICE-T: Implementing AI Responsibly

• Interoperability — does the AI tool integrate smoothly with existing library systems without creating new problems?
• Control — can the library manage, adjust, and oversee how the system operates and where data is stored?
• Explainability — can the system clearly communicate to patrons and staff how it arrives at its outputs?
• Teachability — is the system easy enough to teach patrons and staff to use effectively?
• Policy must also evolve: libraries need updated guidelines on Internet filtering, patron privacy, and opt-out rights for AI systems
• Key to adoption: involve skeptical stakeholders early, educate users, and establish clear organizational plans before deploying any AI tool

AI Literacy

• AI literacy = understanding how AI works, evaluating its outputs critically, using it effectively, and navigating its risks
• Distinct from data or algorithmic literacy — closer to information literacy, focused on source quality and reliability
• Key competencies: understanding training data, recognizing bias, ethical use, combating misinformation, and navigating AI policy
• Users must remember: AI models give a single "definitive" answer with no references — unlike search engines that return multiple sources to compare
• Hallucinations (fabricated facts) are a real risk; all AI outputs should be verified against reliable sources

Teaching AI Literacy – Libraries Leading the Way

• Libraries are natural leaders: already expert in information literacy instruction and public education
• Academic sessions should cover: model comparisons, prompt engineering basics, and academic integrity/policy
• Public library sessions should focus on: what AI is, hands-on interaction with tools, and distinguishing ChatGPT from AI broadly
• Prompt engineering is the new research skill — zero-shot, few-shot, role-playing, iterative, and conversational prompting all improve results
• Update AI literacy content quarterly — the pace of change demands it more than any prior literacy effort
• The ideal AI literacy librarian blends data literacy, information science, ethics awareness, and strong instructional skills

How May AI Transform Education?

• Adaptive learning — AI tailors lesson content, examples, and assignments to individual students' disciplines, cultures, and skill levels
• Adaptive testing — question difficulty adjusts in real time based on student responses, enabling more precise assessment of knowledge
• AI tutors — chatbots built on LLMs provide personalized, on-demand support to multiple students simultaneously
• Learning analytics — student data (grades, engagement, writing) can identify at-risk students early and guide them to support resources
• AI is unlikely to replace instructors — classroom management, motivation, and human connection remain irreplaceable

AI and AI (Artificial Intelligence and Academic Integrity)

• Every course needs its own AI use policy — a single school-wide policy cannot account for the variation across disciplines and assignment types
• Clear syllabus statements should define: what AI use is permitted, what is prohibited, and what penalties apply
• Students view writing an entire paper with AI as misconduct, but are more permissive about partial use — clarity in policy is essential
• Prompt engineering should be taught as a discipline-specific skill, just as subject librarians teach database research skills
• Micro-credentials (like UNT's AI Fundamentals) offer scalable, self-paced AI literacy education — over 1,000 students completed it in six months
• AI is a productivity tool, not a replacement for learning — the goal is responsible, informed integration, not outright bans

AI & Scholarly Publishing – Opportunities and Risks

• AI cannot be listed as an author — it lacks the free will to provide informed consent for publication
• Nearly all major publishers (98.9%) prohibit LLM authorship but permit AI-assisted writing if disclosed
• AI could democratize publishing by helping non-native English speakers overcome language barriers in peer review
• Key risks: hallucinated citations, AI-generated retractions, and the potential for humans to be removed entirely from the peer review process
• Safeguards needed: mandatory AI use disclosure, human-in-the-loop peer review, and AI systems that validate claims against external sources

AI Tools for Research and Popular Theoretical Frameworks for Studying AI

• Scite.ai — classifies citations as supporting, contrasting, or neutral, making citation counts more meaningful
• SciSpace — generates literature reviews using only real, indexed publications to prevent fake citations
• Elicit — extracts specific findings from papers into tables, ideal for systematic reviews and meta-analyses
• Research Rabbit — maps networks of related articles from a single seed paper
• Research on AI in libraries needs more social science perspectives — patron behavior, equity, and adoption patterns are understudied
• Three useful theoretical lenses: Diffusion of Innovations (societal adoption patterns), Technology Acceptance Model (individual psychology), and Actor-Network Theory (how people and tools shape each other)

Conclusion – Future of Libraries with AI

• AI is not magic, nor is it a threat — it is a human-created tool that reflects the quality of its data, design, and oversight
• Every major technology disruption in library history — information systems, the Internet, ebooks — ultimately expanded the librarian's role rather than eliminating it
• AI introduces real risks that demand attention: bias, hallucinations, privacy violations, exploitative labor, and environmental costs
• Libraries are uniquely positioned as trusted, public-facing institutions to lead AI education and ethical adoption
• The most important factor in any AI implementation is not the technology itself — it is whether people understand, trust, and can actually use it

Conclusion – Future of Libraries with AI

• The "intelligent library" is emerging — AI handling routine queries, cataloging, and analytics while librarians focus on instruction, advocacy, and complex patron needs
• New professional roles are taking shape: AI literacy librarians, prompt engineering specialists, and ethical AI implementation leads
• Libraries must act now: develop AI policies, evaluate tools using frameworks like ICE-T, and involve all stakeholders in the process
• Public AI education is a democratic imperative — an informed citizenry depends on institutions willing to cut through hype and misinformation
• The mission never changes: connect people with reliable information and foster an engaged, informed society — AI is simply the newest tool to fulfill it

On the Academic Book Publishing Process

Brady Lund – University of North Texas

Steps in the Publication Process

Proposal is accepted before writing begins!!

