Mixing Humans Into The AI-Powered IDV Workflow Loop

In the world of Digital Identity Verification, convergence is when AI-tech and human intelligence converge to keep the cybercrooks at bay, says Jumio CTO Labhesh Patel. It’s also the focus of Jumio’s new Montreal-based AI Labs where, he says, the world’s brightest minds are converging to push the boundaries of machine and human potential in authenticating identities in a digital world.

Oh, Canada? You might want to consider the country one of the next great hubs of authentication technology – with some biometrics thrown in to boot.

News came at the end of last month that Jumio, which provides identity as a service to companies across verticals such as travel, retail and finance, has launched Jumio AI Labs in Montreal. That lab will focus on the creation and deployment of machine learning and deep learning technologies, as we spotlighted late last month.

One question that might be top of mind: Why would a firm with more than 100 million authenticated transactions under its belt need to boost its AI efforts? Fraud detection is an ever-evolving pursuit, especially as commerce is going increasingly mobile and global. The bad guys are ever adept at cobbling together IDs across any number of data sources across the Dark Web. And without the aid of technology fast and smart enough to get its arms around the deluge of data that comes at firms day in and day out, it is increasingly difficult to be certain that someone on the other side of the transaction is who they say they are.

The new lab ties in with efforts that are already in place in Vienna, noted the company in an announcement on the Montreal location. The activities at the new lab will be focused on creation and deployment of machine learning and deep learning technologies. Day-to-day activities will span identity verification, data extraction, fraud detection and risk scoring.

In an interview with PYMNTS’ Karen Webster, CTO and Chief Scientist Labhesh Patel said that the firm has chosen Montreal due to that location’s emergence as a notable tech hub. The city is known for its academic powerhouses in the field, such as Yoshua Bengio, professor at Université de Montréal.

He also noted that immigration policy is progressive enough so that the firm, and any number of entities involved in high-tech and artificial intelligence (AI) endeavors can find the talent they need and hire them with relative ease.

The stage is set, then, said Patel, for advancements in AI to continue. “We have the entire workflow for ID verification mapped out,” he said. For Jumio, the Montreal location will help examine every part of the ID verification workflow to determine where AI and machine learning would provide additional benefits. The approach is, and will be, governed by a simple philosophy, said Patel.

The Philosophy Governs the Workflow

“My philosophy has been ‘do not use any tool unless it makes a difference in the workflow’,” he told Webster.

Drilling down into that workflow, Patel said that in order to make sure that ID verification is optimal, the company enlists an integrated “human in the loop” flow that is partly done by machines and partly by humans.

“We want to make sure that there are two layers where all the good actors get convergence in the ID verification workflow and all the bad actors get shut down,” he told Webster.

There is certainly enough data in hand, he noted, across the hundreds of millions of IDs already verified where machines can cull those bad actors through initial scans.

“There are a few things happening. One is that just by the sheer volume of the IDs that are being processed,” he said, “we can see if the ID is being used in a second geographical location … or for a second customer. And even before looking at that ID, we can come up with a risk score.”

But upon looking at identification and in getting more granular with detail — especially in onboarding activities — Patel said, the coordination between humans, machine learning and AI algorithms can cut through inefficiencies in optical character recognition (OCR) technologies.

He stated that traditional OCR is used to white backgrounds and black characters. “That model really does not work directly when we are trying to deal with plastic ID,” he said, as the IDs or selfies have been pictured with all kinds of “noise” tied to less-than-optimal-quality cameras.

Here, again, experience counts, said the executive, as Jumio has dealt with thousands of types of IDs, linking them to customer selfies and making sure they match through machine learning — “something that we are already doing.”

Though it is several years out before mass adoption of digital IDs takes place, merchants need to make sure the entire scope of IDs is covered, said Patel, and desire a single uniform interface. Jumio can act as the gateway to making sure those IDs are legitimate, he said.

But beyond the vagaries of identification across tangible documents, he noted that even as verification is key, “the other part of the story is finding fraud,” he said. “We know the kinds of mistake that fraudsters make.” By way of anecdote, he recounted recent incidents in the UK where a number of IDs misspelled the United Kingdom — and yet were legitimate. Real fraudsters, he said, would never make such a glaring error, as they spend hundreds of dollars to get their hands on the data that they use to ply their trade.

As the Montreal location ramps toward critical mass — now with roughly two or three people on staff, with offers out to another eight — 10 to 12 positions will be filled by the end of the year. The tally should double by the end of next year, he said. As he stated, “Montreal’s ecosystem is just buzzing with machine learning and AI energy.”