Year: 2025
Type: dissertation
Abstract: Risk registers are a structured way to document an organization's risks and are a central part of the risk management process. Over time, risk registers can become incoherent, filled with redundant risks and missing relevant ones. Such risk registers may become overwhelming for experts to manage. This thesis developed an artificial intelligence (AI) assistant framework designed to address common issues in risk registers. The AI assistant utilizes large language models (LLMs) enhanced with structured tools and a retrieval-augmented generation (RAG) search. The AI assistant can read information from the organization's risk register, RAG knowledge base, and user inputs to create a comprehensive working context. Using structured tools, the AI assistant can suggest modifications to the risk register. Multi-step reasoning enables the completion of more complex tasks that involve multiple tools and data sources. A case study with Inclus Oy examined how an AI assistant affects the speed, quality, and handling of complex data in the risk register while maintaining a reliable process. Analytical generalizations were made based on observations of the risk register, AI assistant, and perception case study participants. The study yielded promising results, where the AI assistant suggested merging duplicate risks, harmonizing descriptions, and adding new risks. A survey revealed that the use of an AI assistant improved risk register coverage and made it more accurate and up-to-date. While an AI assistant substantially speeds up risk management tasks and improves the quality, it is still prone to hallucinations. The study demonstrated that AI assistants must operate under human oversight and that users must retain the authority to review and confirm every proposed modification to the register. Additionally, human oversight ensures that participants are committed to mitigating the risks. AI assistants can enhance the participatory process by accelerating and improving the quality of data processing.
Source: Aaltodoc (Aalto University)
Author: Westergård, Alexander
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7076957801