Exploring the potential integration of Digital Humanism within Digital SocialInnovations (DSIs), this research focuses on the responsible application of LargeLanguage Models (LLMs) to enhance user autonomy and self-determination. WhileDigital Humanism emphasizes protecting human values and autonomy in digitaltechnology design, a paradox exists where systems intended to enhance capabilitiescan inadvertently undermine human agency. This research investigates how tosupport human capacities through the deployment of LLMs. In this thesis, themethodological approach employed integrates the multidisciplinary orientation ofsocial innovation research and argumentative principles rooted in pragmaticjustification through collective discourse, drawing inspiration from Rorty'sphilosophy (cited in Philosophy Overdose; 2023). Design research provides theprimary methodological structure, with interactive research serving as the basis forquality control. This strategy begins with reviewing existing literature to establishthe problem and potential solution space. Subsequently, the proposed innovationundergoes a process of iterative refinement and construction, informed by collectivecriticism and building, ultimately seeking validation through its demonstratedutility through the emergent consensus within the relevant community. BehavioralEconomics insights were incorporated to analyze cognitive biases and theirinfluence on interactions with AI systems. Insights gained from interpretation andsynthesis revealed a tension between the streamlined utility often prioritized indigital platforms and the value of diverse perspectives. Challenges identifiedinclude the potential for current AI development, especially concerning LLMs, tocreate innovation barriers due to dominant actors and proprietary models. Concernsalso arose regarding the difficulty in ensuring the reliability of LLMs and possibleconsequences. With the purpose of exploring, understanding, and proposing adefensible solution, a path forward is presented involving decentralized,community-governed AI systems, inspired by Elinor Ostrom's work on commonpool resources. Such system is designed to integrate localized values andknowledge, and ethically advance behavioral insights to promote well-being andinformed decision-making. This approach aims to counteract homogenization,emphasizing the fundamental contribution of continuous, diverse human-authoredcontent for maintaining epistemic vitality. While outlining this conceptual approach, the research acknowledges remaining questions regarding practical implementationand scaling.