Researchers used experimental deep learning techniques to uncover illicit activity among $6 billion worth of Bitcoin transactions.
Blockchain security company Elliptic teamed up with researchers from Massachusetts Institute of Technology (MIT) and tech giant IBM to use artificial intelligence and deep learning to analyze 203,769 Bitcoin transactions.
The idea was to harness ‘graph convolutional networks’ to create better tools to detect or even predict illicit activities such as money laundering, sanctions violations, and terrorism financing.
Billions laundered through crypto
Billions of dollars of criminal proceeds are laundered through cryptocurrencies each year.
The researchers determined that two percent of the transactions (equating to around 4075 transactions in total) were illicit – whether darknet, money laundering or terrorism financing.
That is double the one percent volume of illicit transactions estimated by Chainalysis – which equated to $2 million a day.
Back in 2012, the proportion of illicit transactions on the Bitcoin network was seven percent.
Around 21% of the transactions were classified as lawful, but the vast majority of the transactions (77%) remained unclassified.
So there may be many more illicit transactions to find.
The ‘unclassified’ figure is expected to decrease as the techniques become more sophisticated.
— elliptic (@elliptic) July 29, 2019
New techniques are promising
Elliptic researcher Mark Weber said the methodology shows promise.
“Graph convolutional networks are still a young class of methods, and we’re in the early days in these experiments, but we do believe GCN’s power to capture the relational information in these large, complex transaction networks could prove valuable for anti-money laundering,” he said.
Chief Scientist and co-founder of Elliptic, Tom Robinson, said the company used the most advanced techniques available to help banks and financial institutions identify suspect transactions.
“Elliptic uses a range of advanced techniques, including machine learning, to facilitate financial crime detection in cryptocurrencies.
“Our work with researchers from the MIT-IBM Watson AI Lab builds on this, to ensure that our clients have access to the most accurate and effective insights available, reducing their compliance costs and ensuring that their services are not exploited by criminals.”
Eliminate the false positives
Elliptic is frequently hired by law enforcement agencies so the company doesn’t just want to estimate illegal transactions – they want to identify them with a high degree of accuracy.
“A big problem with compliance, in general, is false positives. A big part of this research is minimizing the number of false positives,” Robinson said.
“The key finding is that machine learning techniques are very effective at finding transactions that are illicit.”
The dataset has been made public to encourage open source contributions to help anti-money laundering efforts.