Nearly a year since the inception of ChatGPT, the risk management teams within financial institutions find themselves in a continuous learning curve when grappling with the challenges and opportunities posed by generative AI (GenAI). As regulations for GenAI in the banking sector loom on the horizon, risk teams are at a crossroads, striving to strike a balance between effectively governing this transformative technology and leveraging it to streamline their risk-related processes.
Recently, Evalueserve hosted a gathering of seasoned risk executives in the vibrant heart of New York City. Against the backdrop of delightful drinks and dinner, a distinguished panel comprising risk leaders and AI experts engaged in a discussion around the nuances of establishing robust oversight for GenAI, the practical use cases they’ve ventured into, and the strategies for a successful GenAI implementation. In this blog, we distill the key insights gleaned from this illuminating discussion.
How GenAI differs from other types of AI/ML:
GenAI stands out by harnessing large language models to perform specific tasks based on textual prompts or commands. Its primary focus lies in two key functions: summarization and generative outcomes. Summarization streamlines documentation tasks, while the generative aspect creates results based on input. For example, GenAI can generate different code versions following specified logic from a given code template. In contrast, traditional AI/ML primarily centers on numerical analysis and signaling.
Regulation and governance:
Panelists discussed the evolving regulatory landscape for GenAI. Regulators are actively engaging with banks, seeking answers, conducting surveys, and consolidating responses to establish appropriate standards. One panelist shared insight into the governance of GenAI within their organization, highlighting the importance of second-line model risk management in setting the tone. Cross-functional oversight, including legal, compliance, and technology departments, plays a pivotal role.
Firms are cautiously exploring GenAI’s potential, starting with low-risk, non-customer-facing tasks. These early use cases include data summarization, policy and report generation, and extracting ESG data from publicly available sources. These applications improve efficiency in data processing and summarization. Evalueserve has been actively collaborating with various industries to identify successful use cases, with compliance and enhanced due diligence (EDD) standing out in the risk sector.
While GenAI offers promise, challenges abound. Firms must evaluate whether their experiments yield the desired outcomes. It’s important to strike a balance between cost and benefit, but it can be difficult to determine the right level of efficiency gains versus costs. Additionally, buyers also need to know how to ‘properly’ test a tool to accurately gauge the efficiency gains.
The potential risks of generative AI are a major hurdle that firms need to overcome. Vendors claiming to use GenAI technology raise questions about new risks and vendor controls. The bank may want to incorporate it into its governance framework for managing vendor risk.
Despite challenges, firms can maximize GenAI’s potential. As one speaker highlighted, “It’s crucial to understand it’s ‘garbage in, garbage out.'” Effective prompt engineering is vital, emphasizing the importance of asking the right questions for accurate answers. Ensuring data quality is also critical, necessitating a robust data quality control process using a mix of manual and automated methods. Model training and quality checks led by subject matter experts are essential for accurate results. Continuous performance monitoring is imperative.
Before using generated content, a human review process should be in place to prevent inaccurate investment decisions. Establishing rules between prompts and expected outcomes and comparing them through multiple iterations helps gauge model accuracy, although this assessment can still be subjective.
Rigvi Chevala, Evalueserve’s CTO, encourages firms not to wait for all issues to be resolved but to explore GenAI’s potential with certain limitations. “Remember when the cloud first emerged, people had concerns about data security, which is no longer an issue at this point,” Rigvi stated. Firms should continue to experiment with low-hanging fruits for various qualitative modelling and work their way up.
In conclusion, the exploration into the world of GenAI has shed light on both the challenges and opportunities that lie ahead for financial institutions and risk management teams. As GenAI regulations inch closer, these teams find themselves at a pivotal moment, tasked with the responsibility of navigating the uncharted waters of this transformative technology. The insights from our panel discussion serve as a compass, guiding us towards a better understanding of how to govern and harness GenAI’s potential effectively.