While industry analysts and venture capitalists argue about which nascent technology trends will stick around, one federal agency is betting its software can answer the question for them.
The Intelligence Advanced Research Projects Activity, a group created to coordinate research for the intelligence community, has designed an application to identify technology that isn’t well-known today but might be in three to five years.
Called “Foresight and Understanding from Scientific Exposition,” or FUSE, it scans large volumes of academic journals, patents and other formal scientific documents for hints of emerging technologies. Government scientists could eventually use the analytic system to prioritize research and development. For instance, it might point them to new energy technology worth further investment or investigation, said project manager Dewey Murdick.
It isn’t an entirely new concept. For years, technology forecasters have considered the number of mentions in academic literature when predicting whether a concept will stick. FUSE uses this data but also categorizes the technology by examining the words scientists employ to describe it: Is it just a much-hyped idea, or is the scientific community actually bringing it to fruition?
Though IARPA oversees the project, several contractors do the bulk of development. A handful of FUSE teams work across BAE Systems, Columbia University, Raytheon and SRI International, among other parties. The project’s budget usually runs between $10 million to $15 million a year.
FUSE operates sort of like a search engine today, except users don’t type in phrases. Instead, the application continuously churns through a database of publicly available scientific literature in English and Mandarin Chinese. It periodically surfaces lists of technologies, ranked by the likelihood they will eventually become prominent. FUSE defines prominent technologies as those whose mentions in scientific journals are expected to grow significantly over the next few years — 100 or 200 percent.
So far, FUSE has pointed toward topics including quinone, an organic compound that could be used for better energy storage, and perovskite, a material used in tin-based solar cells. FUSE ranks quinone high because scientists often write about its potential applications across industries, Murdick said. “It’s moving out of the organic chemistry space, and it’s moving into batteries.”
The software also tries to gauge a technology’s maturity. Patents involving an early-stage technology often spend several words explaining it, for instance. When writing about quantum computing, which relies on atomic particles known as “qubits” instead of ones and zeroes, scientists don’t say, “I am using quantum computing to do X,’ ” Murdick said. “They say, ‘I hope to develop a part of a component that could be used in quantum computing.’ ”
Scientists developing nascent concepts are eager to contrast their work with existing research when filing patents and articles, said Olga Babko-Malaya, a senior research engineer at BAE Systems working on FUSE. FUSE counts the number of times a term is contrasted with similar technologies and assumes that as the term becomes more commonplace, this number will decline, she said.
FUSE also assigns more weight to technologies mentioned in highly reputable journals, and to certain companies filing patents — this helps it gauge whether the concept is attracting the best scientific research talent.
IARPA has designed FUSE to crunch academic research and patents, Murdick said, but has intentionally left out market data about business and consumer adoption. After FUSE surfaces potentially prominent technologies, another IARPA project, called “Forecasting Science and Technology,” surveys tens of thousands of human technology experts to assess how viable a certain technology is. Experts weigh in online through a Web site called SciCast. Collectively, for instance, they believe there is only a 31 percent chance utilities will install a quinone-based battery by January 2017.
Without human input, the system still struggles to separate fads from longer-lasting trends. “There’s still a lot of noise in the system,” Murdick said.
The team is still tweaking FUSE’s algorithm to ignore fads, he added. Right now, it just has a working definition: “If it lasts for three years, or two to three years in a community [publishing] 20 papers per year . . . then we know it’s much more likely to emerge as a [more permanent scientific] community, as opposed to a passing fad.”