Values-driven AI in libraries and archives: Introducing the Viewfinder toolkit

1 Sara Mannheimer, Scott Young, Yasmeen Shorish, Hannah Scates Kettler DLF 2025 A toolkit for values-driven AI in libraries & archives 2 Today’s session Project motivation and goals Viewfinder toolkit development Viewfinder activity Discussion Project Team 3 Hannah Scates Kettler Iowa State University Sara Mannheimer (PI) Montana State University Jason A. Clark Montana State University Bonnie Sheehey Montana State University Yasmeen Shorish James Madison University Doralyn Rossmann Montana State University Scott W. H. Young Montana State University Natalie Bond University of Montana Project Advisory Board 4 Dorothy Berry Smithsonian Nat’l Museum of African American History & Culture Stephanie Russo Carroll University of Arizona Bohyun Kim University of Michigan María A. Matienzo University of California, Berkeley Thomas Padilla, Bristlecone Strategy 5 Project funding IMLS National Leadership Grant LG-252307 https://www.imls.gov/grants/awarded/lg-252307-ols-22 Project motivation ●How can we use AI to improve services while also upholding our values and protecting our communities from harm? 6 Goals of the project ●Develop practical resources that support responsible use of AI in library and archives contexts. ●Consider values and ethical implications as we embark on new AI projects 7 8 “Through the slow and careful adoption of tech, the library can be a leader.” — Kate Zwaard Cordell, R. (2020). Machine Learning + Libraries: A report on the state of the field. LC Labs. Viewfinder development Phase 1: Literature Review Phase 2: Case Studies Phase 3: Practitioner, Administrator, and Student Workshops Phase 4: Data Analysis and Tool Development Phase 5: Final Tool 9 Phase 2 of 5 Case Studies 16 17 Case Studies ●Selected 8 author groups from academic libraries, research institutes, health science libraries ●Projects include: ○AI chatbots for reference services ○Computer vision to generate metadata ○AI-enhanced search ○Vendor/library machine learning partnerships ○AI/machine learning for automated text extraction 18 Case Studies - Ethical Considerations ●Describe any ethical issues that arose as you implemented the project, potentially including: ○How you considered and addressed potential harms associated with the implementation. ○Did you engage with stakeholders? ○Did you refer to existing documentation, policy, or best practices? 19 Responsible AI Case Studies Volume 13, issue 1 of Journal of eScience Librarianship, Spring 2024. https://publishing.escholarship.umassm ed.edu/jeslib/issue/59/info/ Phase 3 of 5 Practitioner & User Workshops 20 21 Practitioner & User Workshops ●16 workshops with 62 participants (library/archives practitioners, library/archives administrators, and users) ●Workshop goals: to produce original research data that can inform the development of a practical ethics tool ●Values circles: Which values are relevant for AI in libraries? ●“Dark side” exercise from the Design Method Toolkit: What does irresponsible AI look like? Phase 4 of 5 Data Analysis, Tool Development, Assessment, and Validation 22 Analyze workshop data ●Qualitative content analysis of workshop data ●Review of existing similar tools 23 Phase 5 of 5 Final Responsible AI Tool - Viewfinder 24 25 A toolkit for values-driven AI in libraries & archives Step 3: Choose Stakeholder (2 min) Now choose a stakeholder card at random that represents a different perspective from your own. Step 4: Stakeholder Values (5 min) Consider which values are of concern to that stakeholder in the given scenario. Identify 3 values that are of most concern. Step 5: Reflection Prompts (15 min) Reflect on and respond to the Reflection Prompts. Record notes and key insights in the worksheet. There are three different sections of the toolkit: Scenarios, Stakeholders, Values. Step 1: Choose Scenario (2 min) Select one scenario card OR draft your own AI implementation project or scenario. Step 2: Your Values (5 min) Consider which values are of concern to you in your scenario. Identify 3 values that are of most concern. Step 3: Choose Stakeholder (2 min) Now choose a stakeholder card at random that represents a different perspective from your own. Step 4: Stakeholder Values (5 min) Consider which values are of concern to that stakeholder in the given scenario. Identify 3 values that are of most concern. Step 5: Reflection Prompts (15 min) Reflect on and respond to the Reflection Prompts. Record notes and key insights in the worksheet. 36 1. What is one insight from the activity that you want to share with the full group? 2. What else can our team provide to help you implement the Viewfinder tool in your local context? 37 Thank you! lib.montana.edu/responsible-ai/ bit.ly/aiviewfinder https://osf.io/yue9s 38