By Dan Grimm
“Why did you decide to help build a facial recognition business? Isn’t that a little… scary?”
I have been asked that question numerous times since leaving Amazon to lead SAFR by RealNetworks. The short answer: I believe excellent facial recognition technology can enhance security and convenience. Yes, it also presents risks: it’s extraordinary technology. The choices made by companies in this space will be major determinants in how facial recognition changes daily life. That’s a serious responsibility and challenge.
I understand why facial recognition raises alarm. American political culture was founded out of reaction to government overreach. Books like 1984 and movies like Minority Report condition us to associate facial recognition with repression and limitless surveillance. However, dystopian deployment is not inevitable.
Like most technology, facial recognition can be used for good or evil. It can be used to discriminate out of ignoble intent or, inadvertently, when users deploy systems with unacceptable accuracy variances between lighter and darker-skin tones (not all facial recognition systems are created equal). It can be used to power on-going government surveillance: monitoring anyone, anywhere or even everyone, everywhere. It can be used to egregiously violate privacy: harvesting facial signatures for permanent use.
However, in trustworthy hands, facial recognition can also manifest great benefits: enhancing security and convenience in an age when we’re too often forced to sacrifice one for the other. We’ve seen this first-hand. Parents are using SAFR to gain immediate access to their child’s school campus without waiting to be buzzed in, which means less time for staff to manage visitor entry and more time helping to foster education. School administrators can also configure SAFR to notify them if, for example, a parent with a restraining order appears on a camera.
As a full-featured facial recognition platform, SAFR can similarly improve the security professional’s ability to keep us safe. At a stadium or airport, would you want security staff trying to match photos of individuals of concern from a 3-ring binder to dozens or even hundreds of video feeds in real-time or have the help of an AI-powered platform like SAFR?
I believe facial recognition will become ubiquitous in the next five years. For me, this is the key question: How do we ensure it is deployed in ways that bend our society towards justice, and not repression?
There are two ways: 1) thoughtful regulation and 2) principled design, development, and distribution.
On the first, RealNetworks supports the call for thoughtful regulation from companies like Microsoft. Regulation can help avoid a race to the bottom and set a clear foundation for competition. Such regulation should require greater transparency about the performance of systems like SAFR, require customers to provide notice when in facial recognition is in use, and limit on-going surveillance of specified individuals. We also believe such regulation should acknowledge differences between facial detection (is there a face in this scene?), characterization (what is a person’s apparent age, gender, or sentiment?), persistent unique identification (what is this person doing in a limited context, without knowing other PII?), and recognition (does this person match to someone in a database?).
On the second, RealNetworks is committed to bringing our technology to market in ways that solve meaningful human problems. That’s why we started by offering SAFR for free to K-12 schools in the United States and Canada: the team saw that SAFR could improve school security and access. Now, we are expanding beyond to needs in other industries and have begun working with a range of partners globally to make it simple for customers to use SAFR.
As we do, we wanted to make clear – to ourselves and the public – the Guiding Principles we’ll use to make decisions about how we design, develop, and deploy SAFR:
- Value Both Customers and the Public. Our customers are the organizations that deploy SAFR. We consider both their interests and the interests of the people seen by SAFR in all we do.
- Recognize the Inherent Dignity and Equal Worth of Every Person; Ceaselessly Reduce Bias. All facial recognition platforms exhibit bias. SAFR already ranks as one of the least biased in the world, but we will not stop improving until SAFR detects, characterizes, verifies, and recognizes people of different skin tones, genders, or ages with equal accuracy. We do not train SAFR to recognize ethnicity.
- Earn Trust and Keep Trust. We submit SAFR for independent performance evaluation, communicate its strengths and limitations, and guide customer’s choices about how to deploy it. We help our customers safeguard user information as if it were our own, using bank-level encryption.
- Make it Simple to Protect Privacy. We incorporate privacy tools into SAFR by design so that it’s easy for our customers to give users notice, enable users to opt-in, and enable users to opt-out and be forgotten. Our terms require customers to comply with all applicable privacy laws; we encourage practices beyond the legal minimum.
- Sell to Trustworthy Customers. We take all reasonable steps to ensure SAFR is not sold to customers that will use SAFR to violate individuals’ right to life, liberty and security of person without due process or just legal mandate.
We have already begun implementing these principles. Here are a few examples:
- We submit SAFR to the US National Institute of Standards and Technology (NIST) ongoing Facial Recognition Vendor test. That assessment shows our algorithm as the premier solution for live video based on its combination of accuracy, speed, and efficiency. Further, it showed SAFR was also among the top five least biased algorithms to differences in skin tone. That’s a great position, but we will not stop improving until we reach perfection.
- We published best practices to guide how SAFR could be deployed for secure access and other security use-cases while safeguarding privacy.
- Our contracts already prohibit partners from selling to sanctioned entities. As of this month, we are implementing a process to further evaluate potential partner sales to state-owned enterprises or government entities in a list of countries and territories where we are concerned about abuse (informed by the latest Freedom House evaluation). We will walk away from business where the risk of SAFR being used to violate individual rights is too high.
We aspire for SAFR to become the most trusted facial recognition platform in the world. Our launch for K-12 schools in the United States and Canada was a great start. We can’t wait to see how security professionals use SAFR to make offices and campuses more secure, how event organizers use SAFR to enable faster and more secure access to stadiums and other venues; how retailers leverage SAFR to provide more personalized offline shopping experiences; how other developers incorporate SAFR into their software and hardware; and so much more.
We are at the beginning of an exciting journey with our customers. Their needs will set the direction and our principles will guide our design, development, and distribution choices along the way. If you have a challenge that SAFR could solve, please contact our sales team.
P.S. If, like me, you are attracted to the challenge of developing SAFR into the world’s most trusted facial recognition platform, then join us! We’re hiring across product, design, engineering, data science, marketing, and sales.
—
Dan Grimm is the GM of SAFR by RealNetworks based in Seattle, WA.