Subscribe to our blog
Get ahead with our award-winning insights on the latest developments in the fraud and compliance landscape. Enter your email below to receive our blog posts directly to your inbox.

Scott W. Bauguess is the Acting Director and Chief Economist at the Division of Economic and Risk Analysis (“DERA”) from the US Securities and Exchange Commission (“SEC”). He recently gave a Keynote speech: “The Role of Big Data, Machine Learning, and AI in Assessing Risk: A Regulatory Perspective”, that many people missed. In it he covered how the SEC is taking advantage of these technologies, and more importantly where they are not, at least not yet.

The speech touches many areas of Machine Learning that we as technologists and vendors of a compliance solution face daily. First, let me agree with Scott W. Bauguess that the use of machine learning has great benefits. We see it with our clients every day. Second, let me also agree that it is not, at least not yet, a substitute for trained humans

I say not yet, and perhaps not ever, because, the law and regulations have many gray areas and loopholes. Laws aren’t precise, laws have interpretations, and so do regulations. Moreover, the widespread use of technology – AI and machine learning – in this space is at the mercy of regulators.

It is hard for a machine to interpret law and regulations. The same legal text can be read differently, to the point that it can be used to support completely opposing positions, making this conceptual task so opaque, that it is beyond current technology’s capacity. A machine can interpret law and regulations, so long as there are enough cases on which a legal text has been interpreted in the same way, but it cannot come up with new interpretations because the human process to come up with them is a highly unstructured task, and there is no existing data for the model for it to work. Machines are rigid and can’t understand nuances of language, or understand the intent of laws, less so when these are a set of rules written by lawyers. Also, to meet regulations, you must be able to explain why certain decisions are made, they need to be justifiable. Machine learning decisions are basically the results of historical learnings but often can’t explain exactly why it made a decision.

It’s also worth mentioning that new legislation and regulations on the same topic come out, or interpretations change over time – they evolve with social and legal customs, or even with guidance dictated by a regulatory body on the subject.

The reality is that both risk and compliance analysts can (and do!) use machine learning to inform their operational workflows. Machine Learning can effectively highlight situations that require attention, and it can also remove noise from their daily operations.

In our platform, we apply machine learning to encapsulate the value of our Identity Graph in a simple score – a score that shows the risk of a transaction based on the complexity of an Identity Graph’s properties and associated behavior (you can read about it in these blog posts). As a newly integrated part of our platform, machine learning adds to the overall value — the graph score reduces false positives, which in turn reduces manual reviews, and increases efficiencies in day-to-day operations.

I bring up our implementation because it is important to understand what machine learning can do for you, and where it makes more sense, and most importantly that it is ONE tool in your bag of tools. One that you should use if you have the resources, either to build or to partner with a vendor.

We have this conversation almost daily with our potential clients, who by looking at the marketing material of the latest machine learning company- and there seems to be a new one every week –  believe that all they need is machine learning to address risk and compliance. It isn’t quite that simple, in fact, any vendor that is claiming that can do all of it, by itself, be it with machine learning or without it, is probably looking at a very narrow issue.

Machine learning can help solve a variety of problems, especially those problems where you have enough data. And depending on what data you have, how good the data is, and what problem you want to solve then you can train a model to help you with your problem. We agree with Scott that data quality is very important. I know this based on our clients’ results – the better the data, the better the results, and more data isn’t necessarily better at the expense of quality.

To summarize:

  • Machine learning is an important tool for financial crime analysts.
  • It is not the end all/be all for regulatory compliance – laws are complex and are subject to interpretation
  • The effectiveness depends on the amount of data but more importantly in its quality

I appreciate this perspective of a regulator into the matter, and him sharing his use and views into the use of this important technology. It resonates with us based on our own experiences with current and prospective clients.