Why not use machine learning to crack down on money laundering?
An interesting article in the MIT Tech Review, “The pandemic has changed how criminals hide their cash — and AI tools are trying to sniff it out”, describes how the coronavirus pandemic, with its lockdowns, travel disruption and restrictions on economically or cash-sensitive activities, is a headache for criminals trying to launder money.
Each year, according to UN estimates, between $800 billion and $2 trillion of criminal proceeds are laundered and reintroduced into the economy, usually using small businesses that primarily handle cash, and moving it illegally across borders. But the restrictions imposed by the pandemic have forced criminals to seek new strategies to launder their ill-gotten gains that are proving difficult for the authorities to tackle.
The solution could lie in using machine learning’s ability for anomaly detection. Applied to a set of data, anomaly detection identifies outliers that may indicate fraud or data quality problems, without the need to have previously labelled the data. Unsupervised learning tools for anomaly detection — those I am most familiar with are developed by BigML, a company to which I am a strategic advisor — assign a score to each case in a data set of between 0% and 100%, and scores of 60% or more are usually considered outliers.
