Two weeks ago, Intel announced the launch of its Saffron Anti-Money Laundering (AML) Advisor service, based on software it acquired with Saffron, and running on top of its Xeon CPU hardware. Promising an unsupervised machine-learning capability, Intel has signed the Bank of New Zealand (BNZ) as the first customer for the system, which will hunt for signs of fraud and money laundering, in a notable win for Intel.
The key to the new approach is Associative Memory, which Intel says will find and explain ‘multidimensional patterns’ inside bank or insurance data. That data is often rather unstructured, which has been the stumbling block for automated systems in the past. Thanks to the new approach, Intel says that the Saffron AML system can unify that unstructured data with other data from enterprise systems – to help carry out its task.
This would allow someone like BNZ to combine multiple distinct sources, adding things like customer email correspondence, support ticket information, or even data from government agencies, to its own banking and investment records – without having to rely on an army of programmers to ensure that the system runs slowly.
That’s the promise, at least, and a lot of its sounds like a solid pursuit for enterprises. However, we have no idea on pricing and uptake, or any method of running some form of benchmark to see how the new system might compare to alternative methods. We plan on keeping an ear open to hear of Saffron’s success in this space, but this appears to be an example of how an AI-based application can significantly disrupt established enterprises.
VP and GM of Intel’s Saffron AI Group, Gayle Sheppard, said “Saffron’s mission is to minimize the time and effort it takes to reach confident decisions. The amount of data that banks and insurers collect is growing at massive scale, doubling every two years. While the quantity of data is growing, so are the types and sources of data, which means today much of the data isn’t queried for insights because it’s simply not accessible with traditional tools at scale. Investigators and analysts will depend on transparent AI solutions to meet the ever-growing demands of consistency and efficiency from a business, regulatory and compliance perspective.”
The other key promise is that the system makes it easier for enterprises to comply with regulations, which are becoming more of a priority for businesses that handle data. The terms ‘Safe Harbor’ and ‘GPDR’ will provoke panic attacks for some IT departments, as those deadlines and workloads loom – and a wave of potential lawsuits and liabilities should be panicking legal departments too.
Intel’s launch underplayed the significance of the Associative Memory that powers the system, which mimics the way that a human brain associates objects and patterns, in order to process huge data sets. Intel says that it does this in a transparent manner, which is “paving the way for ‘white box AI’ in enterprise applications.” It claims that early results indicate the ability to catch money launderers with unprecedented speed and efficiency.
Speaking to EE Times, Saffron Director of Financial Industry Solutions, Elizabeth Shriver-Procell, explained that Associative Memory is a different branch of AI, one that excels at examining diverse databases.
“Take the example of a banking customer named Mary. Mary goes to London every other week and shops at Liberty store. John, who lives in a different country goes to London about the same time when Mary is there, and does something entirely different. Is there any relationship between the two? What are the commonalities between the two? Can we take a look at their IP addresses? Do we find any similarities in their log-in patterns? Is there anything that shows if any nefarious activities are going on there?”
Importantly, it can be examined and explained, unlike many of the machine-learning algorithms used in object recognition, which perform tasks predictably but use mathematical models that are very hard or impossible to explain.
Procell explained that ‘white box AI’ refers to the new explainable approach, which is a change from the ‘black box’ approach to fraud detection that financial institutions had to deal with in the past – where they had to trust the software housed inside boxes supplied by vendors, noting that in the past. “They can’t see what’s inside the black box, and they can’t tell if it was working properly.”
Consequently, if what Intel is saying is true, the white box approach will mean that a customer can go to a regulator with confidence in their ability to explain a decision made by an AI-based system – crucial in high-stakes games like banking and insurance. The significance of this ability seems to be underplayed, in that it is an AI-based application that can be explained by its operators – who should be able to follow its decision-making process all the way through to its conclusion, unlike a cat-spotting neural network.
Intel cites UN estimates that between 2-5% of global GDP is laundered each year, some $800bn to $2tn. Part of this was due to some 15.4m consumers that were victims of identity theft and fraud, which alone is pegged at $16bn in losses.
Intel adds that companies like banks often have over 50 applications that require access to the same sets of consumer data, and that replicating the data is costly and increases their risk of attack or compromise. Moving that data around a cloud is already an expensive process, but the Associative Memory techniques offered by Saffron should allow a bank to more efficiently manage that data – sharing it across applications, and in turn boosting the insight of each of those applications.
“We’re excited to be working with Intel Saffron on truly bleeding edge technology that will enable us to understand our customers far better than we ever have before, and help them make smarter decisions,” said BNZ’s Director of Products and Technology, David Bullock. “By staying at the forefront of AI, we can help ensure we have access to the latest, innovative technologies that enhance our business.”