Do not write the whole book before receiving an acceptance!

Step 1: Identify a Publisher

Step 2: Establish Contact with an Editor

Step 4: Editorial Board and Peer Review

Step 5: Sign a Contract

Step 6: Write the Book!

Step 3: Put Together the Proposal

Step 7: Review and Editing

Step 8: Publication!

Identifying a Publisher

• University Presses – Generally looking for books with broader appeal rather than discipline-specific texts. Often specialize on books on particular topics (e.g., Texas History) and may favor authors affiliated with the institution.
• Academic Text Publishers – Publishers who publish a wide range of textbooks (e.g., Pearson), may also be major publishers of academic journals (e.g., Springer, Wiley).
• Academic Imprints of Publishing Houses – Often small presses associated with specific disciplines that have been acquired by large publishing houses (e.g., ABC-Clio and Rowman & Littlefield have both been acquired by Bloomsbury).
• Main Imprints of Publishing Houses – Books that make the shelves of your local bookstore. Generally require topics with very broad interest, connections/celebrity, or catching the attention of a literary agent. Much more difficult path but could be rewarding for a truly innovative idea.
• High Volume Publishers – publishers that publish a very large volume of books – may charge authors for publication or may be free but charge extravagant prices to purchase, ensuring low readership (e.g., IGI Global)

Identifying a Publisher

• I can speak most to the academic imprints, as that is where my books have all been published. These are a nice option for having a mix of scholarly rigor (all book proposals and final manuscripts are reviewed by peer reviewers) and visibility (associated with a major multi-national publisher with a presence on Amazon, Barnes & Noble, etc.). Many of these publishers also have a relationship with large professional associations (e.g., the American Library Association) that helps ensure them/you a minimum number of sales for every published monograph.
• Academic Imprints of Publishing Houses – Often small presses associated with specific disciplines that have been acquired by large publishing houses (e.g., ABC-Clio and Rowman & Littlefield have both been acquired by Bloomsbury).

Establishing Contact with an Editor

There are a few paths to starting this process:

• If your research is well-known or presented in a visible place (e.g., keynote or major presentation at a national conference), it is possible that an editor will reach out to you and encourage you to submit a proposal. This happened with the first book I wrote (Casting Light on the Dark Web).
• More commonly, you will want to submit an unsolicited proposal. You can find the instructions for submitting a proposal somewhere on the publisher's website (here is Bloomsbury’s, for instance: https://www.bloomsbury.com/us/discover/bloomsbury-academic/authors/submitting-a-book-proposal/). You complete a proposal form and then send to the appropriate editor.
• Note the process is different for non-fiction versus fiction books (e.g., fiction imprint of Bloomsbury would generally require a literary agent in order to have any realistic chance of acceptance).

Once you have formed a relationship with an editor, you can just reach out to them whenever you are interested in writing and discuss whether they would be interested in your proposal!

Putting Together the Proposal

• Even with academic books, editors are still looking for topics that will sell. If your book doesn’t have a reasonable chance of selling at least 1000 copies, they probably won’t be interested in publishing it.
• You also need an original idea, not something that already exists in the market. You need to show that there are similar books, but that your book adds something unique.

Editorial Board and Peer Review

Academic publishers have several rounds of review and approval after a proposal is complete:

• Review by the subject/acquisitions editor (the person who would be your contact with the publisher)
• Review by full editorial board (usually all editors with a certain department of the publisher meet monthly to review proposals that pass through initial review)
• Review by at least two anonymous peer reviewers
• Final review and recommendation by the subject/acquisitions editor
• Approval from lead acquisitions editor.

The proposal could be rejected at any of these stages. Revisions may also be needed.

The Financial Reality

• Publishing academic books are not a great money-making venture.
• Generally, publishing contracts pay 6-10% of net proceeds to the author. Net proceeds are calculated as the sale price of the book minus all costs for publishing and marketing the book. On a $60 book, the net proceeds may be $35 (after print costs, taxes, marketplace fees, etc.). This nets the author about $2 - $4.
• Even if the book sells 1000 copies, the average income will be around $3000, before income taxes.
• If the average academic book is 50,000 words, and it takes 1 hour per 250 words to research, write, and edit, then that is 50000/250 = 200 hours
• $3000 / 200 = ~$15 per hour
• Again, that is assuming you work fairly fast and the book is relatively successful.

The Academic Reality

• The value of a book towards tenure, promotion, and annual review varies across universities, departments, and disciplines. Some departments do not place any value in publishing a book. Some will allow it to substitute for 2-3 peer reviewed journal articles.
• In most situations, the value proposition will not be in favor of writing a book. You could publish several complete research studies with a similar amount of effort.
• Who you publish with can be an important factor in how the book is perceived by colleagues.
• Junior faculty might be discouraged from writing a book for these reasons.

So Why Publish an Academic Book?

Book publishing is itself a rewarding experience:

• See your book on major booksellers websites.
• Share your knowledge.
• (Generally) More freedom over your writing.
• Even with low pay, still pays more than academic journals!