Library AI Initiative
The vision of a library driven by artificial intelligence (AI) which Ed Feigenbaum shared thirty years ago may be upon us sooner than he predicted. The machine-augmented human intelligence work, or ‘narrow’ AI that Feigenbaum envisioned for the library is a practical application of search and pattern matching. This is distinguished from the more philosophically engaging and, so far unattainable, ‘general’ AI that some futurists anticipate will surpass human intelligence. The potential applications of narrow AI within the Library are clear: machines can discover patterns and make classifications in images, text, and audio, much faster and more efficiently than humans. We are creating and collecting information today at a pace that is not sustainable without power tools to do the mundane work so that Library staff can attend to the creative and intellectually engaging work of making information useful for research and teaching.
The Library AI initiative at Stanford is a program to identify and enact applications of artificial intelligence —machine perception, machine learning, machine reasoning, and language recognition— that will help us make our rich collections of maps, photographs, manuscripts, data sets and other assets more easily discoverable, accessible, and analyzable for scholars. We have already begun to increase awareness and familiarity with relevant concepts and new research in AI and encouraged SUL-wide conversations about the possible applications of AI. This effort will continue with the "Library AI Conversations" series, blog posts, an email list, Slack channel, brown bag lunches, etc. The objectives in coming year, are: 1. Provide training for staff; 2. Build partnerships across Stanford, with other academic libraries, and with industry; and 3. Establish a lab to model design and development practices for Library AI.
The objectives of training, partnerships, and an experimental environment for development apply to staff across the Library, in all areas of expertise, not only to staff in technical positions. As our first Library AI Conversations guest, Peter Norvig, explained, machine learning is not linear and not at all like developing discrete software applications. Systems cannot be developed in isolation, they need to be in a constant dialogue with experts who are giving feedback and adjusting the machine to the domain-specific task. Also, optimization models are not generically applicable. The one-size-fits-all approach to application development does not work with AI. As Andrew Ing has pointed out, AI requires new roles for an entirely different design process and we do not yet know what they are.
Though there is no consensus about the definition of AI, applications of AI are already transforming knowledge management and knowledge production. Andrew Ing describes AI as the new electricity; not just a new technology, but a technology that will fundamentally change existing systems. In anticipation of that not so distant future, the Library’s AI program must be, from the very beginning, purpose-driven rather than technology-driven, to protect us from moving rapidly down a path of a solution seeking a problem. It is not a digital library project, but a library project that brings everyone into the conversation, the planning, and the implementation.