Clearview CEO pitches developer focus as key to Consent facial recognition success
The new Clearview Consent identity verification product from Clearview AI brings the company’s facial recognition algorithm to an entirely new set of customers and use cases, but the strategic shift also brings the company closer to its developer roots.
Back before Clearview burst into public consciousness with a prominent New York Times article in early-2020, the company had built its facial recognition for a variety of different applications, CEO Hoan Ton-That tells Biometric Update in an interview.
The company tested use cases around security alerts from IP cameras, and physical access control. “It was actually a surprise,” Ton-That says, that law enforcement wanted to use Clearview’s facial recognition – along with its database of images from the web.
Having commercialized as a law enforcement provider, Clearview has seen a lot more interest since it’s strong showing in NIST’s FRVT, according to Ton-That.
“We realized it would be a good idea to start providing this algorithm to other entities, because there are a lot of use cases,” he explains about the development of Clearview Consent.
Ton-That says that because of his background as a software developer, the company’s approach is to look at the market to see what’s missing and then build a product to address that need. Clearview Consent makes the company’s facial recognition appropriate for various use cases, he says, with an algorithm he notes has placed second for accuracy in the FRVT’s Wild category.
The developer product is available in cloud or in a Docker container for on-premises deployments, with a version for deployment on edge coming soon.
Clearview Consent is being pitched as infrastructure for developers, and Ton-That compares it to Amazon and Microsoft’s offerings that come as part of larger image recognition kits. Those companies, however, are not fully invested in facial recognition, which Ton-That argues is reflected in falling NIST rankings for Microsoft. Businesses also cannot deploy their solutions in Docker containers or at the edge.
Ton-That believes Clearview Consent is also differentiated by the company’s U.S. origins, and its easy-to-understand developer API.
“We’re doing it in a way that we think by being something that’s inexpensive, and easy to understand, a software developer or somebody building something can use it, test it quickly, and get results.”
It also provides clarity in a market where Ton-That says, “it’s unclear how much certain products cost.”
“The thing about Amazon and Microsoft is that easy-to-understand pricing,” he adds.
Clearview is also banking on ease of use and low, transparent pricing helping to increase adoption.
Facial recognition for developers and partners
“From somewhat of an outsider perspective, as a software developer, you have Stripe for example for payments, or Twilio for text messaging, so having that developer friendly part – not that we won’t support people all the way up to large enterprises, which we have as well – but I think that accessibility part is there,” Ton-That explains. “So our developer documentation is right on our website. The pricing for the cloud version is right on our website. And we think that’s a great way for a lot of people to get started.”
From there, they can move to on-premises deployment as they grow and it becomes necessary.
“We want to be flexible and very business-friendly,” Ton-That emphasizes.
The demand is evident from investments that flowed in even when the technology’s accuracy was not good enough, he says.
Clearview Consent can be built into KYC provider’s workflow, white-labeled so end-users are not even aware who is providing the algorithm they are using.
KYC providers often charge about a dollar per check, he says, so Clearview is providing its algorithm at a cost that can be built into that price.
Clearview’s presentation attack detection (PAD) technology is coming in “about a month,” Ton-That predicts.
The company plans to enable KYC and other solution providers with their own sales teams and go to market plans. Clearview’s algorithm can help them build out their workflows with authentication, support devices with lower quality cameras, or onboarding in poor photo conditions.
“With a higher quality algorithm, you’re going to have less of a drop-off rate there.”
Deepfake detection will follow PAD in Clearview’s product roadmap, and Clearview plans to continue learning from the market as it expands its business.
“It’s a big market,” Ton-That assesses. “And I think it’s going to continue to grow, especially since the accuracy levels in the last few years of all these algorithms have increased dramatically. The bias issue’s still a concern but when you look at the scores of the NIST 1 to N, that’s trying to pick a face out of 12 million, and we’re at 99.85 percent for that, and there’s minimal demographic bias.”
It is difficult to determine the actual addressable market size, Ton-That says, unpacking the facial recognition element from the related technologies.
Initial adopter Vaale is an example of a use case which may not be immediately obvious. Loan sharks are a big problem in Columbia, often charging 10 percent interest per day, Ton-That says, and a huge number of people use Buy Now Pay Later (BPNL) transactions. Providing them credit with an ID photo and selfie gives them better terms and cuts criminal gangs out of the process.
Ton-That expects that success will increases customer satisfaction and therefore spend, not just at Vaale but in general.
The company also has a KYC provider in Nigeria planning a pilot.
A question for regulators
It remains to be seen if regulators, such as in Europe and Canada, will take issue with new uses of the algorithm trained with data they say should not have been used for the purpose, Ton-That admits, but says, “We don’t really anticipate any issues or pushback on that level.”
It has not come in correspondence yet with regulators, he tells Biometric Update.
Consent-based facial recognition is “the least controversial version of facial recognition that’s out there,” though people will have questions around storage and retention, Ton-That anticipates, which can be addressed.
Consent, he says, is a big part of all legal models. He also believes the difference between a search function and a “better padlock” function is well understood at this point.
Ton-That is confident that as more people experience the benefits of facial recognition, acceptance will increase.
The discussion turns to the adoption curves of new technologies, and he notes that early adopters of Clearview are generally not in a position to run public relations campaigns about it.
“Sometimes we’re the only voice defending law enforcement use cases,” he says. “It’s starting to change.”
The banking sector is leading in adoption of consent-based facial recognition with no backlash, Ton-That observes, and they are “the second-hardest people to sell to.”
“Financial institutions and lower are all probably fine with this. And that’s similar to what I’d say with the adoption of cloud, for example, where it took more time in the government, but it got there.”
Government is smart to define standards, in Ton-That’s opinion, and in the cloud computing domain, today the U.S. government has finally moved some of its own systems to the cloud “which sounds crazy” in the context of the early days of cloud computing, when the technology’s benefits were still largely theoretical.
For facial recognition, Ton-That seems to believe we are closer to that early part of the adoption curve.
With its new product, Clearview is looking to work with other biometrics developers and integrate further into the broader digital identity industry.
“In this market, we’re going to be a vendor out of many,” Ton-That says, “so we’re really open to collaboration with other providers, say they provide fingerprints, and to be part of the whole ecosystem.”
In this way, the release of Consent could represent a fresh start in a new direction for Clearview. The algorithm is the constant.