Smile for the Camera: The reality of AI in retail age verification
– by Ed Heaver, co-founder and CEO of Serve Legal
Imagine walking into a store where the shelves seem to know your age better than you do. Welcome to the future of retail, where age estimation technology is set to revolutionise the shopping experience. With just a glance, sophisticated AI systems can pinpoint your age, transforming experiences around age-restricted sales. Say goodbye to guesswork and hello to a smarter, more personalised shopping journey.
Facial biometric technology
Facial biometric technology is now finding its way into everyday retail environments, promising better efficiency in regulating sales of age-restricted products like alcohol, vapes, tobacco, and cigarettes. These facial biometric systems, designed to estimate a customer’s age with a quick scan, are an emerging tool for retailers looking to enhance compliance and streamline operations. Yet, as these systems move from concept to checkout counters, it brings a range of challenges that need addressing. The benefits of efficiency must be balanced with the concerns about accuracy, fairness, and privacy.
At Serve Legal, where I work as co-founder and CEO, we’ve spent years ensuring that retailers comply with age-restricted sales regulations. Our work has provided us with deep insights into the complexities of enforcing these laws, and the potential for facial biometric technology to transform this space is undeniable. To that end, we’ve partnered with academic experts from the University of Durham to build a comprehensive programme to audit these systems across various retail sectors. Our goal is to ensure that this technology is not only effective but also equitable and transparent, both to customers and retailers.
How does AI work in retail?
Historically, facial biometric technology was used primarily for security purposes, but now its application extends to age estimation at checkout. The technology operates by analysing an individual’s facial features and through a trained algorithm estimates their age. A threshold which is set (usually 25 in line with the challenge 25 policy adopted by most retailers) results in customers estimated below the threshold being flagged as a potential minor, therefore requiring an ID challenge. This could revolutionise the way retailers manage age verification, offering a seamless, automated solution that reduces human error and speeds up the checkout process. Moreover, reputable age estimation solutions are privacy preserving with users’ facial biometrics not being stored.
In theory, it could even reduce the awkward moments of manual ID checks and prevent potential customer-employee conflict, as well as creating a more fluid and customer-friendly experience.
However, in real-world retail environments, several challenges arise, and accuracy remains a concern for those people looking to implement the technology. While these systems can perform well in controlled settings, the variability of real-world conditions—such as lighting, camera angles, and the diversity of the customer base – can affect their performance. These factors can lead to increased margins of errors in estimations, resulting in a potential for minors to not be flagged by the system.
Tackling Bias
The industry is well aware that these systems can exhibit bias, particularly against people of colour. For example, data suggests that current facial biometric systems are typically more accurate at identifying white male faces than those of black women. This disparity not only undermines the effectiveness of the technology but also raises concerns about fairness and race/gender-based discrimination. (See NIST Report: https://pages.nist.gov/frvt/reports/aev/fate_aev_report.pdf)
For retailers adopting this technology, providing due diligence is paramount. Failing to address these concerns can lead to accusations of unfair practices or even racial prejudice, particularly if the technology is perceived as discriminatory or biased.
Liveness Detection
Another crucial element of facial biometric tech is designing systems that can effectively counteract fraudulent methods. This is where liveness detection comes into play. Liveness detection is a key component of biometric age assurance and digital ID technology. It works by verifying that what is being presented to the biometric system is a live human being, rather than a photo, prosthetic, or a digitally injected video designed to trick the system.
Liveness detection methods can be broadly categorised into passive and active approaches. Passive liveness detection requires no specific actions from the user, minimising friction in the verification process. In contrast, active liveness detection may require the user to perform tasks such as blinking, turning their head, or adjusting the camera distance. While active methods might introduce a slight inconvenience, they are often considered more secure, reducing the chances of successful spoofing attempts.
Current testing frameworks can be misleading and fall short in thoroughly evaluating these age-estimation systems, as there are many different methods for evaluating liveness and the accuracy of these methods can vary significantly, providing different levels of confidence. Serve Legal has access to the large cohort of engaged auditors, who can conduct live presentations at a scale that is not exhibited by existing testing protocols, identifying bias and giving a more accurate and realistic demonstration of the resilience of these liveness detection models.
Promoting Fairness Through Rigorous, Independent Testing
Our testing framework, developed in collaboration with a leading academic expert from the University of Durham, is designed to rigorously evaluate these systems across diverse demographic groups. Thousands of our young auditors voluntarily participate by taking selfies, or performing live presentations, that are uploaded to the system for age verification analysis. We then generate detailed reports based on these tests, identifying bias, providing liveness detection analysis and highlighting inaccuracies that emerge. Retailers can also assure the reliability of their technology by utilising our vetted in-situ tests.
The insights gained from this process are invaluable. They allow us to work closely with our retail clients to recommend best practices, ensuring that they navigate the complexities of deploying facial biometric responsibly.
The Bottom Line
To someone like me, who’s been immersed in the industry for over three decades, the promise of a more efficient, checkout experience enhanced by AI is truly exciting. This technology offers a revolutionary approach to age verification, streamlining the process and reducing manual oversight.
However, realising this potential requires more than just enthusiasm, it demands continuous monitoring and thoughtful development. AI systems have the unique ability to learn and adapt, which can be highly beneficial if we commit to refining and developing them responsibly. Regular updates and rigorous testing are essential to ensure these systems remain accurate, fair, and compliant with evolving regulations.
Maintaining transparency in data practices and staying ahead of regulatory changes will be critical in building and preserving customer trust. By focusing on these areas, we can leverage AI-driven age verification to enhance the retail experience while upholding the highest standards of fairness and privacy.
