The Writing is on the Wall for Artificial Intelligence.
Publisher Springer Nature has released a scientific text generated by artificial intelligence. "We are thrilled to finally publish this new type of research content and make it available for the global research community," says Henning Schoenenberger, director of product data and metadata management, Springer Nature.
The book's publisher is the first to caution that the text, which offers an Al-generated summary of the insights and data in 150+ scientific papers on lithium battery research, should be seen as a prototype. (The text, "Lithium-Ion Batteries: A Machine-Generated Summary of Current Research," is available for free download at https://link. springer.com/book/10.1007/978-3030-16800-1.) Still, it has triggered serious expectation that the emergence of high-quality machine-generated text is only a matter of time.
AI has made inroads into writing in a variety of contexts. An entire news agency built around AI-generated writing, RADAR, has emerged in the UK. The agency employs five journalists, who create story templates that its AI tool uses to pump out 8,000 to 10,000 localized news stories each month for various news publishers. RADAR has clients as large as the UK's JPIMedia Ltd. and as hyperlocal as Wales's Caerphilly Observer. Traditional news providers are paying attention. "The use of artificial intelligence in preparing news reports is still at the earliest stage of its development," says Yong Huang, director general of China's Xinhua Europe Regional Bureau. "But we are convinced that it will seriously influence the development of journalism and will help news agency employees to do their work faster and better."
Other applications apply AI to the writing of, for example, product descriptions and business analytic reports. Ginnie.ai mines the product data retailers already have in their databases to create descriptions that would normally be generated by copywriters, saving time and money. According to Isaac Wanzama, developer of Ginnie.a, the software provides "customized, enriched--and always unique--content for all of the products in a seller's catalogue."
Arria NLG, a maker of tools for Al-generated writing, recently announced that its software can now generate text narratives from popular business intelligence dashboards like Microsoft Power BI, Tableau, MicroStrategy, and Qlik. Essentially, the application enables users to create text-based analytic reports on marketing, sales, product development, and other business processes with the click of a mouse, using the data stored by the business software programs. Arria's platform enables users "to develop narratives on massive amounts of data," says Adam Heitzman, managing partner of HigherVisibility, an Arria NLG user.
For the tech underlying Springer Nature's spin on AI-generated academic writing, Springer turned to Christian Chiarcos, an assistant professor and researcher in applied computational linguistics at Goethe University Frankfurt. The resulting tool, Beta Writer, quickly produced summary text, chapter introductions, and a table of contents. It clustered articles that showed patterns of similarity. And it automatically provided references, in the form of hyperlinks, for material it quoted.
Like virtually all AI-generated text these days, Beta Writer's writing style is no threat to its highly creative human competition. The AI's prose has all the flash of the regulatory warning on the back of a pill bottle and all the pathos you might find in the directions for use on a tube of hair gel. Moreover, passages churned out by Beta Writer occasionally devolve into incomprehensibility--for example, try decoding this sentence: "That might consequence in substantially high emphasizes and henceforth cracking or delamination."
Even with its shortcomings, Springer Nature's first foray into machine-generated scientific text is turning heads. And many researchers believe it is only the beginning. "I believe that we will see more efforts like this, in particular in areas with a very large literature," says Frank Keller, a professor at the University of Edinburgh who specializes in natural language processing and cognitive modeling. Rico Sennrich, an assistant professor at University of Edinburgh who also specializes in natural language processing, concurs: "Technically, automatic text summarization is a task that AI is very suited for. It plays to the strengths of current AI algorithms--learning patterns from large collections of texts and their summaries, and then applying the same patterns to new texts."
Encouraged by Beta Writer's first outing, Springer Nature is already hard at work refining the algorithm and moving ahead with plans to generate additional scientific texts with it, Henning says. The academic publishing house is currently putting together AI-generated prototype texts based on work in the humanities and the social sciences.
In the longer term, Henning sees a perfected AI writer capable of summarizing ongoing developments in countless research niches generating scores of academic texts. Those texts would provide researchers easy access to new knowledge and allow them to more easily, quickly, and thoroughly stay abreast of new developments in their fields. Keller believes that the summaries would be particularly useful "in large, fast-moving research areas." He adds, "Researchers would hopefully use them with caution, just as they use other tools (like Google Scholar) with caution." Mats Rooth, a professor at Cornell University who specializes in computational linguistics and natural language semantics, believes such texts would be especially welcomed in niche areas of academic research, where summary tomes are too often a rare find.
But other academics are a bit more skeptical of machine-generated academic writing. "I don't think automated summaries will or should attain a status comparable to human-produced texts," Edinburgh's Sennrich says. "Researchers already rely on summaries, in the form of paper titles and abstracts, to guide their attention. Beyond that, for example in selecting which new publications are worthy of attention, I'd be wary of automatic algorithms. Current algorithms may be able to identify and select 'prototypical' publications, but very innovative and impactful research is often non-prototypical. Technical limitations aside, the value of automatic summaries will also depend on them being used sparingly, ethically, and with humans in the loop to ensure high quality."
Sennrich also says there is a risk that automated science writing could morph into a kind of McResearch Monster. "If people see automatic summaries as a cheap way to produce content to monetize, such content could become a new kind of spam that would create additional burden on researchers to sift through."
One thing is certain: No matter how the evolution of AI-generated scientific texts ultimately plays out, the release of Springer Nature's machined prose heralds, at the very least, the emergence of a new era in academic writing. More broadly, it heralds a new era in how writing of all kinds will be produced. It was only a few years ago that journalists started joking that their jobs would be taken over by robots. So far, that's still a pretty good joke. But while the jobs of most writers appear safe for the time being, they may increasingly find themselves rubbing shoulders with their silicon counterparts. And they may find their new coworkers arriving much sooner than they expected.
New York, New York
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||News and Analysis of the Global Innovation Scene|
|Date:||Nov 1, 2019|
|Previous Article:||Breaking the Battery Bottleneck.|
|Next Article:||The 2019 Maurice Holland Award.|