Industries become highly regulated where failure has significant consequences. Sectors like healthcare, education, and finance are not subject to the same risk calculation as a retail store. In such industries where the impact is higher, new technologies are harder to integrate, particularly when the output of such technologies is not well understood nor explainable.
The Need for Intelligible AI
Currently, few financial regulators have a strong understanding of the mechanisms of AI or ML or what the implications of their use could be. This lack of understanding and perceived absence of thorough testing remain obstacles to more prolific implementation.
When Microsoft’s AI chatbot Tay was released on Twitter, it was shut down after 16 hours of interacting with other users as its tweets had become racist, demonstrating the potential impact of human bias in training data. If AI systems are to be used in situations where they may have a material impact on people’s lives such as providing someone access to finance, blocking a transaction or closing an account, any such bias is unacceptable.
Financial regulators rightly expect institutions to be able to justify the reasons for decisions and actions. This requirement further enhances the complexity of the situation, as AI systems typically produce results and decisions that are not always easily explained or untangled. Unsurprisingly, this opacity has limited the use of ML and AI in such domains, and has meant that some of the best technologies have remained under-utilized in the fight against money laundering and financial crime.
Global Regulatory Initiatives
The Financial Conduct Authority (FCA) in the UK has been actively engaging with private industry as well as research institutions to investigate potential uses of ML in the detection and prevention of money laundering. In May 2018 they conducted a TechSprint and recently concluded “Data Science Week” which included discussion on how AI is changing the work of regulators.
The Monetary Authority of Singapore (MAS) is developing a guide to promote the responsible and ethical use of AI and data analytics by Financial Institutions (FIs) and, in November 2017, announced a 27 million dollar grant to support this. In September 2018, the Financial Action Task Force (FATF) held a forum with regulators and private industry to discuss the possible uses of new technology in the regulation of cryptocurrencies, identification of individuals and companies, and information sharing. AI and ML were raised during many of the panels. Each of these initiatives is expected to produce further output in the near future.
Balancing Regulation & Innovation
It is the regulators’ role to protect consumers and the stability of the financial markets. Many are realizing that with the explosion of available data and the expansion of regulatory responsibilities, AI may offer one of the only ways to appropriately manage risk. It is expected that over time regulators will adapt to the changing technology landscape. However, this may be problematic in an industry like anti-money laundering (AML) where the challenge is essentially a battle between those trying to launder funds and those trying to detect and prevent them. If the perpetrators of such crimes are quicker to adapt to new tools and technologies to enable their trade, FIs will be critically hamstrung by their response. For its part, industry needs to seek out innovative solutions that also take into account regulators’ concerns. The use of ML in finance is paying dividends. While barriers remain to the use of these technologies in active decision making on people, transactions, or accounts, they can still be used to generate the data and information on which to base decisions. Using AI to collect and collate data may be more palatable initially than models which predict and monitor behavior.
There is a common fear that the new wave of technology in finance will lead to an uptick in financial crime. Critics see the digitization of money and the efficiency of modern payment services as enabling criminal activity—especially when managed by firms who may not have the necessary expertise to conduct proper due diligence and monitoring. The counter- argument is that as technology enables faster payments and exponentially increases the data to be monitored, similarly, technology can and must be used to uncover deeper insights and leverage massive volumes of data in ways human analysts can’t.
Time Is of the Essence
Globally it is estimated that 1-3% of money laundering is detected. It is only possible to conclude that existing tools are failing. In the arms race that is money laundering and financial crime, only one party is exploiting technology to its maximum potential. A critical element in leveling the playing field is to bring the same weaponry to the fight, and in today’s world, that means big data techniques and machine learning.
Executed the right way, technology can align the interests of industry and regulators. Compliance is becoming harder; not only is regulation changing frequently, but the amount of data to monitor is growing exponentially. Traditional tools cannot keep up. Regulators focused on the achievement of important outcomes recognize that industry needs new ways of fulfilling their obligations and that technology may offer those solutions. The right technology will be the key to better compliance, better detection of financial crime, and more efficient processes, meeting the needs of both regulators and financial institutions.
About the Authors
Livia Benisty is Head of Financial Crime at ComplyAdvantage. Previously, Livia was Head of AML for Citibank’s Trade and Treasury business in EMEA, and then Head of AML Risk for Digital Payments based in New York. She joined ComplyAdvantage from BBVA where she was Global Risk and Control Manager for Open Platform, BBVA’s API platform connecting FinTechs to the bank’s core.
Luke Mawbey is VP of Engineering at ComplyAdvantage and oversees the development of their AML data and solutions, including the machine learning team that identifies, collects, and processes data used to detect and prevent money laundering.