In a 2020 Rise of AI Virtual Chat, Fabian Westerheide interviews Shaun Moore, CEO of the computer vision company Trueface, about AI surveillance. Westerheide opens by surveying current uses, from France detecting mask compliance to Baidu's face and temperature scanning in China and Israeli border behavior analysis, and notes the EU's temporary ban on facial recognition alongside its own border control ambitions. Moore describes Trueface's shift from building cameras to selling algorithms and software that run entirely on client infrastructure, its use in touchless stadium access and age verification for alcohol and tobacco sales, and the company's 2017 partnership with the United Nations to build more globally representative training data after finding public datasets skewed heavily Caucasian and male. He argues facial recognition should stay a decision-support tool with a human always in the loop, describes Trueface's ethical screening of clients, and compares surveillance norms across the US, China and Europe. The conversation closes on data correlation as the deeper privacy issue, the three distinct uses of facial recognition, and AI's role in future warfare through infrastructure disruption rather than autonomous killing.
Key takeaways
- Trueface runs its facial recognition technology entirely on client infrastructure and never accesses or stores client biometric data itself.
- Public face datasets are roughly 70 percent Caucasian male; Trueface partnered with the United Nations in 2017 to build more globally representative training data.
- Trueface's age-verification tool, built for alcohol and tobacco sales, has a mean average error of about two and a half years.
- Moore says Trueface requires a human to make the final decision in any consequential use, with the system providing only a confidence score.
- Moore distinguishes three uses of facial recognition, one-to-one, one-to-few and one-to-many, arguing surveillance and access control get wrongly conflated in public debate.
Detailed notes
What counts as AI surveillance today
Westerheide frames AI surveillance broadly: camera-based behavior and face analysis, France checking for mask compliance, China's real-time censoring of text messages critical of the government, credit card fraud monitoring, Baidu's combined face detection and temperature scanning in China, and Israeli border control combining behavior, movement and temperature analysis to flag potential threats. He notes the European Union has temporarily banned facial recognition technologies while simultaneously developing an EU border control system, and frames the discussion around how much AI surveillance can serve a greater good against privacy.
Trueface: from cameras to algorithms, kept on client infrastructure
Moore describes Trueface's pivot from building physical cameras to selling algorithms and software, with a core product enabling touchless door access via face recognition. All Trueface technology runs on client infrastructure; the company holds no access to client data, which he calls its pledge to "data transparency." Clients train the pretrained system on their own database, and data is never shared across clients. He describes stadium deployments giving season ticket holders and VIP guests touchless, temperature-checked access, and says he is not aware of US police departments actively using facial recognition on protesters, though he expects the issue will require regulatory guardrails.
Age verification and tackling bias in training data
Moore explains Trueface's age-estimation tool, built primarily for automating alcohol and tobacco age checks, with a mean average error of about two and a half years. He says the system is trained on millions of tagged reference images without personal identifiers, and that public face datasets are roughly 70 percent Caucasian male, producing higher accuracy for that group. In 2017 Trueface partnered with the United Nations to collect more globally representative data, later publishing its results on the third-party FairFace benchmark, which Moore says makes Trueface the first company to do so; he adds that, contrary to common concern, Trueface recognizes women more accurately than men. Clients can retrain the algorithm on their own target population for better regional accuracy.
Human-in-the-loop decisions and where Trueface draws the line
Moore says facial recognition should never be the sole basis for a consequential decision; Trueface's software licensing agreements require a human to make the final call, with the system providing a confidence score as one data point. He says Trueface has refused to work with private investigators and declined certain regions over humanitarian concerns, vetting clients for four to six months before technology goes live. He argues current US regulation is fragmented compared with the UK and EU and says Trueface supports balanced federal regulation.
US, China and Europe: different baselines for privacy
Discussing cultural differences, Moore says Chinese attitudes toward privacy and state monitoring differ fundamentally from US and European norms, and that he does not expect similar acceptance to emerge in the US given its checks and balances. He argues no US law enforcement agency he has spoken with uses facial recognition in real time, only after a crime to generate investigative leads. A live poll during the session found about 40 percent of attendees would trust enterprises to manage their data, versus 20 percent trusting government. Moore adds that data correlation, not raw collection, is the harder privacy problem: tying a single recognition event to a person's broader pattern of movement and behavior.
Q&A highlights: three silos of facial recognition and AI in warfare
Moore distinguishes three uses of facial recognition: one-to-one phone unlocking, one-to-few access control for buildings, and one-to-many investigation or surveillance, arguing these get conflated in public debate; he cites San Francisco's citywide ban, later amended after it accidentally outlawed employees' iPhone face unlock. On AI in warfare, Moore says Trueface's position is that a human must always make critical decisions, and that Project Maven drone operations reportedly kept two humans in the loop, though he cannot confirm this independently; he considers AI-enabled attacks on infrastructure, such as power or water systems, a greater future threat than autonomous weapons.
Quotes
Our fundamental goal as a company is to teach cameras to see like human beings.
Shaun Moore
Everything that we've done in the basis of our technology is building the ethical use of computer vision.
Shaun Moore
The concern is not the tool, it's the data, it's how is your data being used, who has access to it, where is it stored, how long is it stored for.
Shaun Moore
Questions answered in this talk
How does Trueface address bias in facial recognition training data?
According to Shaun Moore at this Rise of AI Virtual Chat, Trueface partnered with the United Nations in 2017 to collect more globally representative face data after finding public datasets skewed roughly 70 percent Caucasian male, and later published its accuracy results on the third-party FairFace benchmark.
What are the different uses of facial recognition technology?
According to Shaun Moore, facial recognition splits into three silos: one-to-one matching such as phone unlocking, one-to-few matching for building access control, and one-to-many matching for investigation or surveillance, which he says the public often wrongly treats as a single category.
Where does Trueface store the biometric data it processes for clients?
According to Shaun Moore, Trueface's technology resides entirely on client infrastructure; the company itself has no access to or storage of that data, which remains the property of the client.