Can Artificial Intelligence Better Validate Consumer Identities?

Who we are will increasingly be determined by algorithms. The question of who oversees, verifies and corrects those algorithms, humans - or other algorithms - is the looming issue. JL


Jared Council reports in the Wall Street Journal:

A new AI platform uses machine-learning algorithms to assign an identity score to a person based on hundreds of elements, such as phone numbers and shipping addresses. Startups and banks are using AI to tackle the challenge of verifying users’ identities without asking too many questions, which can affect the customer experience. “The amount of information used to describe identity is only going to expand, so we want to make sure that the solution we develop can handle that expansion without human intervention.” 20% of the identities previously believed to be fake were actually real, while 5% of identities believed to be real were fake.
Experian PLC is testing whether artificial intelligence can better validate consumer identities, a task that’s vital to its core credit-scoring operation and other services.
The Dublin-based company, one of the big three credit-reporting firms, is best known for maintaining and selling access to its roughly 1.2 billion consumer credit-history records.

But part of its business involves drawing on its database to help banks, retailers and others authenticate and better understand who their customers are. These business lines rely on Experian’s ability to accurately identify people, a growing challenge in an era of data breaches and fake, or “synthetic,” identities.
The new AI platform, which the company expects to introduce nextyear, uses machine-learning algorithms to assign an identity score to a person based on hundreds of elements, such as phone numbers and related shipping addresses. Experian is testing the platform in its U.S. financial services and marketing data business units.
“We believe identity is a far more complicated world today than it was 20 years ago, 10 years ago and almost really five years ago,” said Eric Haller, executive vice president and global head of Experian DataLabs, the company’s research-and-development division. “The amount of information used to describe identity is only going to expand, so we want to make sure that the solution we develop can handle that expansion without human intervention.”
Experian rivals Equifax Inc. and TransUnion also use AI. Equifax last year introduced an alert system that uses machine learning to detect potential synthetic identity fraud. TransUnion said it has been using AI for at least three years in its IDVision platform, but its acquisition last year of AI fraud-detection company iovation Inc. significantly expanded its capabilities.
Julie Conroy, a research director at consulting firm Aite Group LLC, said startups and several large banks are using AI to tackle the challenge of verifying users’ identities without asking too many questions, which can affect the customer experience.
“If you’re a bank and someone comes into you giving you their username and password, it’s really tough to figure out if this is a genuine person or if this is a bad guy that bought their data off the dark web,” she said. “And that’s where machine learning can help.”
When a user enters correct information, she said, AI can assess other data signals such as IP addresses, device identifiers and behavioral biometrics such as how quickly people enter information in an online form.
Experian developed the AI platform in its San Diego lab, one of four such sites the company has world-wide.
The company declined to say how many employees worked on the platform, though a spokesman described it as the “single greatest priority” for its San Diego lab over the past 18 months. The platform has been tested with enough clients to get a sense of its performance “against the entire U.S. adult population multiple times” over, he said, declining to give the number of clients.
Experian’s AI platform can be used in its other business units, such as marketing, to help retailers and brands learn more about who saw their ad campaigns and made purchases. The platform does that in part by using advertising identification codes unique to every device that hasn’t opted out of targeted advertising and correlating that to people or households associated with that code.
Mr. Haller said he couldn’t disclose statistics concerning the platform’s performance, such as accuracy or a false positive rate. He said the system helped Experian DataLabs determine that 20% of the identities it tested previously believed to be fake were actually real, while 5% of those identities it believed to be real were in fact fake.