Meet the researcher: Zoltán Szlávik & Michael Hind

Posted on May 01, 2019

Zoltán Szlávik and Michael Hind have been awarded a TFI long-term research grant to study the relationship between fairness in AI algorithms and banking consumer trust. Read more about the researchers, their project and why they applied for the TFI research grant!

Artificial Intelligence (AI) has been introduced into our daily lives, bringing smart and efficient services that help us live the life we want to live. AI has also opened up many new frontiers of the possible as applications of the technology are being integrated in our businesses. But AI also has a darker side. Being only as good as the data it has been trained with, AI systems can amplify stereotypes: having gender, age or racial biases that operate unfairly to groups of users. Previous research has shown that these biases may be introduced at all stages of the development and use of AI: at the selection of data used to train the AI model, by the developer creating the model, or in the application of the model.

In this study we want to investigate what biases exist in the banking context, how they can be identified and mitigated, and the impact of these biases on consumer trust. Trust is critical for the adoption of AI - if the decision to approve a loan or mortgage is made by a model, do consumers accept this decision? And when financial advice is modeled to prevent worsening of a problematic debt situation, is this advice trusted? Our expectation is that model fairness will improve acceptance. We will explore how "fairness" is best framed in the evolving banking context, and how to correct models to adopt a higher degree of fairness.

We will evaluate several of these use cases using anonymized data of customers, data models and algorithms, as well as crowdsourced human data. We will apply the AI Fairness 360 Open Source toolkit, a set of research innovations resulting from academic studies, to understand the best ways to identify and correct for biases. Furthermore, we will explore what contextual information is needed to be able to create more fairness in the algorithms. Using the insights from the first three steps, we can test the relationship between fairness and consumer trust, and the relationship between consumer trust and consumer adoption by means of an online experiment on a crowdsourcing platform. By better understanding the complex interactions, we aim to inspire the creation of fairer AI algorithms and to empower everyone to make better financial decisions.

“As more and more banking services are powered by AI, it is critical to understand how AI impacts banking consumer trust.”

What was your motivation to apply for the Think Forward Initiative research grant?

We are quite passionate about this topic of reducing bias in AI algorithms. The grant offers us an opportunity to work with actual processes and cases in the banking industry. Consumer trust has always been key to banking. As more and more banking services, from mortgage approvals to personalized advice, are powered by AI, it is critical to understand how AI, and its “behind the scenes” decision-making, impacts this trust. We hope our research creates awareness of the risk of unfairness in AI models, and builds understanding of how to decrease bias in the delivery of services to customers.

How do you expect that your research will contribute to people’s financial well-being?

The Think Forward Initiative’s aim is to help people make better financial decisions. For that mission to succeed, the advice needs to be appropriate to the user, and consumers need to trust and accept financial advice, enough to change their behaviour accordingly. If the advice is driven/given by biased models, the advice may be inappropriate, or the user may be underserved. We hope our research will help improve both cases, and thereby promote inclusiveness.

Zoltán Szlávik is lead and University relations manager at IBMs Center for Advanced Studies (CAS) in the Benelux. The CAS is a Research and Innovation department in which IBM researchers study (enterprise) crowdsourcing, niche-sourcing and cognitive (AI) computing and apply these concepts on real-world industrial challenges in prototypes and first-of-a-kind implementations. Zoltán received his PhD from the Queen Mary University of London, and holds guest lecturer positions at the TU/Delft (Technical University Delft) and the VU (Vrije Universiteit Amsterdam) and is honorary senior research fellow at the University of Exeter.

Michael Hind is a Distinguished Research Staff Member in the IBM Research AI department at the T.J. Watson Research Center in New York. Michael received his Ph.D. from New York University in 1991. He has served on over 30 programme committees, given talks at top universities and conferences, and co-authored over 40 publications. He received a SIGPLAN Most Influential Paper award. His research interests include explainability and bias in AI and larger societal implications for AI; the software lifecycle for creating, deploying, and maintain AI application; programming models and their implementations, static and dynamic development tools, and middleware for emerging commercial paradigms. Dr. Hind’s team developed the AI Fairness 360 toolkit that will be used in this study.

In addition to ourselves, we’ll be working with University PhD- and Master students to carry out the research.