Thursday, April 19, 2018

Today’s headlines boldly proclaim both the beneficial and threatening aspects of artificial intelligence (AI). Leaving aside the hype and the fear-mongering, there is an important aspect of AI that is gaining more and more attention. The issue: How can AI tools be made more transparent? Governments, organizations, and corporate entities are all calling for greater transparency, accountability, and a right to an explanation when AI techniques are used in decision making that can determine if you get a mortgage, go to college, get a job, or get out of jail. Our first speaker explores the issue of algorithmic accountability and transparency, discusses whether governments should regulate AI, corporations set their own standards, and more. Then hear about the tools one library used to assess their virtual reference service with text analysis research by using 6 semesters of 10,000 chat transcripts. They used Voyant and Lexos software to extract words and phrases from the chat transcripts and establish word counts and frequencies, then compared the vocabularies of librarians vs. students in chat reference interviews to improve communication between librarians and their user base; findings are being applied to reference tools and resources. They used the Topic Modeling Tool, adapted from the original Mallet tool, to trace related clusters of words and perform a content analysis on the chat FAQs. Finally, a sentiment analysis using The Subjectivity Lexicon compared student and librarian sentiment. Procedures for all the text analysis techniques are presented, along with key findings and applications.