Year: 2026
Type: dissertation
Abstract: Generative Artificial Intelligence (GenAI) has rapidly become a common tool in corporate strategic decision-making and risk assessment. Due to the efficiency and capabilities of GenAI tools, the cognitively easiest solution for a decision-maker is often to delegate strategic risk analysis to AI. This tendency easily leads to overreliance, which means accepting GenAI output uncritically without independent verification. The objective of this literature review is to examine overreliance on GenAI tools in the context of corporate strategic risk assessment. The thesis investigates the cognitive consequences of overreliance on the manager, the operational consequences for the company, and methods for designing workflows to avoid this phenomenon and its adverse effects. The research is limited to the micro-level of an individual decision-maker. The literature search was conducted in English and primarily restricted to scientific publications from 2021 to 2026. The results show that, cognitively, overreliance on GenAI might bring out the manager's own biases through selective adherence. Furthermore, a degradation of professional skills and intuition can be observed. On the other hand, GenAI errors can lead to algorithmic aversion, causing the tool's potential benefits to be lost. Operationally, overreliance could expose the company to flawed decisions, strategy homogenization, and ambiguity in accountability. Strict limitation of the AI's role, cognitive forcing functions that create cognitive friction, and GenAI counterargumentation were identified as effective countermeasures against overreliance. In conclusion, companies should build a strategic risk assessment workflow that prevents overreliance on GenAI and mitigates its negative cognitive and operational effects. GenAI should be strictly limited to an assistive role in structuring information while the final decision-making remains with the manager. A secondary AI model should be integrated into the analysis phase to challenge the primary model's analysis and identify errors. Additionally, to maintain professional capabilities, the process should be executed entirely manually on a regular basis.
Source: Aaltodoc (Aalto University)
Author: Nykänen, Kasper
Institution: Aalto University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225846570