You probably wouldn’t pull Lithium-Ion Batteries: A Machine-Generated Summary of Current Research off the shelf anytime soon. But the research book is more interesting than it sounds: Its author, “Beta Writer,” is a machine-learning algorithm designed by researchers from Goethe University in Frankfurt, Germany.
Springer Nature recently published the textbook, the first to be written entirely by an algorithm.
The concept of AI authors has been circulating for some time now. Machines have been recruited to help write sports recaps, financial reports, road trip novels and even “Game of Thrones” installments, to varying degrees of success. But Beta Writer’s debut marks “the first machine-generated research book,” according to the academic publishing company. The book consists of about 250 pages of compiled research, sorted into chapters based on subject matter. The algorithm compiled section introductions, quoted passages with hyperlinks to original texts, and created a table of contents and references—all without human intervention.
“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, the head of the lab that designed the algorithm, said in a press release.
In this case, the writing process didn’t demand a lot of creativity or literary prowess. Instead, it was more of a brute-force job, necessitating the scanning, sorting and summarizing of thousands of pages of research on lithium-ion batteries, the rechargeable power sources for smartphones, laptops, electric cars and more, which it pulled from papers in Springer Nature’s online database.
The dry source material might have been a good thing because Beta Writer doesn’t have a way with words quite yet. “We have succeeded in developing a first prototype which also shows that there is still a long way to go,” Springer Nature’s Henning Schoenenberger acknowledges in the book’s introduction, the only portion of the text authored by humans. The publishing house intentionally did not copy edit or "polish" any of any of Beta Writer's texts like it would have for a human author, Schoenenberger explains, “due to the fact that we want to highlight the current status and remaining boundaries of machine-generated content.”
Jeff Bingham of Carnegie Mellon’s Human-Computer Interaction Institute sees plenty of room for improvement with the technology: "It is quite straightforward to take high-quality input text, spew out extractive summaries pushed up next to one another, and have it look somewhat coherent at a cursory glance," Bingham told The Register’s Thomas Claburn in an email interview. "In fact, the very nature of extractive summary means it will be coherent in chunks, so long as the input texts are coherent. It's much harder to create something that a human reader finds valuable."
That being said, Beta Writer successfully turned a “firehose of data” into a “manageable trickle,” Gizmodo’s Andrew Liszewski points out, giving scientists a more approachable way to dive into the intimidating wealth of lithium-ion research out there. The success of the algorithm’s work, although limited, could show promise for cutting down on the amount of tedious work required to stay current in an age of information overload.
Springer Nature hopes to adapt its algorithm to produce similar books on different subjects, including the humanities and social sciences. That doesn’t mean humans will disappear from the process; Schoenenberger foresees a future in which scientific authors collaborate with algorithms like Beta Writer to make the publishing process more efficient.
Data scientist Ross Goodwin, for one, welcomes that future. “When we teach computers to write, the computers don’t replace us any more than pianos replace pianists,” he's quoted as saying in the book’s introduction. “In a certain way," he adds, "they become our pens, and we become more than writers. We become writers of writers.”