The Special Libraries Association (SLA) offers a wide range of continuing education activities. In addition to the annual meetings, webinars, and in-person training sessions, the local programming often includes gems of professional knowledge that are invaluable. Recently the San Francisco Bay Region Chapter of SLA presented an evening program “Artificial Intelligence Tools for Information Discovery,” offering unique insights into the incorporation of artificial intelligence (AI) applications into library research tools.

The program covered libraries’ expanded approaches to information discovery, and how AI technologies are integrating into the tools available to users for their research. The discussion specifically looked at Stanford University’s recent experience with integrating Yewno (yewno.com), an innovative AI research tool, into its overall discovery interface. Speakers included Mimi Calter, deputy university librarian, Stanford University Libraries, and Ryan Mayfield, strategy and product manager for global affairs, Yewno. As a bonus, the session took place in the historic Bing Wing of the Cecil H. Green Library.

Background of AI at Stanford

Calter began by discussing how several scholars at Stanford have done pioneering work in AI, pointing out the paper by Edward A. Feigenbaum, “Toward the Library of the Future.” This seminal article appeared in Long Range Planning (Vol 22, No. 1, February 1989, pp. 118–123), which was based on a talk Feigenbaum gave at Aston University in November 1986. Feigenbaum and others in the Stanford AI Laboratory have had an influence on how the Stanford University Libraries went about the use of AI in their services. The vision Feigenbaum put forth was to “develop an information infrastructure to serve teaching and research in universities.”

Calter pointed out that interest in AI is pervasive on the campus, and Nicole Coleman, a member of the library staff, is actively investigating the ways that the libraries can both partner with those developing AI tools, and leverage AI to better support library users. She sees the library/AI conversation as having discovery as the common thread, a natural relationship to build on.

Coleman, who is the digital research architect at Stanford University Libraries and co-director of the Humanities + Design research lab, recently updated the Feigenbaum paper in an insightful blog post, “Artificial Intelligence and the Library of the Future, Revisited” (library.stanford.edu/blogs/digital-library-blog/2017/11/artificial-intelligence-and-library-future-revisited). In this piece, Coleman provides an overview of historic and current AI technologies and projects and asks, “How best do we bring the skills and knowledge of library staff, scholars, and students together to design an intelligent information system that respects the sources, engages critical inquiry, fosters imagination, and supports human learning and knowledge creation?”

Yewno: A New Strategic Partner

Mayfield described the essence of Yewno as having “taught an algorithm how to read.” He went on to provide more overview points such as, “No one has or is doing general artificial intelligence,” which he defines as “machines thinking broadly.” Working AI is much more specific and narrow. Historic examples include autopilots for commercial aircraft, spam filters, and online shopping. These “narrow AI” examples are built on computers doing very specific tasks and responding to associated stimuli.

Mayfield sees development of AI as being “segmented” into discrete pieces. Yewno focuses on extracting knowledge from textual information, which includes computational linguistics, Big Data, identification of concepts/topics, entity extraction (people, places, things) and the machine learning aspects of AI. Machine learning is about computers learning from very specific datasets, either from their labels, characteristics, or feedback from humans. Mayfield feels a better term would be “augmented intelligence,” and believes the best organizational applications of AI come from solving problems and advancing learning.

Looking at AI as a problem-solving technology is central to what Mayfield called the “AI Flywheel.”

The AI Flywheel shows an iterative process where problems are related to data; lessons learned are examined; and the cycle starts again.

As to why AI is now such a pervasive technology, Mayfield includes the ease of information exchange and the rise of on-demand computing power, particularly in the cloud. In turn, this drives economic forecasts and shifts in the technological landscape. In 2016 Accenture wrote, “AI could double annual economic growth rates by 2035” (accenture.com/us-en/insight-artificial-intelligence-future-growth). This leads to companies involved with numerous technologies and from various industries jumping on the AI bandwagon. However, this has also led to inflated expectations and unrealistic understandings of what AI could do.

Yewno has its roots in the education and publishing industries. It has expanded to working with finance, biomedical, and government organizations. Mayfield is involved in using Yewno for analysis and problem-solving in global affairs. He said there is an extensive amount of textual data dealing with government and global affairs that makes this corpus a prime candidate for the advanced semantic analysis Yewno provides. This work can reveal insights and priorities into financial and regulatory issues as well as demonstrate relationships across silos of disparate data. Global affairs identifies entities such as agencies and organizations, works well with all levels of government, and points to business and industry opportunities.