Machine Learning in Anti-Money Laundering (AML) Risk Mitigation and Tracing
NOTE: Feel free to email me your suggestions of criticisms for editing.
The breakneck pace of advancement within Machine Learning and Artificial Intelligence (AI) has rightfully given rise to a tidal wave of new concerns for digital forensics investigators, cyber security specialists, and law enforcement officers alike.
As is the case with any tool, however, these technologies can be used for good.
In order to better understand and combat the efforts of those who would put these advancements to use in pursuit of nefarious goals, professionals like those listed above, and countless others, are quickly adopting machine learning and artificial intelligence into their daily workflows.
The field of Anti-Money Laundering (AML) stands out as a strong example of this and as a promising use case to accomplish exactly that, especially with regard to risk mitigation and the tracing of laundered funds.
Anti-Money Laundering risk mitigation is an essential component of any financial institution’s risk management strategy. Put simply, AML risk mitigation can be summarized as the efforts taken to prepare for, identify, and counter attempts made to utilize a financial institution as a means for money laundering, or other illegal uses of funds, such as the funding of terror organizations.
Similarly, Anti-Money Laundering Tracing refers to the process of tracking and monitoring transactions in an effort to flag the kinds of misuse mentioned above. The ultimate aim of which is the identification of the origin (source) and destination (allocation) of suspected money laundering activities, as well as any stops the funds may have made along their journey.
Both Anti-Money Laundering risk mitigation and tracing require the constant, accurate, and comprehensive analysis of datasets larger than you or I could reasonably imagine.
The data generated by the volume, velocity, and variety of transactions in today’s global economy can’t be processed by people alone, technological assistance has always been required. While the traditional, rule-based system approach to AML efforts has been effective to a degree, such systems aren’t nearly as flexible as the evolving suite of tools and strategies applied by money launderers. Luckily, tradition is giving way to something greater.
Enter Machine Learning
Now, those involved in the fight against financial crime are developing, adopting, and optimizing Machine Learning models to overcome these barriers. Machine learning can analyze these datasets far faster and more accurately than human investigators.
Pattern recognition makes up the front line of the battle against money laundering and the foundation on which machine learning was built. This, in conjunction with deep learning, has enabled financial institutions to not only deploy a faster, more accurate means of analyzing financial transactions that runs 24/7 without growing tired, but systems capable of discovering new, complex risk indicators that would not have been possible with manual analysis.
Anti-Money Laundering specialists and financial crime investigators will continue to push these tools, training them up on the latest data to understand the nuances of increasingly niche financial systems and different legal jurisdictions. All the while, criminals will be doing the same as part of a never ending arms race.
Put another way, these tools aren’t going to replace the professionals in this field, as despite constant advancements, the human element remains essential to the AML ecosystem. Rather, Anti-Money Laundering professionals will leave to the machines what can be handled by the machines, so that they can better invest their energy where it’s needed most, such as:
1. Interpreting Results
Machine learning models may be excellent at identifying trends and outliers, but as it is now, the interpretation of these findings often rely on human insight and experience. For instance, even under current systems, perfectly legitimate transactions and innocent people may behave in such a way that a false positive is raised. Their account is flagged, a human investigates further, and in the application of their domain expertise, they may instantly identify this result for what it is, nothing to worry about.
2. Complying with Regulations
Anti-Money Laundering regulators may require justification for decisions made and actions taken by a financial institution, whether that decision was made by a set of rules, a machine learning model, or Anne from her corner office, explanations may be required. As such, human oversight and involvement must be maintained, so as to ensure that no unjustified action is taken.
3. Ethics and Bias
As you might remember from some droning lecturer in Statistics 101, bias is everywhere, difficult to identify, and even harder to account for and remove from an equation. While it may be difficult to imagine why a machine learning model would develop a bias without having been trained for it, it’s important to remember that machine learning models need to be trained on past data. This training data is often curated by and the result of actions taken by real people; real people with opinions.
On the flip side of that same coin, is the possibility that in looking only to the data, without human intervention, that a machine learning model can develop decision making heuristics that are unbiased, but unethical. For instance, an entire demographic can be unfairly flagged and punished by such a model due to a disproportionate amount of abuse originating from individuals within that demographic. Should that be the case, it might be deemed reasonable that the model would arrive at the conclusions it has, but that doesn’t make it ethical to leave it that way.
4. Adaptation and Learning
While Machine Learning models stand to be able to adapt to and learn from behavior, as described by financial transactions, to identify patterns that human analysts may not be able to find, that doesn’t mean it can’t be taught valuable new information by real people. As the living team-members of such models continue to develop in their careers, skill sets, and with regard to the size of their professional networks, they will often come upon new tactics employed by money launderers and those tactics employed to stop them. Where effective, or otherwise deemed worthwhile, these human analysts can work to specifically fast track the recognition of such trends and tactics.
5. Complex Decision-Making
Risk is unreliable. Not that you shouldn’t look to and rely on risk in your decision making, but at the end of the day, risk is about what may or may not happen. That introduces a range of complexities, especially where such decisions made on the basis of this risk are subject to regulatory compliance as discussed above, or with regard to people’s finances. When decisions are made, they better be for the right reasons.
Due to the gray-ness of risk, and the inability to roll out broadly applied black-and-white policy without introducing the financial institution itself to risk, these complex decisions need to be guided by, maintained by, and ultimately decided by one or more people with relevant domain expertise.
Putting the ML in AML
Long story short, machine learning technology is one way that AI and related fields are seeing application within the world of Anti-Money Laundering. This application stands to free up more time for the living, breathing people involved in these fields to do what they do best, leaving the machines to do what they do best.
However, not without the need for checks and balances on the technology. The technology has not surpassed its need for human intervention or guidance, and we don’t expect it to anytime soon. However, if and when it does, we imagine it will be a similar situation, one in which professionals are further freed up to invest their time, experience, and skill where it matters most.
