Inclusive Psychophysiology

The Problem of Bias in Biometrics

Biometric systems have transformed identity verification, but their benefits have not been evenly distributed. Studies consistently show that error rates are higher for certain groups, especially women, people with darker skin tones, and individuals with non-binary gender presentations (Buolamwini & Gebru, 2018; Raji & Buolamwini, 2019). The problem is structural: systems trained primarily on homogenous datasets are ill-equipped to handle global diversity.

These disparities are not merely technical shortcomings. They raise questions of fairness, legitimacy, and even legality under emerging regulations like the EU Artificial Intelligence Act. If biometric systems consistently fail for some users, they cannot be trusted by any. For this reason, the future of liveness detection depends on building systems designed for inclusivity from the ground up.

Evidence of Disparities in Traditional Systems

The most visible example of biometric bias comes from facial recognition. The Gender Shades project (Buolamwini & Gebru, 2018) demonstrated intersectional disparities in commercial systems, with error rates far higher for darker-skinned women than for lighter-skinned men. NIST’s Face Recognition Vendor Test confirmed these findings at scale, reporting significant demographic effects across algorithms (Grother, Ngan, & Hanaoka, 2019).

These disparities extend beyond face recognition. Voice-based systems often struggle with accented or non-native speech. Even behavioral biometrics, such as keystroke dynamics, can produce skewed results depending on age or ability. The result is a landscape in which “one-size-fits-all” solutions systematically underserve large portions of the population.

Psychophysiology as a More Inclusive Approach

Psychophysiology offers a path forward. Unlike static biometrics, which rely on surface-level features like skin tone or geometry, psychophysiological signals are generated internally by the nervous system. Heartbeats, pupil reflexes, and microexpressions occur across populations, regardless of demographic characteristics.

That said, psychophysiological signals are not immune to variability. For example, rPPG (remote photoplethysmography) can be affected by skin reflectance, leading to weaker signals for darker skin tones under poor lighting (Sun & Thakor, 2016). But because Moveris integrates multiple modalities—heart rate, ocular responses, facial micro-movements—no single source of variability determines outcomes. This redundancy ensures that if one channel is less reliable for a given user, others compensate.

By designing around diversity rather than treating it as an afterthought, psychophysiology provides a framework for equitable measurement.

The Moveris Bias-Reduction Framework

Moveris operationalizes inclusivity through three key strategies:

Calibration: Signal processing algorithms are optimized across lighting conditions, camera qualities, and individual differences, ensuring stable performance in varied environments.

Multimodal Redundancy: Multiple signals—heart rate variability, pupil dynamics, facial micro-movements—are fused into a single coherence framework. This reduces reliance on any one channel and improves robustness across demographics.

Demographic Validation: Moveris conducts validation studies with stratified sampling, testing system accuracy across gender identities, ethnicities, and age groups. By reporting subgroup performance metrics, Moveris ensures transparency and compliance with fairness standards.

Ethical, Legal, and Compliance Considerations

Inclusive psychophysiology is not only a scientific imperative but also a legal one. The EU AI Act places biometric systems in high-risk categories, requiring explicit fairness validation. NIST guidelines call for demographic testing and transparency in performance reporting. Organizations deploying liveness detection must be prepared to demonstrate not just accuracy but equity.

Moveris addresses these requirements by embedding inclusivity into its design process. Transparency reports, subgroup testing, and explainable coherence metrics provide regulators and clients with evidence of fairness. By aligning with ethical and legal standards, Moveris positions psychophysiology as a responsible foundation for digital trust.

Implications for Industry Adoption

Equity is not just a regulatory hurdle—it is a market advantage. Systems that fail certain populations will eventually lose credibility and adoption. Conversely, systems that work for everyone can build universal trust. By ensuring fairness across demographics, Moveris expands the potential user base for fintech, media, and security applications, making psychophysiology not only more ethical but also more commercially viable.

Conclusion

Bias is the Achilles’ heel of traditional biometrics. Systems that work well for some but fail for others cannot be the foundation of digital trust. Psychophysiology, with its emphasis on dynamic, multimodal signals, provides a more inclusive alternative. By embedding calibration, redundancy, and demographic validation into its framework, Moveris ensures that liveness detection is reliable and fair for all users.

In the post-deepfake era, the ability to measure life must extend to everyone. Inclusivity is not optional—it is the future of trust.

References (Selected for this Paper)

• Buolamwini, J., & Gebru, T. (2018). Gender shades: Intersectional accuracy disparities in commercial gender classification. Proceedings of the Conference on Fairness, Accountability, and Transparency, 77–91.
  
  
• Cavallaro, L., et al. (2022). On measuring fairness in machine learning for biometrics. Pattern Recognition Letters, 156, 1–7.
  
  
• Grother, P., Ngan, M., & Hanaoka, K. (2019). Face recognition vendor test (FRVT) Part 3: Demographic effects. NISTIR 8280.
  
  
• Raji, I. D., & Buolamwini, J. (2019). Actionable auditing: Investigating the impact of publicly naming biased performance results of commercial AI products. Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society, 429–435.
  
  
• Sun, Y., & Thakor, N. (2016). Photoplethysmography revisited: From contact to noncontact, from point to imaging. IEEE Transactions on Biomedical Engineering, 63(3), 463–477.
  
  

‍