Towards Algorithmic Reform: Ethical Values-Informed Tool Design and Inclusive AI/ML Literacy
Salvato in:
| Pubblicato in: | ProQuest Dissertations and Theses (2025) |
|---|---|
| Autore principale: | |
| Pubblicazione: |
ProQuest Dissertations & Theses
|
| Soggetti: | |
| Accesso online: | Citation/Abstract Full Text - PDF |
| Tags: |
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| Abstract: | Artificial Intelligence (AI) and Machine Learning (ML) systems deployed in public housing allocation accelerate service delivery in the U.S. These systems also pose unprecedented harms to the most vulnerable, many of whom lack the relevant AI/ML knowledge to seek recourse when impacted by algorithmic harms, discrimination and injustices. Oftentimes, these systems are deployed without multi-stakeholder evaluation, and considerations of ethical values such as transparency and agency that are critical to human rights in AI/ML. To actively resist algorithmic harms and promote human rights in algorithmic decision making, this thesis proposes algorithmic reform. Algorithmic reform describes the multifaceted and multidisciplinary endeavor necessary for ensuring and upholding human rights in AI/ML for all. It builds on several key approaches (e.g., ethical system design, AI literacy, algorithmic fairness, fair policy, and algorithm design) and disciplines (e.g., CS, social science, ethics, and philosophy). This thesis particularly uses HCI mixed-methods research methods to explore two specific initiatives that are essential in advancing algorithmic reform: (1) designing transparency and autonomy ethical values informed algorithmic risk assessments to foster trust, agency and positivity; (2) contributing in inclusive AI/ML literacy to build an AI/ML aware society who can meaningfully engage in AI/ML deployment and impacts conversations.Algorithmic risk assessments rely on risk factors or the causes of homelessness to determine service deservingness. It neglects the clients’ strengths that help them thrive. This deficit lens perpetuates stigma, erodes clients’ trust in service staff. Most often, there is also a lack of transparency about what factors are accounted for in decision-making and why. This, in turn, further exacerbates clients’ trust, leaves them dissatisfied, deters future service engagement, and pushes them away from deserved resources. Studies suggest that trust is essential for both clients and service staff, as it fosters building a working relationship. This research interrogates these challenges in two ways towards advancing algorithmic reform. First, designing valuesensitive algorithmic assessments that uphold ethical values (i.e., transparency, and autonomy) — operationalized through strengths, different strength ratings, and strengths explanations. Findings from multi-stakeholder investigation show that the new designs were perceived as positive, uplifting and human. Participants appreciated strengths explanations, however, responses varied concerning rating preferences and algorithmic scoring. Second, contesting and questioning algorithmic harms, bias and injustices requires AI/ML literacy. On that note, I developed inclusive AI/ML literacy approaches that teaches participants how to question on transparency of algorithmic systems as their human right using question-based learning; in another study, participants learn about the invisible and socio-technical nature of ML biases through playing transformational game. Findings from exploratory case studies show the effectiveness, and accessibility of these AI/ML literacy approaches. This thesis addresses the following overarching research question: How can we establish algorithmic reform through the ethical values-informed tool design and inclusive AI/ML literacy? The findings of this thesis contribute toward human-rights centered discourses in HCI, human-centered AI (HCAI), AI/ML literacy and ethics, and potentially inform future public service policy and praxis.  |
|---|---|
| ISBN: | 9798290941868 |
| Fonte: | ProQuest Dissertations & Theses Global |