At Techsylvania’s 2018 conference in Cluj, Romania, our VP Engineering, Luke Mawbey explains the diversity of techniques and approaches needed to fight financial crime, including when to use machine learning algorithms, and how they’re used to decide financial crime risk. Watch the video and read the Q&A below.
Q: What technologies do you use at ComplyAdvantage
We use a variety of techniques and approaches – and fundamentally you want to use the best technique and best approach (and if you can get away without using machine learning then perhaps you should do that!). But fundamentally when you’re trying to analyze such a massive amount of public data, then you need to switch to some more intelligent techniques & modern techniques. Using things like distributed systems and cloud computing is one approach, but getting into modern techniques such as natural language processing and machine learning is a critical step in the fight against these sorts of crimes.
Q: When using AI, to find out for example, which politicians are corrupt, how are you making the difference between the boundaries of good and bad, and how do you correct them?
When you are relying on machines and algorithms you have to be sure that they’re making the right decision. We approach this in a few different ways: the first way, is that when you build an algorithm you have a responsibility to test that to see what sort of data it’s producing, to check the quality of that data, and you need to have a threshold to make sure that it meets your requirements. Once you have data above that threshold, then you’re safe to use it in a production environment.
It’s also important to recognise that people make mistakes as well. And with the sort of data we’re analysing, you need to be thoughtful when you’re doing this analysis. You need highly educated people to do that analysis if you’re going to collect that data manually. And if you’re doing that on a day to day basis, on such a repetitive task, people are going to make more and more mistakes, and you’re going to have fundamental issues in your process. So by having an automated process and by having a machine learning trained algorithm, at least you can be confident, that the data you’re producing is of consistent quality time after time after time.
But finally, I think the biggest point is – we’re collecting data and providing data for people to make a decision on, at the end of the day. And so we provide the information that we’ve collected for a human to analyse to make the final decision on whether or not it is representative of someone being corrupt, of someone being linked to money laundering or terrorist financing or anything else.