Anti-money laundering (AML) laws remain anchored in their 1970s design despite now being used to respond to modern terrorism financing. In the face of persistent threats from terrorism, ignoring innovation can cost lives. Several new technologies could modernize the AML framework to help law enforcement better target terrorism financing and money laundering, making the system faster, more efficient, and safer. This is the second post in our series about the potential for these technologies.
It is well known that technology has revolutionized industries such as music and movies, but its impact is less well known on other sectors such as recordkeeping. As technology advances it often creates avenues to rethink the way we do things. One area ripe for technological innovation is AML. As we’ve mentioned before, the current AML framework was designed in the 1970s before the internet revolution. As it currently stands, bank compliance officers search records for patterns and trends that regulators say constitute suspicious activity. Those officers report suspicious transactions to government regulators which then determine which transactions should be investigated by law enforcement.
Two recent examples of the superhuman powers of computers, IBM’s Watson and Google’s AlphaGo could provide the foundation for revolutionizing AML monitoring. If we’ve learned anything since the invention of computers, it’s that they are much better than humans at monotonous tasks like sifting and searching through mountains of data.
Reporting Under the Current AML System
As we’ve previously discussed, the number of AML reports skyrocketed after the September 11, 2001 terrorist attacks. Some commentators, including Juan Zarate, former assistant secretary of the Treasury for terrorist financing and financial crimes, have argued that much of this increase is due to concerns about reputational risk and a federal law that grants banks a safe harbor for reporting rather than the quality of internal monitoring systems. This provides a perverse incentive for financial institutions to file reports regardless of whether they believe a transaction is suspicious or not. That in turn leads to over-reporting and the production of massive amounts of data that could make criminal detection more difficult for law enforcement.
If we’ve learned anything since the invention of computers, it’s that they are much better than humans at monotonous tasks like sifting and searching through mountains of data.
The problem with an ever increasing store of data is that someone has to sift through it all. Anyone who has used Ctrl + F to find a word or phrase in a long document knows that computers are much better at sifting through data to find keywords. However, if you’ve ever spelled that keyword incorrectly, you know the limits of consumer software. To turn the suspicious activity report database into something useful, we need to leverage new technologies that combine the speed of computers with the insight of humans. Fortunately, tech companies are hard at work making these technologies a reality.
Two Types of AI for AML
Watson: IBM’s Watson is most famous for defeating Jeopardy! champions Ken Jennings and Brad Rutter, and for leaving Bob Dylan unimpressed with its singing skills. However, the underlying technology behind Watson could revolutionize industries that require searching for answers hidden within mountains of data, including AML databases. Since the AML legal framework gives financial institutions strong incentives to report if they have only the slightest hint of suspicion, millions of reports are filed every year which the filers know are very unlikely to implicate criminal activity. Once filed, the reports must be analyzed for criminal activity taking up massive amounts of time and resources. There must be a better way.
Enter Watson. To win at Jeopardy!, Watson was given access to a massive database of information including the entire text of Wikipedia. Using a combination of natural language processing and advanced hypothesis checking software, Watson was able to determine the answers to questions even in the deliberately complicated style in which Jeopardy! questions are presented. Watson breaks questions down into components to determine what they mean and how to go about answering them. Once it has an idea, Watson passes it through a hypothesis checking program that weighs the evidence for and against it. If the program returns a sufficiently high probability that the idea is correct Watson selects it as its answer.
Treasury currently maintains a database of AML reports submitted by financial institutions along with records of previous AML investigations. If Watson were provided with enough data from these investigations, it could use the same techniques it used to conquer Jeopardy! to find matching patterns in the data much faster than a human. This extra speed is non-trivial in a world where cutting funds to terrorists could mean saving lives. It’s also useful in the detection of cyber threats and in fact Watson is already being used in the cybersecurity industry.
AlphaGo: Google’s AlphaGo could be even more revolutionary. While Watson would rely on previous AML investigations to predict new ones that fit a similar pattern, AlphaGo could adapt itself to changing environments using what’s called an intuitional algorithm. Intuition suggests plausible ideas for what steps to take next without surveying the universe of possible choices. Broadly speaking, it is similar to why when you lose your keys it never occurs to you to ask NASA for a ride to the international space station to look for them. AlphaGo gained its intuition in the game of Go by playing 3 million games and making tweaks to its intuition programming each time. The program then went on to defeat one of the top Go players in the world earlier this year.
What makes AlphaGo so spectacular is that it shows the possibility of general-purpose AI that is able to adapt itself to changing environments.
What makes AlphaGo so spectacular is that it shows the possibility of general-purpose AI that is able to adapt itself to changing environments. An AlphaGo-style program could monitor transactions and patterns in the current AML framework to refer potential cases to law enforcement. If a referral led to a successful prosecution, the program could tweak itself accordingly. In light of the recent incidents with self-driving cars, AML monitoring would only be fully automated once it was determined that the program was more effective than human monitoring. If it’s progress toward becoming the world Go champion is any indication, we wouldn’t need to wait long.
KEYWORDS: ANTI-MONEY LAUNDERING SERIES