Aligning AI Systems with Human Values

Human values shape what stakeholders need, expect, and trust from AI systems. These values, such as sustainability, fairness, or privacy, often differ between groups and can conflict in ways that traditional requirements engineering struggles to address. When systems are designed without recognizing these differences, the result can be miscommunication, misalignment, and loss of trust.Our approach centers on two key principles: 1) Identifying human values is essential for developing AI systems that align with user expectations and needs. This involves prioritizing these values and resolving conflicts to ensure AI systems are ethically sound, and 2) Employing large language models to uncover and analyze these values allows us to leverage broad knowledge bases.Rather than replacing human judgment, LLMs serve as tools to scale and support value elicitation and analysis. For example, they can be used to generate (AI-)personas, extract concerns from interviews, or surface value conflicts in stakeholder feedback. This helps ensure underrepresented voices are not overlooked during early stages of design. Our goal is to enable AI systems that are not only technically robust but also socially responsible, trusted, and inclusive.