LIR HEAnet User Group for Libraries provides a forum for discussion on the use of electronic resources and networks by HEI libraries and assists in development and training for their effective use. WIT Librarian David Kane chairs the LIR Committee. This year’s seminar was held in the CIT School of Music, Union Quay, Cork and had the theme “Artificial Intelligence / Automation: Short term and Long term thinking in libraries”. I was invited to give a seminar on the topic of AI and research.
There was a strong and varied programme of talks. Highlights included Dr Andrew Cox (University of Sheffield) reviewed how libraries have been transformed by new models of publishing and information retrieval. Adrian Vaughan & Michael Costello (CIT) used IBM Watson to power an information service, presented as a bot, to help answer queries for CIT library users. Michael Upshall described and demonstrated the unsilo service that uses AI and concept mapping to make it easier for researchers and innovators to find analogous concepts and novel ideas from different industries and fields. Dr Tony Russell-Rose (University of London) described how Natural Language Processing (NLP) enabled many moderna applications and described some of the challenges and recent developments, notably the development of BERT. Brenda O’Neill from WIT’s Insyte-Cooley Centre described how undergraduate students are assisting with documenting, tagging and analysing artefacts using AI tools.
AI and Machine Learning – their impact on research and knowledge
The talk introduced AI – what it is, how it relates to machine learning, big data and data science, etc. The ethics of AI were considered briefly. The use of AI for knowledge generation was considered, see Figure below.
Knowledge Generation for Research
One of the critical requirements for any research result is that it should be validated and explained. Current focus is on prediction accuracy, but the lack of contextual awareness can mean that prediction results are brittle in the sens that small changes to the inouts can result inb very large changes in the outputs. There is a growing concern that deep learning, in its current form, is not sufficiently reliable to be used in safety-critical systems like driverless cars. Research procedures might have something to offer here.
The attendees were very interested in what AI and the automation it enables had to offer in libraries. They also had concerns about how to keep up to date with the latest techniques for information retrieval, and to recommend them effectively to library users. For TSSG, it is clear that our research in AI and Machine Learning could have an impact on research itself.
Written by Bernard Butler