Want to learn about lithium-ion batteries? An AI has written a tedious book on the subject
Automated assembly of articles on lithium-ion battery research tests technology and patience
Writers looking to make their names typing impenetrable technical tracts be warned: the machines have arrived and they're already penning scholarly books few will ever read.
On Monday, Springer Nature published what it claims is the first machine-generated book from an academic publisher, titled "Lithium-Ion Batteries, A Machine-Generated Summary of Current Research."
That claim to be first, qualified though it may be, is difficult to verify. Ross Goodwin's "1 the Road" (2018), is sold as "the first real book written by an AI," although it probably isn't. Journalism written by code has been a thing for several years, mainly for rote reporting like earnings reports, earthquake alerts and sports scores.
The book for battery boffins, available as a free download, provides an overview of lithium-ion battery research, summarizing more than 150 research papers published between 2016 and 2018. It was assembled through a process overseen by Christian Chiarcos, assistant professor of the Applied Computational Linguistics (ACoLi) lab at Goethe University.
"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," said Chiarcos 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."
As literature or even plain prose, it's not quite a page turner. More than a few passages sound like they were written by algorithm. Consider "Numerous researches on CuO/graphene composites utilized as Li-ion batteries anode have been indicated" or "Improved electrochemical properties than the pristine CuO nanorods were shown by the composite electrode."
The book's flaws, however, are acknowledged in the preface, penned by Henning Schoenenberger, director of product data and metadata management at Springer Nature, which explains the process for creating this battery research compendium.
Schoenenberger suggests the project is focused on building a functional production workflow that integrates natural language processing and related technologies; it's not attempting to advance the state of the art for natural language processing, which hasn't quite kept up with rapidly advancing fields like AI image recognition.
Cute, but is it up to snuff?
In an email to The Register, Jeff Bigham, associate professor at Carnegie Mellon's Human-Computer Interaction Institute, found the project underwhelming.
"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," he said. "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."
As a one-stop shop for lithium-ion battery research, "Lithium-ion batteries" is possibly functional, not so much for reading as for finding reference links to academic papers. But even for that, Bigham suggests concentrating a list of Google Scholar search results might work just as well.
Schoenenberger says that there have been more than 53,000 academic articles published in lithium-ion batteries over the past three years. He characterizes future battery tech advancements as an existential issue for humanity, one that necessitates the application of automation.
"The future of mankind depends on progress in research on lithium-ion batteries, and we need to think of innovative ways to enable researchers to achieve this progress," he wrote in the preface. "This is where the potential of natural language processing and artificial intelligence (AI) comes in that might help researchers stay on top of the vast and growing amount of literature."
Schoenenberger acknowledges that automated publishing raises some issues that haven't yet been addressed, such as who is accountable for machine-generated content from an ethical standpoint? Part of the goal for this project is to start grappling with these questions because there's more machine-generated publishing on its way.
According to Schoenenberger, Springer Nature plans to look beyond chemistry for future machine learning prototypes to fields like the social sciences and humanities. Something to look forward to. ®