AI is frequently used to power biometric analysis. Biometrics are measurements relating to human characteristics that are frequently used to enable access to systems and services, for example fingerprint, voice, and facial biometrics as used by smartphones to authenticate users.
Biometric verification using AI
AI can be highly effective at matching facial biometrics. For the purposes of KYC compliance and fraud prevention, businesses often use a document and biometric based solution. A user provides the business with an image of a photo ID to be analyzed for authenticity. They also provide a selfie taken at the moment of onboarding. AI is used to match the facial biometrics between the photo ID and selfie to provide the business with assurance that the identity document was submitted by its rightful owner, and not by a fraudster who may have illegally obtained the document.
AI is particularly well suited to this task. A 2014 study found that even trained passport officers averaged a 14% error rate when matching faces — but the figure is likely to be even higher for untrained individuals conducting this examination for businesses.
However, AI is not infallible — and should be trained ethically and monitored for algorithmic bias. Algorithmic bias refers to systematic and repeatable errors that result in unfair outcomes, such as discriminating against or providing undue advantage to one group of people over another. Algorithmic bias in facial recognition can mean that the biometrics demographic groups are not accurately analyzed, leading to false rejections or false matches during the identity verification processes. At Onfido, we use balanced data sets to train our AI models, and monitor results — you can learn more about how we mitigate AI bias here.
AI for document analysis
As part of identity verification processes, AI can also be used to analyze identity documents to help businesses meet their KYC requirements.
Document verification is a high assurance method of verifying user identity compared to validating user supplied data against trusted sources. If processed manually by human agents it can introduce friction by increasing turnaround times — the time taken from receiving a capture of a document, analysis being completed, and action taken by the business to either reject or accept the applicant. AI can automate this analysis and complete accurate document analysis in a matter of seconds.
How does Onfido use AI?
At Onfido, we use AI to provide automated digital identity verification for 800+ businesses worldwide. Our proprietary AI, Atlas™, has been built over 10 years by a dedicated team of hundreds of researchers and engineers to make our analysis fair, fast and accurate. Atlas’ analysis helps businesses acquire more customers, navigate global compliance, improve operational efficiency, and prevent identity fraud. You can learn more about ‘what is identity fraud’ in our blog.
What makes Atlas AI fair, fast, and accurate?
- Fair. Atlas has been trained on balanced datasets that are representative of real-world variations in race and ethnicity to mitigate AI bias.
- Fast. Atlas processes 95% of biometric verifications in under 10 seconds – a year-over-year improvement of 54%.
- Accurate. Atlas is built in-house using a unique micro-model architecture that combines over 10,000 models trained to detect specific fraud markers.