Mysterious face-recog AI startup Clearview sued, capabilities questioned after scraping billions of web pics

Also, the most detailed brain map yet took 12 years, $40m to build


Roundup Hello, welcome to this week's AI roundup of news that's beyond what we've already covered, and it's mostly all about facial recognition.

The most detailed map of a brain cost $40m to build – and it’s of a fruitfly: Scientists have created the most comprehensive map yet of a fruitfly’s biological neural network after splashing $40m on the 12-year project.

The map doesn’t cover the whole brain of the humble drosophila, an insect commonly used in genetics experiments, only its hemibrain – a small subsection that controls how it learns, navigates, sees, and smells.

A large team of researchers from the Janelia Research Campus of the Howard Hughes Medical Institute in Virginia, partnered with engineers at Google to map 25,000 neurons, and more than 20 millions, of its connections using machine learning algorithms and detailed images of the fly’s brain.

Miniature slices of the critter’s brain tissue are captured by a series of microscopes until the whole sample is imaged. Next, these images are collated and processed by Google’s software to segment the different neurons in order to build up a full 3D visualization of the brain. These connections are then carefully scrutinized by researchers to make sure the data has been accurately represented.

The whole effort has taken 12 years so far and $40 million in funding, and the team isn't done yet; they hope to map the insect’s whole nervous system by 2022.

“It’s satisfying to see it succeed — it is the scientific achievement that I am the most proud of from my time as director, in part because it required complementary contributions from so many talented people working together for over a decade,” said Gerry Rubin, vice president of HHMI and executive director of Janelia. “But personally, I’m interested in using this knowledge to learn how the brain works.”

You can read more about that process in a paper released this week (bioRxiv).

NYPD casts doubt on how good Clearview’s facial recognition algos really are: Secretive American startup Clearview claimed to have the “world’s best facial recognition system” that could identify people’s faces after scraping together more than three billion snaps from Instagram, Facebook, YouTube, and similar.

The machine-learning based system, we're told, should be able to identify someone from their photo, by matching the pic to the corresponding entry in the image database along with a link to where it came from, thus revealing the person's identity. It still works even if the initial picture is blurry or if parts of the face are obscured, apparently.

No wonder the surveillance technology was, apparently, eagerly used by cops around the world. Clearview boasted in a promotional video that it was able to help New York City cops catch a terrorist, who planted rice cookers suspected as bombs on the subway. But the NYPD has said it didn’t use Clearview’s software at all (Buzzfeed). Some have now cast doubt on the technology’s capabilities, believing it may simply just not live up to the hype.

Illinois citizen sues Clearview: Since we’re still on the topic of Clearview, a resident in the US state of Illinois has sued the upstart.

David Mutnick filed a class-action complaint against Clearview and its founders (Zdnet). Mutnick argues that Clearview has broken the Illinois Biometric Information Privacy Act (BIPA), a law that requires companies to obtain explicit consent to collect biometric data, whether that’s images for facial recognition models or fingerprints.

The complaint calls for an injunction against Clearview, urging the startup to delete any data taken of Mutnick and any other residents of Illinois.

Supreme Court throws out a Facebook facial recognition hearing: Yes, we’re going to continue talking about facial recognition in Illinois.

The US Supreme Court has dismissed a request by Facebook to decide whether or not the web giant can be sued over its facial-recognition technology (The Hill).

As a result, the antisocial network must now deal with a lawsuit filed by Illinois residents Nimesh Patel, Adam Pezen, and Carlo Licata. That suit accuses Facebook of violating BIPA by deploying facial-recognition software that analyzes netizens faces in photos and suggests tags to identify them without explicit consent.

Facebook attempted to argue its technology did not result in "real-world harm." The Supreme Court rejected this notion.

Stop facial recognition, says Google CEO: Google boss Sundar Pichai has backed the European Commission’s proposal to halt facial recognition technology being rolled out in public areas for five years.

The moratorium would give the EU time to mull over the technology’s impacts and how it affects current General Data Protection Regulation (GDPR) laws. Pichai, who spoke at a Belgian think tank conference this week, said: “I think it is important that governments and regulations tackle it sooner rather than later and give a framework for it." (Reuters)

Microsoft president Brad Smith was more lukewarm towards the European Commissions proposal, however.

“I’m really reluctant to say let’s stop people from using technology in a way that will reunite families when it can help them do it,” he said. “There is only one way at the end of the day to make technology better and that is to use it.” ®


Biting the hand that feeds IT © 1998–2020