Use of Generative AI in Large Financial Institutions

QuantUniversity had the second lecture in the QU AI Fall School yesterday

Jacob Kosoff from 美国银行 discussed the use and governance of generative AI in large U.S. financial institutions. It explores the practical applications of AI, such as content classification, summarization, and content generation, and addresses the associated risks, including algorithmic transparency, compliance, and fairness.

Slides and Video:

The slides and video from the presentation is available here:

Register using the code "QUFALLSCHOOL2024" to get access to the video and slides.

https://academy.qusandbox.com/market/670598c150c75511c728b53b

Here is a summary of the talk!

The financial industry is exploring and adopting generative AI technologies, but with a cautious and pragmatic approach. While the public might imagine widespread use of AI chatbots and automated decision-making, the reality in late 2024 is more nuanced.

Generative AI tools are primarily used to assist employees, not replace them, and to enhance employee experience rather than directly interact with customers. This approach focuses on improving efficiency and effectiveness in internal processes.

Jacob highlights this trend by saying that “the output of generative AI is not an auto decision” and that it is designed to “assist an employee” in making “a better decision.”

Use Cases of Generative AI

One popular framework used by financial institutions categorizes generative AI applications into three main buckets: classifying content, summarizing content, and generating content.

• Classifying Content: This involves using AI to analyze information and categorize it. One prominent example is classifying customer interactions. Instead of manually categorizing customer calls, AI can analyze transcripts to identify complaints and their nature. This allows for a more efficient and data-driven approach to customer service.
• Summarizing Content: Financial institutions deal with massive amounts of data, including news articles, research reports, and internal documents. Generative AI can be used to summarize these large volumes of information, making it easier for employees to access and understand key takeaways. This can be especially helpful in areas like loan underwriting, where bankers need to quickly process information about potential borrowers.
• Generating Content: Generative AI tools like Github Copilot are being used to assist programmers in writing code. The AI suggests code snippets, which programmers can then accept, reject, or modify. This can speed up the coding process and potentially improve code quality.? Other examples include creating first drafts of credit approval memos and marketing materials.

Risks and Challenges

While the potential benefits of generative AI are significant, financial institutions are also acutely aware of the risks:

• Regulatory Scrutiny: The rapidly evolving regulatory landscape around AI, with various state and national laws, presents challenges for financial institutions. Ensuring compliance with these regulations adds complexity to the development and deployment of generative AI applications.
• Model Risk: As with any model, generative AI models can produce inaccurate or biased outputs. It is crucial to have robust model risk management frameworks in place to identify, assess, and mitigate these risks. This includes ongoing monitoring of model performance and implementing controls to ensure fair and unbiased outcomes.
• Complacency Risk: Over-reliance on AI outputs without proper human oversight is a concern. Employees must be trained to critically evaluate and challenge AI-generated suggestions, recognizing that these tools are not infallible.
• Cost-Benefit Analysis: Implementing and maintaining generative AI solutions can be expensive. A thorough cost-benefit analysis is essential to determine whether the potential benefits outweigh the costs. This includes considering factors such as hardware requirements, software licensing fees, and the need for specialized talent.

Future Trends

Looking ahead, the use of generative AI in financial services is likely to evolve in several ways:

• Rise of AI Agents: The speaker envisions a future where AI agents handle more complex tasks, such as mortgage applications and travel bookings. These agents would interact with customers and perform actions on their behalf. This raises new challenges around security, privacy, and the risk of unauthorized transactions.
• Increased Collaboration: Successfully integrating generative AI into financial institutions requires collaboration across different departments, including technology, legal, risk management, and data privacy. Clear communication, shared understanding of risks, and harmonized policies will be crucial.
• Focus on Practical Value: While the hype surrounding generative AI is undeniable, financial institutions are ultimately driven by practical considerations. The focus will remain on identifying use cases where these technologies deliver tangible business value, improve employee experience, and enhance customer service without compromising on risk management and ethical considerations.

Join us for our other upcoming events:

1. On demand courses at www.quantuniversity.com

2. Next Lecture:

1/2 day QuantUniversity Workshop on AI & Investing on October 18th 9.30-12.00pm EDT (In Boston and Online)

https://web.cvent.com/event/6cf00e1d-9f3b-4244-9db3-279d9efeef42/summary

Looking forward to your participation!

Sri Krishnamurthy, CFA, CAP

QuantUniversity