Artificial intelligence (AI) has the potential to change lives, literally – aiding in the detection of diseases, for example – while in commerce, it has the potential to stop bad transactions.
We’re getting there, but bringing AI into everyday life is an evolutionary process that differs markedly from the breathless portrayals in the movies, where the rise of sentient computers and robots bodes well, or ill, for humanity, depending on what you’re watching.
In an interview with PYMNTS’ Karen Webster, Rajat Taneja, EVP of technology and operations at Visa, helped separate AI hype from AI reality, laying out what actually goes into the nuts and bolts of building an AI-powered platform. The roadmap is, by necessity, focused on narrow applications first, with an eye on expanding deeply and widely (and more broadly) across several settings.
Deep, Narrow and Focused
The perception of what AI can do – and actually is doing – in today’s world does not necessarily dovetail with reality, said Taneja.
“In technology, terms are often co-opted, and are used in ways that lead to misinterpretation and confusion,” he noted. “They are used in ways that do not totally reflect reality.”
That statement, Taneja said, applies to AI today. We’re a far cry from deploying computers that can think for themselves. In fact, the AI we see today, he explained, is best described as “narrow intelligence,” which means it is very good at performing deep but singly focused and specific tasks, as mandated by its creator – the programmer.
There may be an inflection point in the wings, but we are far from seeing what might be termed as “general AI” take root, which is the computers that ostensibly make decisions truly on their own, with a configuration of moving and ancillary parts that demonstrate motor responses such as would be seen with human beings.
The Building Blocks
As Taneja described it, “the core asset needed to drive AI breakthroughs,” especially with payments, lies with data, adding that “without data, there is no way for any algorithm to find a pattern or make an inference.” The machine cannot detect, learn or take action without having a data platform at its foundation.
Nowadays, the sheer amount of data that can be stored and harnessed for AI is voluminous – enough so that Visa has been investing in the space over the past five years, and not in an insignificant way. The company has spent roughly $500 million through the past five years on data and AI efforts. Along the way, the payments giant has hired several thousand engineers into its ranks, representing a significant portion of its employee base.
Beyond the data lies the second piece of the AI puzzle, as Taneja related: the machine learning platform, where the models are built and “trained” on the data – and it is here that the algorithms learn to detect patterns and behaviors.
(Along the way there is a “deep obligation,” he said, to protect the data and promote data privacy, beyond even the confines and mandates of GDPR and other regulations.)
Beyond the data platform and the machine learning platform, he said there is a third layer known as the AI platform (separate from the other two), which he described as the “online platform where we can apply the algorithms in real time at Visa scale. In the payments realm, all three layers operating, and cooperating, are ultimately tied to authorization, risk scoring and even managing cyber threats.
The Cake Model Is No Cake Walk
Taneja offered a visualization to help illustrate how all of this works together toward embracing actual use cases in payments (and, of course, beyond): The first three layers are horizontal and foundational, he explained, and span an enterprise. “Once you have the three layers, then applications are verticals on it, so you can have multiple applications,” he told Webster. “Think of it as a cake with candles.”
With the three layers working in tandem, the system “begins to infer and detect patterns by itself as opposed to somebody programming a set of rules. This is called ‘deep’ learning, where there are layers and layers and layers of these neural networks that simulate the human brain,” Taneja explained. “Each layer learns from the prior layer by taking its output and performing computations on it. Then its output [is sent] to the next layer.”
The use cases (or, as described earlier, candles) are still in early innings, he noted, and hundreds of Visa AI-powered applications are thus far mainly tied to fraud and risk management – and can even detect and address operational issues (such as those tied to network disruptions) before they become problematic.
Why It Takes a Village
The larger promise with AI remains inherent in the fact that data is able to find patterns that users are not even asking it to find – and it finds meaning where human observers may not yet have thought to look.
Asked how far away we are from tapping into the true power and potential of AI, Taneja said the process is “a constant evolution,” where new concepts and use cases will cascade, especially as his firm and its clients work together.
He noted that Visa opened its technology to clients with standard APIs and interfaces. Data is protected by several layers of encryption, and additional efforts have focused on reducing latency so it is processed even more securely.
Taneja pointed toward the constant evolution of a “parallel industry” as a roadmap of sorts for AI itself: When an iPhone comes out, for example, new apps are continually built, the App Store becomes a bit more densely populated, “and then the platform keeps evolving to support those [apps], and it’s a virtuous cycle,” he told Webster.
“AI is not just about data science,” Taneja continued. “It’s about data engineering. It’s about infrastructure. It’s about a network, it’s about security – and it’s about application architecture. So the underpinnings for AI to really blossom require the whole village to take part. It truly is a massive, coordinated effort across multiple disciplines.”