By : Dena Hamilton
June 02, 2017 12:00 AM
As some of the biggest tech trends of 2017, artificial intelligence (AI) and machine learning promise to offer a great many opportunities to businesses across all parts of the economy, and the banking sector is no different.
Among the most talked-about use cases for this technology in banking is improving fraud detection capabilities. There’s a lot of buzz about how this can transform operations by analyzing transactions in real-time and applying algorithms to determine whether they should be allowed to proceed unhindered, to request further information to verify if it is genuine, or to block it altogether.
Many people use AI and machine learning fairly interchangeably when it comes to this technology, but this could create confusion among users and could lead to some companies ending up with a lack of clarity about what they’re actually getting when they’re buying or using an AI or machine learning solution.
So what difference, if any, is there between the two, and what impact will this have on how a bank deploys the technology in order to combat fraud?
While the two technologies are certainly closely related, and there is indeed a lot of overlap at a casual glance, there are a few key differences. AI refers to the ability of a machine to conduct actions without the guidance of a human operator, while machine learning refers to a particular approach to AI that is able to take data and algorithms and apply it to new scenarios and patterns without being programmed directly.
In essence, artificial intelligence can mimic actions it has seen or been previously taught about without any new intervention, and is used to conduct a wide range of specific tasks. Such ‘applied’ AI has been in use for years now, for activities such as automatically trading stocks based on a predefined set of rules, identifying and sorting images, or playing chess.
Machine learning, however, is often regarded as an extension of AI, and viewed by many as the next stage in the evolution of the technology. The key characteristic of a machine learning algorithm is the ability to ingest large volumes of data and ‘learn’ for itself how to apply its knowledge to future scenarios.
A simple example of this may be an autocorrect feature on a phone or computer. If a user misspells a word, an application may suggest what it believes to be the most likely option based on its inbuilt dictionary. But if a user rejects the suggestion or corrects it to something different, a machine learning system can register this without being specifically instructed. In the future, it can then correct it more accurately to what it believes the user meant to say, or recognize that the unusual word makes sense – much like a child learning a new word through exposure and context.
This does not necessarily mean that machine learning is intrinsically the only option for use cases involving fraud detection. In fact, for certain scenarios, where banks are looking for a narrowly-defined set of parameters, or reacting to a new fraud vector, using rules can be the answer for fraud prevention in real time.
Machine learning, on the other hand, is better-equipped to deal with spotting evolving patterns and reacting accordingly without instruction or human intervention. For example, when it comes to fraud detection, AI may be able to monitor the buying patterns of a consumer and send out an alert if it spots a transaction that differs from the norm. However, with machine learning, the system, can recognize wider changes in habits and bring in data from elsewhere to build its understanding of what a fraudulent transaction looks like without the influence of a human.
Fraudsters are innovative and constantly change their strategies in order to evade detection, and the amount of information generated on a daily basis will be far too much for a human observer to sift through. Being able to use machine learning technology to spot shifting patterns can be an essential part of improving detection rates while eliminating false positives.