18 July 2012
Dodgy dealings at banks are in the spotlight again – but AI-based systems can help spot bad behaviour.
LAST week, another UK banking executive was made to squirm in front of the TV cameras as he was grilled by angry parliamentarians over alleged corruption in his organisation.
Barclays bank was fined $450 million by UK and US authorities for its part in an inter-bank interest-rate fixing operation. Now the bank’s former chief executive Bob Diamond has told a UK Treasury committee that rogue employees operating far outside the bank’s trading and dealing rules were to blame.
“When I read the emails from those traders I got physically ill,” Diamond said of the missives that proved Barclays staff had rigged the rates.
The computer power that has automated financial markets and made lightning-fast trading possible is being turned on the more suspect elements that operate within the world’s financial system. The techniques are based on our love of smartphones, tweets, blogging and email.
Thanks to the US rules of corporate governance that came in after the Enron and WorldCom scandals, firms like Hewlett-Packard’s subsidiary Autonomy Systems are selling software that flags up employees whose trading deviates from the norm. Meanwhile, a European team has developed a system that can be customised to detect anything from money-laundering schemes to insider-trading scams (Digital Investigation, DOI: 10.1016/j.diin.2012.04.003).
Autonomy’s software trawls a company’s information systems for “unstructured” data – as opposed to information held in corporate databases. This could include tweets, text messages, Skype video, smartphone data, emails, or transcripts from phone calls. The software then hunts for examples of behaviour that varies from normal practice.
Autonomy is not the only organisation combing through unstructured data. At the AGH University of Science and Technology in Krakow, Poland, a team of computer scientists led by Rafal Drezewski has written software that helps the Polish state police check for evidence of money laundering in phone calls, SMS and bank statement data.
Their system uses AI routines to seek out and visualise clusters of information. This might include money transfers from vast numbers of sources to a single account, or many people texting or phoning someone of interest. “We detect suspicious patterns and roles played by different people using pattern mining and social network analysis algorithms,” says Drezewski.
It doesn’t end there. This cluster analysis system could be customised to check through bankers’ emails, texts and documents to search for suspicious-looking activity.