A publication about lithium-ion batteries is the first textbook authored by a computer. The 247-page title sets a new milestone in machine-generated publishing. Springer produced the book by collaborating with researchers from the Applied Computational Linguistics (ACoLi) lab at Goethe University in Frankfurt.
- Everything in the book beyond the preface was algorithmically generated , says Springer. They suggest their method opens ‘a new era in scientific publishing.’
- Springer intends to automate publishing more books in other fields, including the humanities and social sciences.
- The achievement: The system assessed 1,200 possible papers in Springer’s files and shortlisted them to 150. It organized them by theme, then extracted and summarised, aggregated or paraphrased the content as required. The system also arranged elements such as bibliographies, scientific notation, introductions, and a table of contents.
- The value: The volume of published findings is growing rapidly and researchers want to keep up or get up-to-speed quickly. More than 53,000 research papers in the field have been published in the past three years. Automated systems will help match supply (of information) to demand.
- This changes the scale for automated text. News autoscribe systems are averaging 100-400 words, produced with templates. The Springer system produced more than 200 pages drawn from complete papers.
- The workflow creates a platform for more titles. Other kinds of long-form NLP publications are inevitable.
- It’s another industry + academic success story. Industry identifies a market need and the data, the research community provides know-how and specialty computing power. Both gain a high-profile outcome, practical learning, and possibilities for revenue or a patrh to future spin-offs.
SEE RELATED STORIES
The first AI-generated textbook shows what robot writers are actually good at
THE VERGE | April 10,
The First Machine-Generated Book by a Scholarly Publisher Is a Boring Read
GIZMODO | April 9,