Vernacularizing Taxonomies of Harm is Essential for Operationalizing Holistic AI Safety
Project Overview
The document explores the role of generative AI in education, emphasizing the necessity of tailoring AI safety principles to specific educational contexts. It argues that AI systems are sociotechnical, meaning their impacts—both positive and negative—are shaped by societal norms and values. The authors advocate for a participatory approach to AI safety, which involves engaging local stakeholders to address unique challenges and ethical dilemmas within educational environments. By vernacularizing taxonomies of harm, the document highlights the importance of context-specific adaptations to avoid unintended consequences and to enhance the effectiveness of AI applications in education. The findings suggest that successful implementation of generative AI in educational settings requires an understanding of local cultural dynamics and ethical considerations, ensuring that AI tools are beneficial and aligned with the values of the communities they serve.
Key Applications
Google Deepmind’s LearnLM
Context: AI-supported learning tasks in educational settings, targeting students and educators.
Implementation: Developed using participatory methodologies as a testbed for holistic evaluation methods.
Outcomes: Supports learning tasks while promoting ethical and safety considerations.
Challenges: Requires a shift from traditional 'translation' to 'vernacularization' of AI ethics and safety principles.
Implementation Barriers
Operational Challenges
Difficulties in operationalizing general taxonomies due to structural obstacles and lack of trained personnel.
Proposed Solutions: Incorporating participatory design methodologies and ensuring local actors are engaged in the vernacularization process.
Resource Limitations
Under-resourced educational sectors may struggle with the costs and efforts required for vernacularization.
Proposed Solutions: AI actors need to allocate sufficient resources for research and community engagement.
Project Team
Wm. Matthew Kennedy
Researcher
Daniel Vargas Campos
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Wm. Matthew Kennedy, Daniel Vargas Campos
Source Publication: View Original PaperLink opens in a new window
Project Contact: Dr. Jianhua Yang
LLM Model Version: gpt-4o-mini-2024-07-18
Analysis Provider: Openai