Artificial Intelligence against digital frauds

This article was published by Ignacio Nordmann on 11/12/2022 at LinkedIn

A new Era

Financial services are undergoing profound transformations that can be characterized as a "change of era" in which it is essential to adopt new ways of "doing banking". The empowerment of the “new clients” of generations X and Y; the increase in transactions carried out in digital channels with increasingly innovative products adapted to the lifestyle and consumption habits of these “new consumers”; the appearance of "new players" delivering financial services (“FinTechs”) boosted by the Open Banking phenomenon, caused these drastic changes in the financial services and payments industry, opening enormous opportunities for "new players" and causing threats that challenge the preponderance of incumbents and “traditional actors”.

Pushed by the preferences of "new consumers" on digital channels, banks deepened their digitization (87% in Brazil according to FEBRABAN). For this reason, the volume of transactions in digital channels increased significantly, while fraud also increased exponentially.

Let´s see some recent examples:

• Computer Weekly LATAM Survey, October 2022:

  • "90% of companies had an increase in online fraud in the last 12 months."

  • “Scammers are targeting consumers who increasingly rely on digital channels for their shopping, entertainment, and banking activities.”

• Forrester Survey, October 2022:

  • “The rapid transition to digital payments and e-commerce in emerging countries is a two-edged sword. While significantly improving customer shopping and payment experiences, boosting the region's digital economy, it leaves many systems and channels vulnerable to fraud."

  • “As the cost of fraud continues to rise in all emerging countries, it is clearly important that fraud detection solutions detect the growing variety of payment fraud across different digital channels, without many false positives.”

• FEBRABAN cites in its 2022 Technology Agenda in relation to AI:

  • "Financial services is one of the industries with the highest volume of customers and data, and the banking sector is under pressure to offer hyper-personalization of products, services, and prices, for all segments, with immediate, seamless, and personalized experiences."

  • "With Artificial Intelligence and Machine Learning we go from regression models to predictive models for understanding consumer behavior."

  • “AI has the ability to improve efficiency, increase differentiation and influence the Customer Experience, being one of the main commitments of banks in 2022”.

In summary, the use of Artificial Intelligence and Machine Learning is in the plans of practically all banks, FinTechs and payment processors. According to FEBRABAN, in 2021 78% of Brazilian banks used AI + ML and in 2022 100% do or will.

Prevention & detection with AI + ML

Scammers, fraudsters, and cybercriminals have also quickly adapted to the new times, becoming true technology specialists, capable of writing Big Data Analytics code and using powerful tools to constantly monitor financial activities online, in order to detect vulnerabilities. And when they identify an opportunity, they escalate rapidly, causing significant losses in seconds or minutes, not hours or days as before.

Thus, the need to completely review anti-fraud strategies and defenses was born, giving rise to the so-called “third generation”, based on the exhaustive use of artificial intelligence and machine learning (AI + ML) components, but which also has other characteristics. that define it:

• Transparency: Analysts and Investigators of financial institutions can view and even modify (optimize by fine tuning) the rules automatically generated by AI + ML ("white box" concept).

• Model Factory: the proliferation of vectors and tools used to make frauds determines that it is not advisable to combat them with a single tool, but instead, with the collaborative combination of different types of automatic engines. This concept of convergence of various engines with different natures, including AI + ML, predictive analytics modeling, neural networks, other statistical engines and even rules written by tool specialists such as Python, is what is called "model factory", recognizing that "there is no silver bullet" in counter fraud activities.

• Cognitive platform: the combination of the best of two worlds: the functionalities of automatic prevention and detection engines (the only ones capable of dealing with thousands of transactions per second and large volumes of data) with the essential human specialized knowledge of the bank's Analysts, is what characterizes a platform as truly cognitive, in which the control and the last word always belong to the Analyst.

• Direct vs. indirect detection: second-generation solutions perform indirect detection because they only use past data from different financial entities (consortium data) to identify criminal patterns. Third-generation platforms also use consortium data, aggregating the precision provided by machine learning on their own data and casuistry, in what is known as direct detection.

