AI Textbook

In what is believed to be the first example of using artificial intelligence (AI) to write a book, academic publishing company Springer Nature recently released Lithium Ion Batteries: A Machine-Generated Summary of Current Research. While it’s not likely to make the bestseller list, it does represent a new era in scientific publishing.

With vast numbers of studies generating literally thousands of papers per year on any given topic, using AI to scan and summarize the information could save scientists valuable time by enabling them to keep abreast of new developments while continuing to do what they do best – research. If they feel a need to explore a subject further, links within the book provide easy access to the original source material.

This type of “robot writing” is probably not suitable for all types of books, but for formulaic texts and extractive summaries (for example, financial news or sports summaries), machine learning would definitely reduce the drudgery of poring over thousands of pages of scientific data. The method still needs some fine-tuning before it can be relied on to generate consistently coherent prose, and it may never be able to mimic the structure that human writers can generate, but this important milestone will provide publishers with much-needed information regarding the challenges and potential risks of AI publishing, and it will undoubtedly get better over time (as machine learning always does).