To understand how artificial intelligence is fundamentally changing how Jumio verifies online IDs and identities, you must first understand two concepts you’ve likely heard of: machine learning and deep learning.
Definitionally, machine learning are algorithms that parse data, learn from that data, and then apply what they’ve learned to make informed decisions. So rather than hand-coding software routines with a specific set of instructions to accomplish a particular task, the machine is “trained” using large amounts of data and algorithms that give it the ability to learn how to perform the task. They’re only capable of what they’re designed for; nothing more, nothing less.
As it turned out, one of the very best application areas for machine learning was classical computer vision, though it still required a great deal of hand-coding of specific features to get the job done. At Jumio, we use classical computer vision to create ID templates for specific ID types (e.g., a California driver’s license or a French passport). We created hand-coded classifiers like edge detection filters so the program could identify where an image of an ID started and stopped and advanced OCR technologies to recognize characters and letters in different languages. From all those hand-coded classifiers we developed algorithms to make sense of the ID document and “learn” to determine the type (driver’s license, ID card or passport) and country of origin of the ID.
Classical computer vision and traditional machine learning are good, but not mind-blowingly great — especially when the ID is captured in less than ideal circumstances (the picture of the ID is captured with bad lighting or the image is tilted or blurred). There’s a reason classical computer vision and image analysis didn’t come close to rivaling humans until very recently, it was just too prone to error.
This is why deep learning holds a lot more promise.
In practical terms, deep learning is a subset of machine learning.
A deep learning model is designed to continually analyze data with a logic structure similar to how a human would draw conclusions. To achieve this, deep learning uses a layered structure of computing units called artificial neural networks (ANN). The design of an ANN is inspired by the biological neural network of the human brain. This makes for machine intelligence that’s far more capable than that of standard machine learning models.
Andrew Ng, one of the stalwarts of deep learning and one of the leaders of the Google Brain project, shared a great analogy for deep learning with Wired Magazine: “I think AI is akin to building a rocket ship. You need a huge engine and a lot of fuel,” he told Wired journalist Caleb Garling. “If you have a large engine and a tiny amount of fuel, you won’t make it to orbit. If you have a tiny engine and a ton of fuel, you can’t even lift off. To build a rocket you need a huge engine and a lot of fuel.”
“The analogy to deep learning is that the rocket engine is the deep learning models and the fuel is the huge amounts of data we can feed to these algorithms.”
Jumio & Deep Learning
How is Jumio leveraging deep learning?
Jumio has processed more than 100 million IDs over the last few years and this data contains important patterns — patterns often indistinguishable to the human eye. Jumio uses deep learning in three important ways: data extraction, fraud detection, and risk scoring.
- Data Extraction: Jumio uses deep learning to extract key data from ID documents. Jumio’s scale means we can feed our algorithms with lots of data (aka fuel) to not only improve their ability to recognize specific ID documents, but also to know how to extract the data and make sense of it.
- Fraud Detection: Deep learning helps us spot characteristics of fraudulent IDs, such as understanding the unique fonts, pictures, and security features of a specific country-issued ID. If an ID document has been manipulated or changed and does not conform to the pattern, our deep learning algorithms flag it for closer review.
- Risk Scoring. We also leverage deep learning to identify patterns and probability of fraud based on a combination of high risk variables. These insights have allowed us to identify that 1% of the ID documents (with a unique risk profile) account for about 15% of known fraud. We use this scoring to alert our verification experts to pay special attention to high risk IDs.
There are still instances where the machine and deep learning models fail to identify what is wrong with an ID. This is where verification experts can apply their expertise to determine what error occurred and teach our algorithms how to spot the issue in the future. This feedback loop creates a different method of learning where new information is constantly being input to the learning model so that the model can improve.
The Wisdom of the Crowd
Unlike automated solutions, Jumio has always been delivered as a hybrid model — leveraging technologies (including computer vision, machine learning, deep learning) and verification experts. This has proven to extremely important when it comes to face matching.
With Jumio’s Netverify Identity Verification, we go one step beyond ID verification, by verifying that the picture of the selfie and the picture of the person on the government-issued ID actually match. This is pretty important because if I stole your driver’s license and used it to create a bank account, there is no doubt that the ID itself is authentic, but the person behind the account creation is clearly not the person pictured on the ID.
But matching these images is a very difficult technical problem. The ID itself could be 5 or more years old and this could mean the person taking the selfie naturally looks different than the person on the ID. The person’s weight could have changed, the hair color could have been dyed, the person now wears glasses, and even the way the picture was taken makes a comparison exceedingly challenging even for the most seasoned verification experts.
This is another area where deep learning is having a dramatic impact.
We already leverage machine learning to perform “similarity checks” of the two images (the one on the ID and the one pictured in the selfie). Jumio has a large team of verification experts around the globe who are apply their judgment to accept or reject these image pairs and better inform our DL models.
We are also looking at image pairs of identity checks that were rejected and using deep learning to spot additional unique features that will train our algorithms to better spot discrepancies which may signal fraud. Once again, our ability to identify these trends and nuances is based on the amount of fuel we have in our rocket ship.
A Closing Word
The promise of AI, machine learning and deep learning is enormous. But, in order to harness this potential, identity companies have these ingredients in place:
- Big Data. For identity companies, this means capturing government-issued IDs in large volumes to train their algorithms to spot patterns and better detect when an ID has been manipulated or altered in any way.
- Human Review. To better inform the algorithms, there needs to be a continuous feedback loop where every ID (and face matching image pair) is labeled as pass or fail. When only a small fraction of transactions are reviewed by humans (as is often the case with automated solutions), this limits the ability of deep learning.
- Data Scientists. Increasingly, data scientists are in high demand to help build deep learning models. Make sure, your identity verification vendor has made these personnel investments to exploit the potential of artificial intelligence.
- Years of Experience. Companies that are new to the identity verification space are disadvantaged in some pretty material ways when it comes to ML and deep learning. Usually, they have not amassed much data to inform their algorithms. But, just as importantly, their verification experts often don’t have the experience to know how to tag legitimate and fraudulent ID documents and are often only reviewing a handful of online verifications. Leading companies that have long relied on human review are in a much better position to recognize fraudulent IDs and face matches because of their experience and training.
- Strict Privacy Compliance. Jumio treats data privacy compliance as a mandatory requirement of any solution released to the market. Fortunately Jumio’s scale allows for the application of DL models in a fully compliant way which would not be possible for smaller identity verification providers.
We’re at the cutting edge, but have already seen the power and possibility of AI. Increasingly, our customers will reap the benefits from these investments to make our solutions faster, enable us to detect more fraud, and streamline the experience of legitimate customers.