Machine Learning Image Recognition
Machine Learning Image Recognition : London Metropolitan Police Commissioner, Cressida Dick, has been pushing back against critics of facial recognition claiming that these arguments are highly inaccurate and ill-informed. Commissioner Dick makes a responsible argument for the proportionate use of technology to prevent and resolve criminal investigations.
Facial recognition has proven to be a game changer in the areas of identity theft, fraud and money laundering. The UNODC estimated in 2009 that up 5% of global GDP is money laundered annually, of which less than 1% was getting seized or frozen. The next generation of online identity verification (IDV) companies are attempting to address this 99% failure rate by making it more difficult for offenders to conceal their identity.
Coming out of Europe, a document ID and biometric verification platform called Onfido is using facial biometrics and machine learning to fight identity theft and provide reliable Anti-Money Laundering (AML) and Know Your Customer (KYC) solutions to their clients. As with all EU projects, privacy regulation is central to the design of all software applications following the introduction of the General Data Protection Regulations (GDPR) to ensure that the civil liberties concerns are addressed to reduce government overreach.
Managing Cybersecurity in a Rapidly Evolving Landscape
Onfido’s Global Head of Fraud, Michael Van Gestel, told Rebellion Research how he always has a copy of the relevant compliance requirements on hand whenever he is engaged in product development. This is to ensure that the company develops responsible next generation technology that strengthens IDV into the future.
Van Gestel also points out that it is important to their clients that they comply with regulations and help to fight fraud. Onfido aims to balance both of those interests without sacrificing their integrity. From the point of view of the company’s top detective, compliance is essential to designing responsible solutions that the public and their clients can trust.
Van Gestel is an experienced document verification expert, including 12 years in the Royal Marechaussee. Over that time he has dedicated his work to making life harder for the bad guys, including partnering up with INTERPOL from time to time.
The hindrance of criminals is something he remains very passionate about as he tries to improve systems that assist real people to make informed decisions in real-time. He points out that identity theft is the fastest growing crime in the US and accounts for over half of all fraud in the UK. Onfido works to combat this in accordance with three pillars of their solution:
- By leveraging Biometrics
- Using a hybrid AI approach
- Knowing how to handle 3D documents
AI Robot Security – Making the US a Safer Place
While biometrics is an exciting new area of technology research, Onfido’s ‘Hybrid’ AI approach checks a person’s uploaded image against the presented identity document. It merges human expertise and machine power to deliver a fast solution, turning around decisions in as little as 15 seconds. Any users who can’t be processed automatically are passed to one of Onfido’s trained human experts for review.
This reduces the complexity of what is being asked of Onfido’s human analysts by doing most of the tedious work for them. The merit of this approach is that it enhances the status quo, of humans checking documents against the image presented, by supplementing it with a software tool that is proven to be better than humans at running these checks at scale in real time. The process detects fraudulent documents and spoofing attempts, while providing legitimate users a fast and frictionless onboarding experience.
The hybrid AI approach acknowledges that humans in the loop are still necessary to help improve machine learning models as fraud keeps evolving. It also shows that technology can be used to vastly improve current processes and make user experience more efficient.
As with the development of safety protocols in industry, the development of responsible algorithms will become its own growth area. Onfido is taking this challenge head on, working with the UK’s Information Commissioner’s Office (ICO) to identify and mitigate algorithmic bias in machine learning models used for remote biometric identification.
Where Van Gestel’s expertise comes into his own is on the third pillar, 3D document security. Most major identity documents can have over 40 different security details embedded to ensure that it is not a counterfeit or fraudulently obtained genuine (FOG) document.
Unfortunately, these are often reliant on in-person checks and only around five of these are easily verifiable from 2D document scans. He works with the Onfido development team to find new ways of increasing the number of security details that are observable to ensure documents can be reliably verified as genuine official documents.
Van Gestel is pragmatic about the complaints against facial recognition when asked about the pushback that he has seen. “There has been a lot of miscommunication around facial similarity. From my point of view, it is important to be able to match the person to the document that they are presenting for the purposes of identity verification.
Dawn of a New Age of Global Crime
Obviously, there are different ways to use facial biometrics. We are specific with how we make these verifications, comparing against the face of the person presenting an ID but also to previously seen faces by each individual customer. However, the use of this technology can be quite scary in the wrong situation. China is where it gets scary. Same with some private sector companies that are covertly working in this space. We are here to help catch the bad guys. It is an ongoing cat and mouse game. It’s why I am still doing this for 20 years. We will never win this game but it is up to us to make it harder for the fraudsters that are just looking for the easiest way to commit their crime.”
It is estimated that 1 in 60 transactions online are fraudulent. This is a significant level of economic activity that requires IDV to ensure that the people involved can be identified and tracked. Van Gestel and Onfido’s CEO, Husyan Kassai, have both pointed to their Fraud Index as a tool for customers to identify countries where their experts have found high levels of document fraud, which they are ideally positioned to mitigate. Identifying risk in emerging markets is a key consideration for expansion in these growing economies. By providing technological support and expertise that reliably reduces these risks, Onfido is demonstrating the positive power of biometric similarity analysis in a responsible way.
The use of computers to run machine learning algorithms to process big data efficiently to fight crime, is the modern equivalent of the industrial revolution to improve the standard of living across society. To point to the failings of the technology is to ignore the multitude of benefits that it can achieve. Onfido is demonstrating a strong case for the use of facial biometrics in identity verification and authentication as it reduces risk of human error. In a world where criminals are becoming more technologically savvy, companies like Onfido are working to protect the general public from fraud and identity theft.
At the end of the day, technology is not inherently good or bad. As suggested by Commissioner Dick, much of the messaging around facial recognition overlooks the many ways it can be safely implemented to achieve positive results. The application of sensible regulation and the development of compliant software helps to grow trust in the service providers. The use of machine learning and facial biometric analysis can be a net positive for society.
Through the identification and mitigation of risk, many concerns can be addressed and often reduced or eliminated. While this may not be the silver bullet to all problems, it is certainly a step in the right direction for platforms that want to be diligent and innovative in their approach to fraud and AML compliance.
Machine Learning is Important for Fraud Protection
Written by Paul Marrinan & Edited by Alexander Fleiss
Machine Learning Image Recognition
Machine Learning Image Recognition
