ai4Libraries virtual-conference

quick notes

5 min, 873 words

conference info


Morning keynote by Dave Hansen

  • terrific
  • Some issues around mining-for-AI aren't new, legally; text-data-mining has been around for a long time.
  • lots of lawsuits recently filed (against Stability, OpenAI, Meta) -- should have more clarity over the next year or so
  • Is it permissible to use copyrighted works for use as training data? -- for non-commercial academic-research, probably, based on text-mining settled law.
  • Open questions:
    • what about when copying is't just for search/analysis, but for creation of new works?
    • how should we think about systems that are used by users to create new works that may in some way substitute for the original?
    • "snoopy problem" -- generative AI can be used to create known-copyrighted-characters.
      • do the products/tools have safeguards against creating copyrighted outputs? If not, does that put the tools and tool-creators at legal risk?
      • to what extent are the products/tools (vs user) responsible for resultant infringing works?
  • Terms of service can have implications.

AI, Libraries, and Higher Education (Panel

  • AI tools will play a role in retrieval of info. Either because users want to use them, or because regular tools we all use will begin invisibly using AI.
  • Libraries have a historic curation role -- offering not only the info but info but also attributions of where info is from -- unlike AI.
  • Some AI tools are super useful for academic-research projects -- eg fingerprinting project.
  • Fit technology to goals -- good calculator analogy: if the purpose is "learning arithmetic", don't use calculator much. But if it's learning higher-level calculus, a calculator is essential.
  • Our ways of knowing / demonstrating-knowledge -- have been through writing. Now that writing is easy and cheap and not necessarily connected to a human author, writing may be less valuable.
  • One aspect of protecting ourselves from copyright infringement is to clarify that AI-use is educational/for-learning -- not commercial.
  • Assertion for good discussion: "AI won't replace humans, but humans with AI will replace humans without AI."

AI and Ethics: Considerations for the Scholarly Communities

(points below are a mix of points made and my thoughts)

  • Should it be clear that presented data may have been (partially) AI generated?
  • Should our resources be permitted to be harvested for AI?
  • How to keep AI tools from harvesting user-info potentially inappropriately?

(points below are a mix of points made and my thoughts)

  • Ithaka S+R is working on an AI-Resiliency project, with 19 universities
    • because I never remember: the "S+R" stands for "Strategy & Research"
  • Areas AI will be affecting: search, collections-as-data, process-automation, analytics, smart-sensors.
  • Sandboxing one technique to minimize PII leakage. One example mentioned in chat:

Tech-Services panel

(points below are a mix of points made and my thoughts)

  • Examples of how back-end processes are and will be affected by AI: transcriptions, OCR, using AI to write scripts, auto keyword-generation, enhancing OpenRefine-ish capabilities; auto-cataloging to MARC-format.
  • Defining licensing-protections should be a goal; challenging.
  • Challenge of communicating Library desires to vendors. They may be receptive, but if the main point-of-contact is salespeople, not constructive.
  • Conveying that AI is basically "statistical guessing" may reduce staff fear.
  • Would be nice to investigate ways staff knowledge could be used to train tools.

Lightning talks

  • Leveraging AI tools for automating metadata extraction -- UCLA folk used OCR, fed data into spaCy for "named entity recognition", tried using chatgpt for DC records (good for title, description, language; bad for date, type, physical description, rights)
  • Moles and Martians: A New, LLM-based Library Search, by Tim Spalding, LibraryThing -- gave examples of human exploratory searches that were handled well. See and the youtube video there.

interesting projects/articles mentioned