Large robot holds a voter and voting booth in its hand.

Out, Damned Bot!


The 2016 presidential election revealed how much a few social media accounts can influence world events. A group of Russian trolls and bots—people and algorithms posting online content on behalf of political organizations—reached millions of Americans. Well before their sway in that election became known, however, members of NYU’s Social Media and Political Participation (SMaPP) lab had taken interest in their role in Russian politics. “We were studying Russian bots before Russian bots were cool,” says Joshua Tucker, an Arts and Science professor of politics, codirector of SMaPP, and director of NYU’s Jordan Center for the Advanced Study of Russia. “We were interested in the larger question of how authoritarian regimes respond to online opposition.” The first step to understanding bots is finding a

way to identify them on a mass scale. In a study published last year in Big Data (the leading peer-reviewed journal on the topic), they demonstrated such a method. Using more than a thousand Twitter accounts they’d labeled as human or bot, they created a machine-learning algorithm to recognize the telltale signs, based on a combination of 42 features such as friend count and use of hashtags. They learned that the bots didn’t necessarily tweet about Russian politics more than people, that they often tweeted headlines without links (perhaps to affect trending topics), and that they sometimes were anti-regime. Now that the bots have hit closer to home, the team hopes to see if their tools can transfer to American politics.
—Matthew Hutson
(Illustration by John Tomac)

Naming Rites


If you’re hoping to start a new company or launch a new product with a catchy name, chances are you’re already too late. The landscape of memorable, pronounceable trademarks is vast but not infinite, and its colonization is accelerating. A study in Harvard Law Review by Barton Beebe and Jeanne Fromer, professors at the School of Law, quantifies how slim the pickings have gotten. They looked at nearly 7 million trademark applications and 128 million dot-com names and found that most of the words we use daily are taken, as are most last names and short made-up words. “You could invent words, put words together, change the spelling,” Fromer says. “But we might actually be running out of competitively effective trademarks.” Among the duo's proposed solutions? They advocate for the release of unused trademarks and increasing maintenance fees, which seems smart—that is, unless you want your start-up to be named Ljksfdksdl.
—Matthew Hutson
(Composite by Nathaniel Kilcer)



Gas station/convenience store with the name “Although” on the canopy.

Pictures For Health  


Breast cancer will kill 40,000 women in the United States this year, and catching malignancies early can mean the difference between life and death. Researchers at NYU’s Center for Data Science are using artificial intelligence to scan mammogram images for irregularities, training the algorithms on the largest data set of its kind: more than 800,000 high-resolution labeled pictures. Their method processes four images in parallel—two perspectives of each breast, using 100 times more pixels than typically used—and then combines the analysis to catch asymmetries. And their analysis doesn’t just classify scans as cancerous or not, but it also gives a probability of a normal or benign outcome, highlighting precisely where in the breast it suspects problems. The method’s accuracy doesn’t quite match that of a human doctor, but human and machine combined are better than either one independently, says Krzysztof Geras, an assistant professor in the School of Medicine’s radiology department. Some doctors fear losing their jobs to robots, but in the end they will likely collaborate with them. According to Geras, those who have seen the system “are actually very enthusiastic about it, because it can make their work a lot easier.”
—Matthew Hutson

Licking Zika


When detecting viruses, speed saves lives. Daniel Malamud, a professor at the College of Dentistry and an adjunct professor at the School of Medicine, had developed convenient saliva-based tests for HIV, Ebola, and malaria. Then the 2015 Zika outbreak in Brazil led to thousands of infants born with malformed heads. “When I saw the babies, I said I’ve got to do something,” Malamud says. He and his colleagues developed a saliva-based Zika test that’s far superior to earlier detection options because the virus disappears from blood in one to two weeks but remains in the salivary glands for several months. Results are delivered in less than an hour (rather than the weeks needed to screen blood or urine), the device that does the testing is portable, the results are more precise, and it estimates the age of the infection. Malamud’s method can be applied to other pathogens. “In my mind the most important thing that I’ve done with my lab is create a protocol,” he says. As soon as a virus’s nucleic acids and the body’s antigens are known, a saliva-based test can be created.
—Matthew Hutson

The Disaster Artist


Just about the worst time to plan a disaster response is during an actual disaster, when emotions are high and communication may be compromised. Joshua Epstein, professor of epidemiology at the College of Global Public Health, works to build game plans ahead of time. Since you can’t subject cities to attacks simply to take notes, he builds models of cities in computers and populates them with millions of virtual citizens, complete with social networks, commuting patterns, and rudimentary beliefs. His lab has modeled scenarios such as evacuation planning in response to toxic plumes in Los Angeles and Zika virus outbreaks in New York City. For the Big Apple model, “what we really built was an entire urban health dynamics simulator,” Epstein says, where contagious disease can be studied as well as “the entire range of public health threats, in principle including addictions and even violence.” Such work can inform travel restrictions or vaccine distribution, or just illuminate human nature by the light of a computer monitor.
—Matthew Hutson

Abstract cityscape with red toxic plume.

Courtesy of the NYU Agent-Based Modeling Lab

Dissing What’s Disingenuous


More than a trillion dollars’ worth of counterfeit goods are sold each year, but Courant Institute of Mathematical Sciences researchers are coming to the rescue. They’ve developed a gadget that attaches to a smartphone and takes microscopic photos of items, then uses artificial intelligence (AI) to compare the items’ surfaces with an online database. The AI has been trained on 3.5 million examples of real and fake items including clothes, pills, electronics, handbags, and toys. Several of the researchers launched a company called Entrupy to commercialize the service, focusing first on luxury handbags. Lakshminarayanan Subramanian, a Courant computer science professor, says accuracy on the 16 brands they cover is 98 percent and that retailers and resellers have checked more than $50 million worth of luxury items so far. Entrupy’s software looks for subtle clues in the leather, stitching, logo, printing, and finishing. And this futuristic testing isn’t strictly for new products—it’s authenticating vintage handbags, too, some more than 100 years old.
—Matthew Hutson

Designer handbags deemed fake.