Inside | Real news, curated by real humans
Inside AI

Inside AI (Dec 20th, 2019)

1. A newly released study by the National Institute of Standards and Technology (NIST) has confirmed that most of the world's advanced facial recognition algorithms have inherent biases against people of color, women, children, and the elderly. The agency tested 189 algorithms from 99 organizations that collectively power the majority of the world's facial recognition systems. A majority displayed worse performance on any faces that were not white, according to the landmark study released Thursday. In the case of “one-to-one” matching, or matching photos of a single person in a database, the majority of systems produced more false positives (finding a match when there wasn't one) for Asian and African-American faces compared to Caucasian faces, in some cases by a factor of 100. The worst false-positive rates were for African-American women. As a result, lawmakers called the results "shocking" as they seek to better regulate the technology, and the NIST is urging face recognition developers to find ways to mitigate these biases. - BBC

2. Facebook and Twitter jointly removed a sprawling disinformation campaign that was spreading pro-Trump messages using accounts with realistic-looking avatars created through AI. Saying that the campaign marks a new escalation in the war on such disinformation-spreading tactics, the companies stated they removed hundreds of fake accounts that were purporting to be owned by people with AI-generated photos. The companies say the campaign is linked to pro-Trump propaganda site TheBL.com, whose Facebook page was also removed Friday. The BL is the creation of Epoch Times, the media company tied to the Falun Gong religious movement, many of whose adherents are Chinese immigrants living in the U.S. to escape persecution. The Falun Gong has been outspoken in their support of President Trump, because he has stood up to the Chinese government. And The Epoch Times made headlines in August after it was found it had spent $1.5 million on pro-Trump ads on Facebook, and was spreading the QAnon conspiracy theory. - WAPO

A version of this story first appeared in Inside Social.

3. AI experts Rodney Brooks and Kai Fu Lee challenged Elon Musk's timeline of producing a million fully autonomous robotaxis by the end of 2020. In April, the Tesla CEO predicted that Tesla would have at least that number of robotaxis on the road next year "with no one in them,” although he did warn investors, “Sometimes I am not on time, but I get it done.” Of course, there have been no model robotaxis so far, or any fully autonomous production cars for that matter. In a joke made earlier this year, Brooks and Lee vowed that if there are actually a million Tesla robotaxis functioning on the road in 2020, they would "eat them." The Next Web, which analyzed the technological breakthroughs and government red tape that Musk would likely face to make this a reality, reports that "it’s safe to say Lee and Brooks won’t be eating any cars" by this time next year. - TNW

4. Santa Fe-based Descartes Labs is using AI to detect wildfires in less than 10 minutes. The company's wildfire detector tool leverages two geostationary satellites, GOES-16 and GOES-17, that hover above the Earth and stream back thermal infrared images every five minutes. Any fires will appear as bright distinct spots in the images that stand out from the surrounding area. Once a hot spot is detected, the Labs' AI algorithms filter out false alarms such as fireworks shows and controlled agricultural burning. It typically takes nine minutes "from the point of the satellite capturing the image in space to us sending out an alert," according to Descartes applied scientist Clyde Wheeler. - BUSINESS INSIDER

5. Joelle Pineau, a machine-learning scientist at McGill University and Facebook, wants more AI researchers to open up their code and methods because she says it could lead to better reproducibility. Pineau launched a reproducibility competition - challenging researchers to re-create each other's work - at this month's Conference on Neural Information Processing System in Canada. Organizers encouraged attendees to submit their code and review a special checklist for submitted papers, which asks for details of models and other characteristics. The efforts could be paying off: A year ago, half of accepted NeurIPS papers contained a link to code, compared to 75 percent this year, Pineau noted. - NATURE.COM

6. A Facebook AI solved 500 calculus problems with 98 percent accuracy, according to a new paper published on arxiv.org. Facebook AI researchers François Charton and Guillaume Lample trained the AI on millions of calculus problems involving integration that were randomly generated by a computer. The AI used then natural language processing to tackle the problems, beating the 85 percent accuracy score of a comparable standard program. Charton says additional training could one day help the AI solve math problems that are too difficult for people. - NEW SCIENTIST

7. Reuters is offering a free online course - called Identifying and Tackling Manipulated Media - that's designed to help journalists spot deepfakes. Facebook’s Julia Bain said the goal is to "work across industries" so people can identify and deal with deepfakes and other manipulated media, which has the potential to spread disinformation. The course is already available in English, French, Spanish, and Arabic, and will open up to more languages next year. - PRESS GAZETTE

Written and curated by Beth Duckett, a former reporter for The Arizona Republic who wrote a book about the solar industry and frequently covers hobby and commercial drones. You can follow her tweets about the latest news in artificial intelligence here.

Edited by Sheena Vasani, Inside Dev editor.

Subscribe to Inside AI