Artificial intelligence is our past, present and future
Artificial intelligence (AI) is the current buzzword of business. Computer software is helping us choose our films and music, dealing with our customer service enquiries, and improving our healthcare. It is also helping fraud managers across the world to spot and stop fraudsters, keeping us safe online.
Concepts of artificial intelligence have existed for hundreds of years. More recently, mathematics and engineering have combined to produce algorithmically-driven probability-based systems that can perform pre-defined tasks. The advent of today’s digital technologies has opened infinite possibilities and fostered tremendous progress in the theories and capabilities behind these algorithms.
Essentially what we mean when we refer to artificial intelligence is the ability of a machine to solve tasks. The ultimate goal is often seen as mimicking human intellect as closely as possible, but the reality is that we are seeking automated solutions that make our lives easier.
In modern fraud fighting, machine learning is the essential reasoning system that enables fraud managers to track and respond to the rising threats faced around the world. More of us now transact online, which means greater potential rewards for fraudsters. But all of us leave a trace online, in everything we do, generating huge amounts of data. Using “stream processing”, we can constantly ingest huge streams of data, which enables faster reactions to emerging threats – asignificant enhancement on rule-based fraud systems that process data in batches.
AI learns from this data, meaning that businesses can address their specific fraud problem rather than requiring expensive custom-built solutions that drain resources. More data also means we are beginning to understand how fraud itself works, which means we can get ahead of it.
Let’s look at a specific fraud issue as an example. From June 2015 to June 2016, we saw an increase of up to 300% in account takeover attempts on our ecommerce customers. The relative ease with which fraudsters can access less secure accounts is causing a major headache for retailers. Fraudsters can buy login details from the black market, or steal them through malware or phishing attacks, or sometimes simply run through the most common passwords to crack a customer’s online shopping account.
The challenge in tackling account takeovers is that they’re typically hard to detect. Fraudsters operate from within genuine and trustworthy user accounts, often with an impeccable purchasing history, but will make small changes to the account so they can obtain goods to sell for a profit. Importantly when it comes to AI, fraudsters will also change their tactics frequently, whether it’s going for lesser-known brands, smaller value items, or masking attempts via proxy services or numerous devices.
In Europe, data privacy regulations are stricter than elsewhere, and will become more so after GDPR is implemented in 2018, but by anonymising the data, we can still identify fraudsters based on these ever-growing data pools. Artificial intelligence provides us with levels of information above and beyond individual fraud. The speed and scale at which AI can analyse and make connections between different data points enables us to identify orchestrated attempts of fraud; in other words, we can stop multiple attacks in parallel.
Many fraud attacks take place within a given timeframe and AI provides fraud managers with the capabilities to find similarities and connections with relative ease. Fraud departments can then find not just individual fraud, but use one fraud case as an anchor to find much more fraud information, protecting the business and customers from other risks.
This interaction between man and machine is essential in building the best defence against fraud. Without support from each other, connections and cases can get missed or wrongly interpreted. Together, human intelligence (HI) and artificial intelligence (AI) learn more about their fraud problem.
Unlike old anti-fraud systems, AI software adapts to changing data and fraud managers can label cases as they discover them. So, if the software incorrectly identifies something as fraud based on multiple risk factors, the human expert can say no, and the system learns. This means if new fraud trends began in the last week, companies can spot them quickly and act on them. Giving the system feedback enables it to absorb changes in patterns and provides stronger fraud defences.
Tuning fraud on a consistent basis is particularly important, as different companies have a different risk appetite depending on the nature of their business. Economics drives every aspect of the commercial world and some businesses are more willing to accept risk if it means more custom. AI can help compute false positive rates if it makes economic sense. The purpose of anti-fraud engineering is not simply “to stop fraud”, it’s to limit damage to the company.
Alan Turing’s “imitation game”, better known as the “Turing Test” is world famous for identifying the level at which a machine can be identified as intelligent. Turing proposed that in order for a machine to truly think, it must be able to answer questions in a way that is indistinguishable from that of a human.
Fraud prevention should work in similar fashion: the HI and AI elements should be blended seamlessly so that it is almost indistinguishable between the two; the priority is in reducing the risk for businesses.
What could once only be imagined is becoming reality. The Turing Test has been passed (albeit subject to interpretation), and each day, huge volumes of data are processed by AI algorithms, creating greater accuracy as software becomes stronger. The journey is ongoing, but with constant advancements in the capabilities of AI the future is looking bright. The same cannot be said for fraudsters.
Download the original article published in PCM Magazine, issue 08 2017 (PDF)