Poster ID
P-33
Poster Title
From Staff AI Practice to Genomic AI Governance: An Institute Case with Reference to GA4GH AIWS Principles
Authors
Ricky W.F. Man¹, J.T.J. van Lunenburg¹, Crystal K.L. Tik¹, Venus L.Y. Yung¹, Don C.K. Tai¹, Mullin H.C. Yu¹, Brian H.Y. Chung¹,²
Affiliations
¹ Hong Kong Genome Institute, Hong Kong SAR, China
² Department of Pediatrics and Adolescent Medicine, School of Clinical Medicine, Li Ka Shing Faculty of Medicine, The University of Hong Kong, Hong Kong SAR, China"
Affiliations
¹ Hong Kong Genome Institute, Hong Kong SAR, China
² Department of Pediatrics and Adolescent Medicine, School of Clinical Medicine, Li Ka Shing Faculty of Medicine, The University of Hong Kong, Hong Kong SAR, China"
Abstract
Background: The Global Alliance for Genomics and Health (GA4GH) Artificial Intelligence Work Stream (AIWS) outlines core principles—transparency, replicability, governance/ethics, and effectiveness—for responsible AI in genomics. At the Hong Kong Genome Institute (HKGI), staff AI use grew ahead of formal policy, creating both opportunity and risk for scientific AI systems. HKGI therefore engaged an external consultant for AI assessment and governance-plan work to build policy foundations, and is developing scientific-track policy with AIWS as an international reference.
Method: In July–August 2026, HKGI ran an institute-wide staff baseline survey across all units (107 valid responses; 99.1% of 108 target staff). As diagnostic discovery rather than a maturity audit, it assessed adoption, tool types (public generative AI, embedded assistants, coding tools, local models), and practices in a scientific cohort (n=36) across genome curation, lab operations, and bioinformatics, plus approval awareness, data-input concerns, and human verification habits. Findings informed a two-track policy: workplace productivity enablement and risk-gated genomic/clinical AI governance.
Result: AI adoption reached 85%, above recent Hong Kong workplace estimates of ~73% weekly/daily use (HKUST Business School, 2026). Strengths include active scientific use for coding, genomic data processing, literature synthesis, and analysis support; strong self-reported output validation; and AI CURA, an automated LLM workflow for genetic variant classification (Ma et al., Sci. Transl. Med., 2026). InsightQL, an LLM platform that processes patient data from clinical and curation systems to support curator review, is under controlled development. A residual gap—78% of active users on public/personal tools, unapproved setups, or unsure approval status—mirrors wider Hong Kong shadow-AI risks. To manage this, the consultant-supported program delivered AI governance policy artefacts, Responsible Use guidance, AI Impact Assessment (AIA), risk-gating, training plans, approved-tool direction, and secure local model infrastructure for sensitive clinical workflows.
Conclusion: HKGI provides an institute model where empirical baseline data, scientific AI delivery, and GA4GH AIWS-guided governance jointly support responsible genomic AI adoption.
Method: In July–August 2026, HKGI ran an institute-wide staff baseline survey across all units (107 valid responses; 99.1% of 108 target staff). As diagnostic discovery rather than a maturity audit, it assessed adoption, tool types (public generative AI, embedded assistants, coding tools, local models), and practices in a scientific cohort (n=36) across genome curation, lab operations, and bioinformatics, plus approval awareness, data-input concerns, and human verification habits. Findings informed a two-track policy: workplace productivity enablement and risk-gated genomic/clinical AI governance.
Result: AI adoption reached 85%, above recent Hong Kong workplace estimates of ~73% weekly/daily use (HKUST Business School, 2026). Strengths include active scientific use for coding, genomic data processing, literature synthesis, and analysis support; strong self-reported output validation; and AI CURA, an automated LLM workflow for genetic variant classification (Ma et al., Sci. Transl. Med., 2026). InsightQL, an LLM platform that processes patient data from clinical and curation systems to support curator review, is under controlled development. A residual gap—78% of active users on public/personal tools, unapproved setups, or unsure approval status—mirrors wider Hong Kong shadow-AI risks. To manage this, the consultant-supported program delivered AI governance policy artefacts, Responsible Use guidance, AI Impact Assessment (AIA), risk-gating, training plans, approved-tool direction, and secure local model infrastructure for sensitive clinical workflows.
Conclusion: HKGI provides an institute model where empirical baseline data, scientific AI delivery, and GA4GH AIWS-guided governance jointly support responsible genomic AI adoption.