Roundup Welcome to this week's AI roundup, where The Reg has - among other things - tried to lift the veil off the Clearview saga, and got the low-down from Nervana co-founder Naveen Rao, who is leaving Intel following the cancellation of the startup's neural network training chip.
Clearview drama: The controversial startup Clearview, exposed for scraping over three billion photos on people's social media accounts, just can't seem to catch a break.
First, a miscreant "gained unauthorized access" to Clearview's internal list of customers to see what companies and organisations had used its facial-recognition software. The data also revealed how many user accounts those customers had created, and the total number of searches they conducted with its tools. It's unclear how the intruder nabbed the data. Clearview said it has patched the security flaw since.
That list was then reviewed by BuzzFeed News; there were about 2,900 different names on it. Some institutions did not have logins or did not use Clearview's software to run any searches. Out of the companies that did, however, it's not immediately clear if they had merely trialled the tool or if they had entered a contract to pay for the service.
Clearview's CEO, Hoan Ton-That, has made it clear that his company's facial-recognition app is for law enforcement. And that is indeed the case. 2,228 clients – the vast majority – are law enforcement agencies across several countries in North America, South America, Europe, Asia, and Australia, including the US, Canada, United Kingdom and Ireland. Altogether, these 2,228 customers have run a total of nearly 500,000 searches.
In the US, some of these law enforcement agencies include the Department of Homeland Security's Customs and Border Protection, and its Immigration and Customs Enforcement has hundreds of accounts. Organisations from the Department of Justice were also on the list too. The Secret Service, the Drug Enforcement Administration, the Bureau of Alcohol, Tobacco, Firearms and Explosives as well as the FBI have collectively conducted over 10,000 searches.
There are also numerous police and sheriff departments in various states – Miami, Philadelphia, Indiana, Chicago, Atlanta and New York – using the tech. In fact, the New York Police Department previously said it didn't work with the murky startup at all. But the latest documents revealed that the NYPD has in fact performed over 11,000 searches – the highest number of any organisation on the list. Over 30 officers have accounts. Chicago Police Department forked out $49,875 to obtain accounts for 30 employees to use the service for two years.
What's more interesting, perhaps, are the customers that aren't from law enforcement at all. These include retail stores like Macy's and Walmart, as well as banks like Wells Fargo and Bank of America. Particularly chilling cases include educational institutions; most are universities and two were high schools. It's not clear why these folks wanted to use facial recognition.
You can read about that in more detail here.
More Clearview drama: Some details about Clearview's facial-recognition app were left sitting in an Amazon Web Services S3 bucket that was public.
Some of the screenshots captured by Gizmodo show snippets of code that hint at various features, including a "private search mode" and something that scans barcodes on people's drivers licences. It's not clear what these things do exactly, and if they are included in the working version of Clearview's app.
The code in the S3 bucket was installed on an Android device, but the reporters noted it was difficult to see how the app really functions without a login. They did see, however, that some of the data in the app had been accessed by Android's Fine Location API, an application that returns an accurate location of the user, as well as GPS, Wi-Fi, and mobile data networks.
The app also tells users to "invite your coworkers or other investigators to Clearview for free. Just press share below to send a link with a free Clearview demo account."
Even more Clearview drama: OK, one more bit: Apple has suspended the startup's iOS developer account for inappropriately sharing its app.
Under Apple's rules, a developer account cannot share the app to any third parties. A developer account can distribute the app to multiple people within the same organisation if there are numerous employees working on the app, for example. But developer accounts are not meant to be shared with other companies.
Clearview, however, has been distributing its tool to its customers in this dodgy way to avoid putting it on Apple's app store. The startup's facial-recognition app is not meant for public use, after all. Essentially, it now means that Clearview's app cannot be used on iPhones.
"We are in contact with Apple and working on complying with their terms and conditions," CEO Hoan Ton-That said in a statement to Buzzfeed News. "The app cannot be used without a valid Clearview account. A user can download the app, but not perform any searches without proper authorization and credentials."
Clearview now has 14 days to respond to Apple before its developer account is revoked.
Nervana's co-founder and CEO exits Intel: Naveen Rao announced he has left his position as head of AI at Intel after the company decided to reshuffle its hardware team.
I have decided to leave #intelai. It has been a remarkable experience building a new brand within Intel and sharpening the focus around #ai at the company! I wish all of my colleagues well as the field of AI matures.— Naveen 0xDEADBEEF (@NaveenGRao) February 26, 2020
Never stop innovating!
The move comes shortly after Chipzilla decided to can the NNP-T, a chip for training neural networks, also known as Spring Crest. The efforts for building Spring Crest was headed by Rao, who was co-founder and CEO of Nervana, a startup that Intel forked out $350m for in 2016.
The Spring Crest vapourware was heavily marketed but faced multiple delays and was renamed a few times. Frustrated with a lack of hardware to compete with its rivals like Nvidia or AMD for a slice of the AI pie, Intel decided to splash out $2bn for yet another AI startup: Habana.
Unlike Nervana, however, Habana already had two machine-learning chips in the market, known as Gaudi and Goya. Gaudi was optimised for heavier workloads for training purposes, while Goya was for inference.
After the Habana acquisition, Intel decided to abandon its attempt to bring out the NNP-T. Nervana's inference chip, the NNP-I or Spring Hill, however, will be given to customers who preordered it.
Employees on the original Nervana team have been leaving, and now its old CEO has finally exited too. Rao told The Register that his team was very close and were "slated to tape out very soon".
"I learned a lot about what it takes to scale teams up," he said when reflecting on his time at Intel. "We were 48 people at acquisition. When we finally got the green light to grow, we went from 48 to over 500 in about eight months. I haven't met anyone that has done it faster! Crazy to see how the dynamics and culture evolves through that. And, it was a distributed international team."
Rao also said he was taking a break before deciding his next steps. "I'm leaving because things are moving well to go to market and I'm much more about creating new companies to drive technology transitions. [It's] much more about entrepreneurship for me. I'm taking some time off before jumping into my next thing."
Facebook pledges to fund research that fights misinformation: Facebook might try to use AI to flag violent or pornographic smut on its platform, but, unfortunately, that tactic doesn't quite work for fake news.
Now the social media giant is calling scientists for help by splashing $2m to fund research that studies misinformation and polarisation online.
"To advance our understanding of how technology impacts people and society, we've strengthened our commitment to conducting social science research in partnership with academics globally," Umer Farooq, who leads community integrity research at Facebook, announced.
Facebook hopes that by having a better understanding of how fake news spreads, how online communities become polarised, and how conflict starts on social media, it can tackle the problem of misinformation and hate speech being posted and shared on its platform. Particular areas of interest include health, politics and news.
Researchers interested in applying for financial grants should do so by filling out this form here by (we're not kidding) 1 April. ®