Designing Trustworthy Algorithms: Operational Environment as a Cornerstone for Governing Risk and Enhancing Trust
Year: 2025
Type: dissertation
Abstract: Algorithms and human lives are, in many ways, symbiotic; algorithms can be found anywhere from personal devices to space exploration on Mars. We rely heavily on algorithms to assist and take care of us, but algorithms can also have fatal unintended consequences. The subject of this thesis is to find a way to increase algorithms' trustworthiness by focusing on identifying and categorizing the unintended consequences of algorithms on the operational environment of algorithms. The objective of this thesis is to provide what we call an Algorithm Power Matrix to aid algorithm developers, regulators, and private sector management in their quest for safely deploying and managing advanced algorithms on a continuous basis. Furthermore, we achieved additional progress in algorithm governance, a development that wasn't originally within the scope of this study but was enabled by our primary discoveries. The study was structured using design science research and design science research methodology. A comprehensive literature review was conducted on algorithms and how they interact with our society. We cover the fundamentals of algorithms, how they are built, operated, and interacted with, what consequences algorithms have, and how they are regulated. This is the first phase to cover the essential parts of the study, which resulted in the creation of the first version of the Algorithm Power Matrix. The second phase involved a group of experts reviewing and suggesting editions for the second iteration cycle. The final phase was to review the iterated framework again, leading to the third iteration cycle and finalized framework presentation. The conclusion was that even experts can have difficulties understanding how different environments affect algorithm performance. Currently, there is an intuitive way in which the developers utilize similar principles that are presented in this thesis, but this was the first time they saw something like this on paper. There is a clear need to develop more tools for under-standing and controlling the effects of algorithms, and the matrix provided in this thesis is just one example. The main contribution of our research is the Algorithm Power Matrix, which is a novel framework for assessing and managing the potential impact and risk of algorithms based on their degree of freedom, autonomy, and human dependency. The matrix can help stakeholders to, most importantly, identify the appropriate level of oversight and accountability required for specific algorithms. Ultimately leading to the design and implementation of ethical and trustworthy algorithms that respect human values. Further studies are necessary to determine more precise thresholds for categorizing algorithms and their operational environment.
Source: Osuva (University of Vaasa)
Author: Harmaala, Rasmus Valtteri
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W6990255454