Artificial Intelligence Explained
Using AI for Trade Surveillance and Fraud Detection
Artificial intelligence, machine learning and deep learning have become dominant forces shaping the banking and capital markets sector in recent years. Financial institutions are allocating significant resources on AI-based applications and talent and regulators are paying increased attention to the emergence of AI and its impact on the financial services sector.
In an episode of the Fraud Eats Strategy podcast series, Ozgur “Oz” Vural and I were joined by Murphy & McGonigle partner Howard Kramer to discuss the expanding use of AI to undertake trade surveillance and mitigate fraud.
AI, machine learning and deep learning are subsets of one another. AI is the development of intelligence systems not produced by nature. An algorithm is the simplest form of AI. Machine learning is when algorithms have learned from data and patterns and produce outcomes without the need to reprogram. Machine learning is when algorithms teach themselves without human intervention. A simplistic example of that is when you type an internet search, omit a letter and the search algorithms prompt you with the corrected spelling of your search term. That’s machine learning and there has been widespread use of machine learning across a broad spectrum of activities in financial markets. The smallest subset of the three is deep learning. Deep learning uses simulated neural networks. These are multiple layers of analysis in which algorithms at each level learn and adjust based on the inputs from the prior layer. The logic underlying deep learning is that the more layers you get and the more reprogramming with different algorithms takes place, the more precise an outcome you can expect. Deep learning requires enormous amounts of data and immense computing power. As a result, it is very expensive in comparison to machine learning.
Agencies that regulate the capital markets are primarily focused AI in two ways. FINRA for example has regulatory personnel devoted to understanding how their members are using AI machine learning. That understanding then informs how FINRA incorporates what they have learned about the use of AI machine learning in their market activity monitoring and analysis of specific member activity.
The use of AI and machine learning solutions and applications to monitor market abuse, perform trade surveillance and monitoring traits has grown significantly. There is wide recognition of the need to transition away from traditional rule-based systems to a risk-based surveillance modeling with AI applications. Some compliance professionals have expressed frustration with the “black box approach” surrounding the use of AI in which the logic underlying the algorithms in use is not transparent to end uses. It is not enough to make use of AI and machine learning if how it is being utilized and the logic underlying the algorithms cannot be readily explained to a non-technical audience. For the use of AI machine learning to continue to gain widespread acceptance amongst financial institutions and their regulators, it needs to be explainable, auditable and traceable.
As AI is built into market surveillance operations, it becomes more difficult to discern what is proper behavior and what is improper or abusive behavior. The widespread use of machine learning creates an added category of activity that must be monitored that is a byproduct of the use of AI itself. For example, if recursive algorithms effectively function based on inputs, and there are multiple layers of these algorithms and they produce a result, that has the potential to be disruptive to the market or lead to a result that is unintended. Given this possibility, use of AI and machine learning requires a separate level of self-monitoring looking for market disruptions and unexpected consequences. Further complicating the use of AI is the need to train compliance personnel on the AI and machine learning in use, the logic underlying it and the possibility for market disruption and unintended consequences. Compliance people need to be sufficiently well versed in the machine learning that is in use at their institution to be able to spot issues that may hold potential for harm to the institution or the markets themselves. As is so often the case with compliance officers, there need to be people with compliance domain expertise who are able to translate machine learning output and make it understandable for leadership and bank regulators.
Equally important for the use of AI in monitoring or fraud detection to be readily explainable is the fact that trade surveillance and fraud detection can give rise to litigation. This will eventually require the entire AI machine learning ecosystem to be examined under a critical lens. The inability to explain and defend the underlying logic in the use of AI could undermine the financial institution’s position in a legal dispute. Indeed, one of the underpinnings of trying to prove a case of fraud or market manipulation is showing intent and with these new forms of AI, it may be more difficult to prove intent. It is something that both regulators and compliance personnel are going to have to work through in terms of overseeing potential market abuses.
We are at an exciting time in compliance and fraud investigations. Advances in technology make it possible to examine every single trade and transaction as opposed to a statistical sample. With these advances come new challenges though. Compliance officers and organizational leaders must be sufficiently familiar with their institution’s use of AI and machine learning, its benefits and pitfalls. It may require a gradual change to the composition compliance and leadership teams to include individuals with an advanced understanding of programming languages, data science and high math. One thing is certain though, AI, machine learning and deep learning is ushering in a new era of trade surveillance and fraud detection and we’re in for a wild ride.
To hear the full Fraud Eats Strategy podcast episode with Howard Kramer and Oz Vural, click here.
Note: The postings on this site are my own and do not necessarily represent White Collar Forensic’s positions, strategies or opinions