AI in library from learning to application
by Gang Shao
With the increasing computational capacity, AI and machine learning, particularly deep learning, are becoming leading technics and have more and more impacts in our daily life, from face recognition phone lock to security camera to the whole smart home systems, from learning information with big data to generating new text, from deep dreaming images to deep fake video generation. And it will continue to act as front edge technology for the foreseeable future.
Applying AI in libraries, archives, and museums, sounds like something emerging but already has a long history due to the high demand for digital library collection, data and other resources. Based on Machine learning + Libraries: A Report on the State of the Field, libraries apply AI and machine learning in many different areas, including but not limited to text and image clustering and classification, Optical Character Recognition (OCR), handwriting recognition, metadata extraction, data annotation. As the new generation of academic librarians for library service 2.0. It is important to understand how AI and machine learning work and how these technics can be applied and benefit our library services.
Learning AI and machine learning was not an easy task several years ago as the educational resources on AI and machine learning were very limited and the available resources were mainly targeting audience with stats and computer science background with a high bar on math and programming skills. Currently, it is a different story as more and more AI for All educational initiatives and open education resources. There are even more YouTube videos on such topics to educate AI and machine learning to the public without experience. Research on artificial neural networks was motivated by the observation that human intelligence emerges from highly parallel networks of relatively simple, non-linear neurons that learn by adjusting the strengths of their connections. With such concepts, it is not very hard to understand the foundations of AI and the algorithms behind, such as how to build a neural network, how to calculate weights in each inceptron and how to build a deep learning model for object detection.
However, knowing such fundamental knowledge may not promise applications to provide efficient and robust services or even be challenging to meet the service quality with no AI components. Take making a library chatbot as an example, with so many existing chatbot platforms (like Dialogflow by Google, and Azue Bot service by Microsoft), it is not hard to build a chatbot. However, such platforms usually provide limited access to the AI model behind the platform. It is very challenging to improve the chatting service without the ability of model modification. To build the chatbot to meet the requirement of quality library services, we still need to hunt solutions for more adjustable platforms. Also, applying AI for quality service requires a deep understanding of the AI model, architecture, and technic updates. Protecting AI models from adversarial attacks is another important knowledge that needs to draw the attention of librarian who are working on AI and machine learning projects.