It’s hard to find fault with one of the latest tests of artificial intelligence (AI).
Medical diagnostic technology built by Google and a sister company can detect diabetic retinopathy — an eye condition that can cause blindness — with as much accuracy as a human doctor. That matters in places such as India, where the AI tech has been deployed, and where a relatively tiny group of eye doctors struggle to serve a massive population.
As one can imagine, the AI project took time to get to the pilot phase — about three years, in fact. That’s understandable, of course. Science is much more about incremental, steady progress than singular moments of discovery. However, the time involved is not ideal when it comes to AI and payments. The people who run financial institutions (FIs) need faster proof that AI can work for them, relatively quick evidence that can help them figure out the return on investment (ROI) for the tech. The payment and commerce markets can be impatient creatures.
The idea is to provide proof of AI’s value to FIs and other payment players within weeks, instead of months or years, and to do so with the data those organizations have on hand. That includes both labeled and unlabeled stores of information that are run through AI algorithms to help with use cases and problems that vex the people who run those companies.
Perception Versus Reality
That’s easier said than done, according to the PYMNTS “AI Innovation Playbook: Perception Versus Reality In Payments And Banking Services” study, done in collaboration with Brighterion. While artificial intelligence is a phrase that carries more currency with each passing day (and, with each new instance, its real and potential use in the world of payments and commerce, including for anti-fraud measures), the real meaning of AI is not always clear to participants in the digital economy.
As that PYMNTS study has demonstrated, true AI involves unsupervised machine learning (ML) — a computer, software and algorithms that can “think” on their own. AI’s less sophisticated (but much more well-known) cousin machine learning needs supervision to learn, though the technology is certainly capable of high-level fraud prevention and business operational tasks — which is why FIs have adopted ML.
“There is a lot of confusion, and a lot of noise” around AI, Adjaoute said. In fact, true AI systems are used by only 5.5 percent of FIs.
Proof Of Value
Can an AI fast lane get those numbers up? To that end, Brighterion has launched a program called AI Express. Its goal: build “a fully functional [AI] model, customized to your unique business situation,” within five to eight weeks.
As Adjaoute told Webster, the “key word is ‘express.’” He said many companies have much more confusion about AI than expertise, with anxiety about length of AI deployments and ROI among the top concerns. In addition, as PYMNTS research has shown, there exists significant misunderstandings in the world of payments and FIs about what constitutes real AI, and what is merely ML.
“The vision here is that we want more companies to see the benefits of AI,” Adjaoute said.
According to his view, it won’t hurt. The key, he told Webster, is to take as much of the work — and as much of the thinking — out of a company’s hands as possible, at least until they see the effort’s results in five-to-eight weeks’ time.
“They don’t need to think about anything,” he said.
That includes worrying about labeled and unlabeled data, and issues associated with AI that can cause delays and headaches for FIs and other companies that want to move forward with such technology. For instance, data related to anti-money laundering (AML) — a main promise of AI in the payments world is that it can spot subtle patterns of fraud — tends to be unlabeled, Adjaoute said. The ideal results of putting AI in the fast lane is to deliver a quick “proof of value” to executives, and provide a solution that is “scalable and resilient.”
Artificial intelligence suffers from a number of misconceptions that go beyond confusing it with machine learning. Among them, according to Adjaoute, is the view that AI is just about a complex, sophisticated algorithm. He used an example of a new structure needing architects, engineers and other workers to be built, or new software requiring a team to see it through to completion, as to why AI is more than just a mathematical formula. “You have to have a team to build a real-time, scalable solution,” he said, and that’s one of the perceptions of AI that will ideally take root in the minds of people meant to use it.
Every day in this digital era brings massive amounts of newly generated data — data that, in payments, can not only combine with AI to fight fraud, but boost customer service, reduce false transaction declines and perform any number of revenue-enhancing toil. However, AI faces perception and time-to-deployment problems at this early stage of the game, and that’s going to be among the main challenges for the next few years.