The financial services industry has undergone significant disruption in recent years, particularly with the rise of “open banking” and the increased control that customers have over their financial data. As a result, the industry has had to adapt by embracing digital technology and artificial intelligence (AI). AI has the potential to revolutionize financial services by enabling demand and revenue forecasting, anomaly detection, decision support, and more. However, the industry is heavily regulated, with strict guidelines in place to protect sensitive data and ensure responsible AI use.
Discover Financial Services, a leading financial institution, has been gradually incorporating AI into its operations to improve efficiency and customer service. Raghu Kulkarni, the Senior Vice President and Chief Data Science Officer at Discover, recognized the need for an AI governance council to establish repeatable processes and safeguards before implementing AI initiatives. This council ensures that AI is used safely and effectively, with input from various stakeholders such as cybersecurity, architecture, model risk, compliance, legal, and data science.
Kulkarni explains that the council’s primary focus is on responsible AI use, which includes ensuring accuracy and avoiding biases. Discover utilizes AI in various areas, such as fraud detection and customer service. For example, they employ generative AI models to summarize customer service calls, but ultimately, human intervention is still crucial to ensure the accuracy and tangibility of the output.
While Discover does not currently have a prompt engineering job role, Kulkarni foresees it becoming important in the future as generative AI advances. Prompt engineering involves fine-tuning AI models to suit specific needs. However, Kulkarni believes that the technology is still a few years away from reaching that stage.
The creation of the AI governance council at Discover was motivated by the realization that AI implementation requires a collaborative approach. It is not solely the responsibility of data scientists but also involves cybersecurity professionals, architects, legal experts, and compliance officers. By working together, the council ensures that infrastructure is robust, privacy is protected, and model risk and compliance are addressed.
The council has established principles and policies for AI use at Discover, which include understanding the problem to be solved, selecting the appropriate model, considering cybersecurity and architectural principles, and adhering to existing guidelines and regulatory frameworks. They take into account evolving policies and standards set by organizations such as the NIST framework and the US Chamber of Commerce to ensure responsible and transparent AI usage.
Kulkarni acknowledges that the greatest threats posed by AI in the financial industry are related to cybersecurity and the potential misuse of AI for fraudulent activities. However, he remains optimistic, stating that there are always solutions to counter these threats.
Kulkarni advises other businesses facing security, privacy, and regulatory challenges from AI to work together and take a comprehensive approach. He emphasizes the importance of simplicity, compliance with regulations, and continuous learning through simulations and simpler models.
Looking ahead, Kulkarni predicts that large language models (LLMs), such as GPT-2, will become more domain-specific as computing power advances. This means that institutions can develop LLMs tailored to their specific needs, rather than relying on generic models. Discover currently uses NLP and GPT-2 models, but they are cautious and purposeful in their deployment, focusing on smaller LLMs that are explainable and manageable.
In terms of connecting LLMs to back-end systems, databases, and documents, Kulkarni envisions a future roadmap that involves the development of a vector database and extensive prompt engineering. However, current regulations and architectural constraints are still being navigated. Currently, Discover uses batch models that involve running the LLMs and then having a human review the results before finalizing them.
Discover Financial Services' approach to AI governance serves as an example for the financial services industry. By establishing a council that includes experts from various disciplines and adhering to responsible AI practices, Discover ensures the safe and productive use of AI, leading to improved efficiency, customer service, and overall success in the industry.
Use the share button below if you liked it.