Navigating the Reality of GenAI in Risk Management
Society’s preoccupation with artificial intelligence (AI), particularly generative AI (GenAI), has ignited high expectations across industries, including finance and specifically risk management. But questions remain over the realistic capabilities of AI technology as it stands today and the appropriate use cases.
According to a Chartis report, Dismantling the Zeal and the Hype: The Real GenAI Use Cases in Risk Management, heavy investments in this area are only serving to fuel inflated expectations around AI’s current capabilities, including Microsoft’s $10 billion investment in OpenAI, as well as new government initiatives, such as the UK’s pledge to invest £250 million into AI. However, putting all the buzz aside, it’s essential to take a close look at where GenAI genuinely adds value in the realm of risk management, and to identify areas where traditional machine learning (ML) techniques may be the more prudent choice.
Breaking Down the Hype
When it comes to GenAI, widespread enthusiasm is often overshadowing reality. A key limitation is that while AI models offer significant benefits in natural language processing (NLP) and text generation, they are computationally intensive and costly to deploy. It’s a misconception to think of GenAI as a general-purpose tool when, in many instances, simpler rule-based systems or traditional ML may be more effective, efficient and less costly for certain tasks.
The Need for Domain-Specific Adaptation
Large language models (LLMs) like ChatGPT are trained on a broad range of general data, but lack deep, subject-specific intelligence in many domains. However, users can fine-tune such models using proprietary data to improve relevance and accuracy. It’s important to bear in mind that adapting LLMs to financial contexts, such as risk management, can demand significant domain expertise and data.
Practical Applications of GenAI in Risk Management
While many speculative use cases for GenAI abound, there are several real-world applications in financial risk management that are beginning to emerge. A few of these include:
Internal Co-Pilots: GenAI models can act as internal “co-pilots,” aiding analysts in risk assessments, report generation, and documentation tasks that might otherwise be manual in nature. By automating routine yet time-intensive processes, GenAI enables professionals to focus on more complex analysis, thus boosting productivity across the business.
Improving Accessibility: GenAI tools are emerging that can help make technical language models more accessible to a non-technical audience. Such tools may be used to produce simplified natural language instructions that are based on more complex domain-specific languages. By introducing natural language versions, GenAI can help broaden the audience scope for certain applications.
Data Retrieval and Summarization: Another promising application lies in information retrieval and summarization, where GenAI models can quickly process large volumes of financial documents, such as analyst reports or regulatory filings. This feature has proven particularly beneficial in compliance tasks and can be instrumental in analyzing company reports for potential risks.
Addressing the Downsides of GenAI
Despite the potential of GenAI, these models do introduce distinct challenges:
Computational Costs: LLMs demand substantial computing power, often involving GPUs and high-performance infrastructure. This expense can limit GenAI’s practicality for smaller financial institutions and highlights the importance of understanding GenAI’s true return on investment (ROI).
Accuracy and "Hallucinations": GenAI models occasionally produce inaccurate or “hallucinated” results, where the model generates plausible-sounding information that is factually incorrect. In finance, where precision and trust is paramount, these inaccuracies pose risks and necessitate human oversight.
Model Governance and Bias: Financial applications require transparency and fairness, which can be challenging to achieve with GenAI. To mitigate risks, firms must focus on robust governance and bias reduction strategies. Ensuring that model outputs align with ethical standards and regulatory requirements is critical for effective deployment.
In many cases, conventional ML approaches remain competitive with or even superior to GenAI. For structured data analysis and repetitive tasks, rule-based systems are often more cost-effective and simpler to implement.
Employing a Balanced Approach
Investments and enthusiasm around GenAI are set to continue. Yet, the financial industry must carefully evaluate where GenAI adds real value and where alternative approaches remain more judicious. The best path forward combines realistic use cases, rigorous oversight, and an understanding of the technology’s limitations, ensuring that firms benefit from all AI has to offer, while managing the risks.
For additional insight around deciphering between AI hype and practicality, check out our podcast: Debunking AI Myths: A Glimpse into Tech’s Future with Don Welch