Privately bringing on-device biometrics and age estimation to diverse market sectors
The desire to protect children from inappropriate content and goods is bound up with broader concerns about trust in the digital world, and running into the same issues around the sensitivity of biometric data.
Privately, a start-up based in Switzerland, is applying its edge computing expertise to putting multi-modal biometrics, age estimation or age verification on end-user devices. CEO and Co-founder Deepak Tewari tells Biometric Update in an interview that the company started with edge computing for self-help use cases, often involving children online.
One example is the detection of nudity in an image before it leaves the device to trigger a warning and different engagement to prevent it from being shared.
The basis of the product came out of a research project funded by the Swiss government and carried out in collaboration with the Idiap Research Institute under the guidance of Professor Sebastian Marcel, Tewari recounts.
With industry leaders like Yoti and FaceTec coming under fire, Tewari saw the situation being addressed in a way that vindicates Privately’s focus on keeping biometric data on-device.
Privately has always operated as a B2B technology provider, working on a BBC kids app and a trial with Epic Games.
Now, Tewari says it is experiencing strong interest from a broad collection of industries to use its face and voice biometrics, alone or in combination, for liveness detection and age estimation.
The device Privately’s technology runs on can include a desktop PC, but again, he explains, “none of the user’s biometrics ever leave the user’s browser.”
Face biometrics enable Privately to be implemented for identity verification with ID documents, while voice enables the assurance of a second modality, or a lighter-touch process that might be comfortable for users in situations in which they do not want to turn on a webcam, like accessing an adult site, Tewari points out.
The voice is also protected from spoofing by prompting the user to read a sentence for the probe data. The complexity of the sentence can be varied to suit the use case.
Proving it works
The UK’s Information Commissioner’s Office has performed an audit on Privately’s application, and Tewari says the ICO loved it as a “zero-data application.”
Privately’s age estimation has been tested and certified by the UKAS scheme, and Tewari says its mean age error rate is 1.26; better than Yoti’s public numbers.
The company is now pursuing opportunities with companies in sectors from advertising to transport to vaping.
In the latter case, a vape pen can be bound to its user, with a spoken phrase used to unlock the device with voice biometrics. As in KYC use cases, Tewari says, this application requires liveness: “it should not be easy to fool the system.”
The operator or data controller has no access to biometric data, however, or need to store it. Neither does Privately. This protects both from the risk of a data breach and potential liability.
“When this page calls for Privately tech, our machine learning models get downloaded to the browser, and they execute completely on the browser,” Tewari explains.
Privately also uses federated learning to learn from the model’s age estimate compared with what an NFC ID document says.
Bringing on-device age checks to the mainstream
Privately is now talking to investors, and considering expansion in North America.
There are a number of scenarios, including companies looking to avoid liability under BIPA and similar regulations, that could benefit from keeping biometric data in the hands of the user. The California Age Appropriate Design Code is sparking interest from lawyers in the U.S. as well.
“From the biometric data handling point of view, the burden to the industry is completely reduced in this kind of an implementation,” Tewari says.
The technology can also be deployed as a scanning feature for retail self-checkouts.
Privately is currently talking to “a pretty big POS player,” according to Tewari, but most companies in the space are currently wait to see how the law develops.
The company is also proposing voice for verification and re-verification in KYC space, but in its zero-data, user privacy centric way, using NFC-enabled ID documents.
Moving processes like age assurance to the device also delivers a benefit in back-end load reduction. Instagram does 100 million checks a year, Tewari says, and at the scale with a server-side implementation, “Amazon would get quite rich.”
It all adds up to significant interest from businesses in a variety of industry verticals.
“The first movers are coming from the adult industry, from the restricted goods industry, and interestingly from the advertising industry,” Tewari says. “The advertising industry feels that a lot of money is being misappropriated, because if Instagram says that I’m serving your ad to a 25-year-old but it’s actually a 12-year-old, that, say, real-estate ad is actually wasted.”
Not all biometric models are light enough to run on a user device, however.
Privately’s model is light enough, Tewari says, even with multiple modalities and the capability to run different kids on checks.
That differentiates the company’s technology and gives Tewari optimism that production deployments are ahead.
