Papers by Ilma A Fiddien
ACM CSCW 2024 Workshop - From Stem to Stern: Contestability Along AI Value Chains, 2024
As Indonesia pursues an ambitious national strategy for artificial intelligence (AI), ensuring th... more As Indonesia pursues an ambitious national strategy for artificial intelligence (AI), ensuring the development of contestable and accountable AI systems is crucial. This paper critically examines the current state of AI contestability in Indonesia's policy landscape, identifying challenges, opportunities, and recommendations. Through document analysis, comparative review, and application of the Contestable AI by Design framework, we find that while Indonesia's AI strategy shows promising intent, gaps remain in institutionalizing contestability principles. To address these gaps, we propose a policy roadmap spanning regulatory reforms, public empowerment, and cross-sectoral partnerships. Key recommendations include mandating Algorithmic Impact Assessments, establishing AI ombudsperson offices, and fostering multi-stakeholder collaboration. We argue that by embedding contestability safeguards and adaptive governance, Indonesia can foster an inclusive and socially responsive AI ecosystem. This paper advances the research agenda on AI governance in the Global South. It offers policymakers and practitioners diagnostic tools and pathways for aligning Indonesia's AI ambitions with democratic accountability, contributing to the transnational dialogue on ethical AI.
Thesis Chapters by Ilma A Fiddien
This dissertation investigates the adherence to clear-writing principles in paraphrased sentences... more This dissertation investigates the adherence to clear-writing principles in paraphrased sentences within the scientific domain. Leveraging a combination of automated systems and human surveys, the study explores the connection between writing principles, clarity, and reader preferences. A total of 740 sentence pairs were annotated with span labels and adherence flags to a set of clear-writing principles, leading to the development of the CLEARMETRICS system. This system achieved F1-scores of 74.9% and 71.7% in detecting spans (subject, verb, character, and action) in scientific and journalistic sentences, respectively, and 60.2% and 44.7% in flagging inadherence toward the specific writing rules. A human survey evaluated the effectiveness of the clear-writing rules, identifying strong preferences for certain rules, while others had minimal impact. This research revealed common challenges and offered pathways for future enhancements, contributing insights to the creation of writing tools and guidelines to enhance the clarity of scientific writing.
Uploads
Papers by Ilma A Fiddien
Thesis Chapters by Ilma A Fiddien