Extending the sustainability-quality model for AI systems
Dula, Mhreteabe (2025)
Diplomityö
2025
School of Engineering Science, Tietotekniikka
Kaikki oikeudet pidätetään.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi-fe2025080781378
https://urn.fi/URN:NBN:fi-fe2025080781378
Tiivistelmä
As Artificial Intelligence (AI) systems become increasingly embedded in critical societal infrastructures—ranging from education and healthcare to governance and security—they introduce unique sustainability challenges. Traditional sustainability assessment frameworks, including the Sustainability-Quality (SQ) Model (1), are insufficient to capture the dynamic, data-intensive, and ethically complex nature of AI systems. New concerns such as fairness, explainability, lifecycle energy use, and adaptability demand a tailored approach to sustainability reasoning.
This thesis presents an extension of the Sustainability-Quality (SQ) Model to support structured evaluation of AI systems from a sustainability perspective. The extended model leverages a graph-based representation to capture the complex interdependencies between system features and diverse sustainability attributes across social, environmental, economic, and technical dimensions.
Implemented using Neo4j(2), the model enables interactive querying and reasoning over trade-offs, synergies, and influence pathways within AI system designs. Its flexibility allows stakeholders to define and analyze context-specific sustainability concerns—ranging from energy consumption and robustness to transparency, ethical governance, or long-term social impact.
The framework is empirically validated through two case studies: EMotivA, a persuasive emotion-aware system, and Perspective API, a multilingual content moderation service. Graph-theoretic metrics such as centrality and trade-off density are used to uncover design tensions and support informed decision-making.
Key contributions include: (i) a formalized and extensible SQ model adapted for AI contexts, (ii) a graph-driven framework for sustainability modeling and analysis, and (iii) empirical validation using real-world AI systems.
This thesis presents an extension of the Sustainability-Quality (SQ) Model to support structured evaluation of AI systems from a sustainability perspective. The extended model leverages a graph-based representation to capture the complex interdependencies between system features and diverse sustainability attributes across social, environmental, economic, and technical dimensions.
Implemented using Neo4j(2), the model enables interactive querying and reasoning over trade-offs, synergies, and influence pathways within AI system designs. Its flexibility allows stakeholders to define and analyze context-specific sustainability concerns—ranging from energy consumption and robustness to transparency, ethical governance, or long-term social impact.
The framework is empirically validated through two case studies: EMotivA, a persuasive emotion-aware system, and Perspective API, a multilingual content moderation service. Graph-theoretic metrics such as centrality and trade-off density are used to uncover design tensions and support informed decision-making.
Key contributions include: (i) a formalized and extensible SQ model adapted for AI contexts, (ii) a graph-driven framework for sustainability modeling and analysis, and (iii) empirical validation using real-world AI systems.