First Machine-Learning-Generated Book Proves AI Lacks Creativity

First Machine-Learning-Generated Book
Pexels / Pixabay

Artificial intelligence and machine learning reached their peak performance in the last couple of years. So far there are algorithms capable of making pizza, exploring exoplanets, beating professional gamers at their favorite games and even see through walls. Some might think computer-generated algorithms are starting to outperform humans. However, the first machine-learning-generated book shows humans do still have the edge over computers in one area.

Springer Nature, a book publisher which focuses on the research community, published its first machine-learning-generated book, and it’s far from interesting. While the book can’t offer the creative quality of Stephen King, J.K. Rowling, Ernest Hemmingway, or any other literary masters, its sole purpose was to see how advanced and creative machine-learning algorithms are compared to humans.

The algorithm was created by researchers at the Applied Computational Linguistics (ACoLi) lab, which is part of the Goethe University in Frankfurt, Germany. The book titled Lithium-Ion Batteries: A Machine-Generated Summary of Current Research doesn’t feature any creative characters, wondrous battles or empowering dialogues. Instead, it describes extensive research performed on lithium-ion batteries. The book is available for free here.

Given that a lot of devices, including smartphones, watches, and cars, are powered by lithium-ion batteries, it’s no wonder there have been more than 53,000 papers and articles on the subject over the last three years.

The algorithm analyzed thousands of titles about the topic to ensure they contained relevant content for the first machine-learning-generated book. After that, it parsed, condensed and organized the publications from Springer Nature’s online database to create corresponding chapters focusing on different approaches to lithium-ion batteries. The book includes summaries automatically generated for articles in the chapters and quoted sentences linked to the original research papers for those who are more interested in a particular approach.

The first machine-learning-generated book could be considered as a sort of Reader’s Digest for researchers, with 180 pages which are easy to browse through, compared to the hundreds of document pages in the research papers it drew from. This may mean that it could introduce a new way to write research papers. The source material was still written by human writers to provide at least some quality control. Clearly, the writing industry will not be taken over by AI anytime soon.

“This publication has allowed us to demonstrate the degree to which the challenges of machine-generated publications can be solved when experts from scientific publishers collaborate with computer linguists,” Christian Chiarcos of Goethe University’s Applied Computational Linguistics (ACoLi) lab explained in a statement. “The project also enabled us to better understand the expectations of authors, editors, publishers and consumers – with regard to both scientific and economic requirements.”

The research-like book doesn’t mean that AI completely lacks creativity. In February, the formerly Elon Musk-backed organization OpenAI published a controversial language model which can transform a small sample of text into a set of convincing paragraphs similar to what can usually be found in tabloid-like news outlets. At the time, it caused a lot of debate about whether such models can be used to manipulate the media and spread false information.