As Mastercard’s chief innovation officer, Ken Moore has team members working around the world. So, it wasn’t unusual for Moore, based in Dublin, Ireland, to get a call from an associate in Singapore. What was unusual was the colleague’s questions:
Was Moore really stranded someplace because he had left his corporate card behind? And did he really need the colleague to get some money to him so he could pay for a taxi?
In other words, the associate was asking, was this series of WhatsApp messages really from you?.
It was a deepfake attack — Moore didn’t need money nor a lift and he hadn’t sent any pleas. But there was believability. Messages to the team member included vague references to Moore having seen the Singapore team recently, along with pictures that could conceivably have come from Moore. Next was some generic chat about how Moore’s year was going, all of which could have been easily cribbed from social media posts and other internet sources.
Deepfakes have been possible for some time; similar techniques have been used in the movies for years.
“But while deepfakes have long existed, generative artificial intelligence gives bad actors many more tools for making them more convincing and for producing deepfakes at scale,” says Moore. GenAI can crank out attempts like those using Moore’s name in volume and even if only a small percentage of recipients fall for the ruse, the bad guys win.
GenAI Makes Payments Fraud Easier and Potentially More Profitable
Both large commercial payments and small consumer payments can be vulnerable.
While corporate accounts typically have safeguards, only a few days after The Financial Brand interviewed Moore, CNN reported that a Hong Kong finance employee was duped into transferring $25.6 million. An early contact by impersonators of company employees had been dismissed as a phishing attack by the employee, according to CNN, but a video call made it appear that real company employees needed the transfer made. The images turned out to be deepfakes.
Ruses backed by GenAI to fool consumers won’t yield as much, but in volume they can be lucrative. Moore adds that many people won’t think hard about a request involving $5 to $10.
This is one of the many facets of generative artificial intelligence that Moore’s team has been working on since GenAI became widely available.
Moore believes that shared standards for the use of GenAI will be essential to maintain trust in payments and other financial technology. Work is advancing in different parts of the world. Often regulatory reactions to such threats take time to get moving, but Moore says some jurisdictions are gaining momentum. Ideally, standards will cross borders in some fashion, Moore says, because the fraudulent activity often does so.
“We are very much part of those conversations,” says Moore. “We believe that regulation in this space promotes trust. We think it promotes security. So we are advocates.”
Tech tools are already available. Moore points to software that can detect “liveness,” or the lack of it, in video images. For example, it is possible to detect the movement of blood beneath the skin, to confirm real images. Moore says there is also software for detecting use of language that is in character and consistent with past communication.
“It will take a collective response,” says Moore. “The only real way to fight these advancements in technology is with more technology on the other side, coupled with customer education.”
If the financial services industry doesn’t nail this, the impact of issues like frauds paid through Zelle — a persistent black eye for the P2P service — will be dwarfed.
It’s just one part of how Mastercard’s innovation labs are tackling AI, but it’s a critical one. Decision Intelligence Pro, Mastercard technology relying on GenAI, was announced in early February for release later in the year. The company says the tech will increase banks’ fraud detection capability by 20% and even higher in some situations.
More generally, the trick, going forward, will be banking’s long-running challenge in anti-fraud efforts: Building nets fine enough to trap the real frauds without generating too many false positives.
Read more: AI Arms Race: Banks and Fraudsters Battle for the Upper Hand
Understanding How Facets of AI Can Be Combined
The impression among the general public, thanks to the mainstream media, is that there are things done with “traditional” AI and then newer things done with “new” generative AI.
Where’s the actual dividing line? “There isn’t one, really,” says Moore. “They overlap.”
Generative AI is strong for interpreting large sets of data, especially unstructured data, according to Moore. It also excels at conversational engagement and producing new content based on its training data.
“But it’s not as good as other forms of AI at predicting things,” Moore continues. “So, most of the applications being rolled out, there’s an amalgamation of more traditional AI together with generative AI. It’s not one or the other — it’s using the right tools to solve a given business problem.”
Combining the two facets of AI can produce superior results, going from 80% on target to 95%, according to Moore. He explains this by analogy:.
Using traditional AI techniques is like asking a librarian to help you find books that could hold the answers you are looking for. When the librarian plunks a stack of books in front of you, then you have to do the digging.
Combining that step with GenAI, says Moore, is the equivalent of having a study buddy at your library table. The combination not only gets you the books, but also finds the answers within them.
Read more: How to Mitigate Risk and Consumer Fear As Banks Adopt GenAI
How Gen AI Will Shape Commerce and Payments
Many people who fear the impact of artificial intelligence on their lives don’t understand that a good deal of their lives have already been touched by the technology, sometimes for years. Numerous aspects of a consumer’s Amazon opening page are personalized by AI. Likewise, the viewing recommendations presented on many streaming services.
Gen AI is only one of many aspects of what Moore’s innovation teams have been developing, and some have nothing to do with AI at all.
“But there was a step change 16 months ago, with the broader advent of large language models and the massive spike in investment. There was also convergence of other technologies that have allowed large language models to do things that they couldn’t have done two or three years ago.”
— Ken Moore, Mastercard
New products will change the direction of ecommerce. Some of these are beginning to debut now; some will evolve over time.
Shopping “copilots,” personalized to consumers’ preferences, are one example. Late last year Mastercard released “Shopping Muse,” a GenAI tool that guides consumers through retailers’ digital catalogs. Shopping Muse can turn everyday language about goods into product recommendations along with complimentary recommendations.
The software can also help shoppers who can’t quite say what they want — ever try to order a part when you don’t know what the part is called?
This is just early days for such technology, according to Moore. In time, there will services that will go beyond a single retailer. This will include the ability to hunt for products a consumer is looking for while simultaneously examining what people have said about this or that product. Items that have garnered bad consumer ratings wouldn’t even show up in the personalized shopping copilot’s results.
Read more about Artificial Intelligence in Banking:
- Banks: Be Prepared to Explain ‘Explainability’ of AI
- The Growth Opportunity Where Artificial Intelligence and Embedded Finance Meet
- Lessons from Ally’s Experiment with GenAI Marketing: Commit, But Carefully
A Taste of What May (or May Not) Be Next
If there is anything fresher than GenAI large language models, it’s a newer wrinkle, “large action models.” An example of this is a device in development called “rabbit,” from a firm of the same name. A sensation at the Consumer Electronic Show, the rabbitOS and the rabbit device are intended to enable a consumer to interact with all of their digital devices using spoken, plain-language commands in lieu of using app menus and commands.
“Interaction with rabbit will be like interacting with a friend, who learns by observing how you tackle a task and listening to your explanations in plain language,” according to Jesse Lyu, founder and CEO of rabbit, in a 2023 funding announcement. The point is that the device can accomplish one-time tasks as well as tasks that get repeated periodically. Some tasks would involve working across multiple apps, as a person might.
Moore says there hasn’t been much seen on this front yet and it’s hard to tell where the technology could go. He’s a bit skeptical about getting people to carry another gadget when they live off their phones.
“Will we really carry another device?” asks Moore. “I think it would have to be very compelling for users to want to do that.”
The functionality might evolve such that large action models morph into parts of large language models, he suggests. Then it could become part of existing devices, much as GPS became a feature on many phones.
“I’d want to see it run for a few more months because I could say much with absolute conviction,” says Moore. “Months” is testimony to the pace of change today.
In early February Mastercard released a special edition of its “Signals” interactive white papers, concerning artificial intelligence issues.