AI Trending

Could Artificial Intelligence Spell Check-Mate for Money Laundering?

Dr Karthik Tadinada, Director of Data Science at Featurespace assesses the potential for machine learning in the fight against financial crime, and separates data-science fact from AI fiction.

The finance industry’s traditional rules-based approach to catching money laundering is at breaking point. As little as 1% of laundered money ends up being seized by regulators. Meanwhile, the scale and ambition of money launderers continues to grow. It’s estimated that up to 5% of Global GDP is laundered – that’s equivalent to $2 trillion every year, roughly the annual GDP of Brazil directly funding organised crime and terrorism. 

Dr Karthik Tadinada

Criminals by their nature aren’t known for their tendency to follow rules. They are as innovative as the legitimate financial services they hide behind. And yet, the finance industry traditionally relies on pre-set rules to monitor transactions and catch money laundering. This casts a very wide net, producing a vast volume of unprioritized alerts for investigators to sift through. The majority of an investigator’s time is spent working false positives (alerts flagging possible ‘suspicious’ activity that turn out to be genuine), leaving little time for the truly suspicious alerts. And novel criminal activity that hasn’t been seen before? Well, that can slip through entirely undetected.

Regulators are naturally keen to see the financial industry adopt new technologies and there’s certainly no shortage of eager salespeople promising an “AI revolution” in the way we fight financial crime. But as with any emergent technology, it’s important to separate fact from marketing spin. A recent European study by venture firm MMC found that 40% of “AI” branded start-ups don’t actually deploy any real machine learning. As the AI-saying goes; ‘garbage in, garbage out’. Data is everywhere but knowing what it means in a real business environment is everything.

As little as 1% of laundered money ends up being seized by regulators.

To a data scientist like myself, AI – or more properly, machine learning – is just a tool. It’s your skill in using the tool that differentiates. Machine learning is in essence an incredible multiplier of human productivity, offering a fast and highly efficient way to make predictions about things that we have historical data for.

Modern computing allows us to build very descriptive profiles for all manner of entities – people, accounts, companies, countries – absolutely anything that you want to follow and build a history and a context for. In this way, we can determine the probability of risk of a particular entity or transaction, and attribute a risk score. The higher the score, the greater the risk. Currently, the best application for machine learning is to give investigators a reason to close an alert or a reason to look more deeply, based on that risk score. 

So, will machine learning ever predict and prevent all money laundering? Well let’s consider an example from another field; medicine. When antibiotics were first invented, people thought they would cure all infections. Now we have antibiotic resistance and need to develop new strategies and treatments to combat it. But no-one would deny that many more people are alive thanks to antibiotics. 

Like deploying antibiotics to fight infection outbreaks, machine learning allows us to combat money laundering on a completely new scale. It can predict suspicious activity and supercharge the effectiveness of investigating teams. Yes, there will be innovative resistance from criminals but by understanding context, behaviour and continually adapting our models, we can stay ahead. 

It’s usually impossible to discuss this subject without at some point being asked whether machine learning will one day replace the hard-working investigator. Well, it turns out that people, institutions and regulators are usually very uncomfortable with the idea of delegating everything to a machine. In the end, they want a human to make the really high-stakes decisions. 

And that is an entirely reasonable position. Look at chess. The strongest possible play comes from Cyborg Chess where an expert human works in tandem with a well-trained computer, selecting the move they think the best from the ones the computer recommends. That combination of human and machine is the real supercomputer, capable of beating AI chess programmes and money launderers alike. 


  • Editorial Director of the The Fintech Times

Related posts

Coconut Uses Open Banking to Transform Bookkeeping for Millions of UK’s Self-Employed

Mark Walker

Proethical People: Oliver Huggins

Manisha Patel

PwC Sells New Independent Fintech Business LIKEZERO

Polly Jean Harrison