• False positives: customer security is a priority, but it cannot conspire against a pleasant and efficient experience when using financial services. Third-generation platforms include entity profiling functionalities (customers, businesses, devices, and others), which, combined with their "cross channel" feature, allows behavioral profiles to be consulted in thousandths of a second with a 360º vision, to reduce to a minimum expression and even to adjust the rate of false positives at will, which is one of the main causes of friction and customer dissatisfaction.

Present & future of Artificial Intelligence

It is clear nowadays that artificial intelligence components with machine learning are the central core of barriers to prevent and detect fraud and attacks in real time, although in the most advanced third-generation architectures they collaborate harmoniously with other automatic engines and the human specialists who operate, manage, and constantly update them.

But we wonder if all solutions from all AI + ML providers are the same. To answer this relevant question, we decided to analyze some aspects of the process of evolution of information technology in general and of artificial intelligence in particular, in both cases not restricting ourselves to software, but also considering the forecasts related to hardware and firmware. To exemplify this analysis, we will reproduce below some paragraphs from an article published by IBM on 10/24/2022:

“The rise of the Internet fundamentally changed the way we use semiconductors. They have gone from simply powering our computers to being an integral part of almost every aspect of our lives. A typical sedan vehicle can have thousands of semiconductors. Entire companies run on smartphones. A pair of AirPods are more powerful than the computers we originally used to send astronauts to the moon. Without access to increasingly powerful chips, our lives will come to a standstill."

“At the same time, there is a growing demand for entirely new types of computing resources. For years, chipmakers have continued to chase Moore's Law, which states that the number of transistors on a microchip will double roughly every two years. But it's not just about packing more processing power on a chip anymore: there's a need to build purpose-built chips, such as those required to run complex AI algorithms and models on quantum computers."

“Some of the earliest semiconductors, those made in the late 1960s, were made using processes that resemble modern techniques. These semis had transistors measuring 20 micrometers across; for comparison, the average human hair is about 75 micrometers thick. For decades, researchers have searched for ways to keep Moore's Law alive by reducing the size of each transistor. By 2005, at the dawn of the mobile internet era, transistors had shrunk to about 65 nanometers, more than 1,000 times smaller than those in the 1960s.”

“In recent years, researchers at the IBM Lab in Albany have made a number of breakthroughs. In 2015, they unveiled the world's first 7 nanometer (nm) transistor, at a time when the state of the art was 14 nm. They also coined the term "nanosheet" and unveiled this new device architecture to help further reduce the size of transistors. The 5nm transistor followed, and by 2021, IBM introduced the world's first 2nm transistor. We're getting to the point where transistors can't get much smaller: after all, a single silicon atom is only around 0.2 nm. At these sizes, the standard laws of physics start to break down. “We are firmly in the realm of quantum physics” said Mukesh Khare, IBM Research vice president of Hybrid Cloud and director of the Albany lab.

“These advancements end up in the types of chips we rely on every day, whether through third-party chip producers who rely on IBM innovations to power their own products, or IBM's own hardware. The newest IBM z16 system runs on an Albany-designed AI inference processor called Telum that gives businesses the power to run AI models while transactions are in progress. A bank can now detect if a charge is likely fraudulent while swiping the credit card."

We recommend the full reading of the article partially reproduced above, retaining the concept that a glimpse into the immediate future allows us to verify that the efficient use of artificial intelligence with machine learning will only be possible through a combination of software and hardware (firmware) components.

And to wrap up for now, we would recommend that when evaluating AI-powered real-time fraud prevention platforms, banks look not only at the AI ​​+ ML components of each solution, but also at whether their provider will be able to follow the planned path of technological evolution and whether these components are integrated into a third-generation collaborative platform.

Ignacio Nordmann
Security & Antifraud Analyst
Member of ACFE (Association of Certified Fraud Examiners)