Open Source AI for Healthcare: From FLOSS, Open Data, and Open Science to AI Governance

IMIA Open Source Working Group Shinji Kobayashi Chair, IMIA Open Source Working Group, Gifu University https://creativecommons.org/licenses/by/4.0/

• https://creativecommons.org/licenses/by/4.0/

AeHIN Hour on FOSS · IMIA Open Source Working Group What does “open source” really mean? Use Study Modify Share for any purpose inspect how it works adapt to your needs with or without changes Rooted in the Free Software tradition, reframed by Open Source, and brought together here as FLOSS. Not free of charge · Not only source-available · Not only open weights

AeHIN Hour on FOSS Open source as a digital public good The UN recognises digital public goods — open software, data, AI models, standards, and content that protect privacy, do no harm, and advance the SDGs. DHIS2 OpenMRS OpenCRVS openIMIS GNU Health health information systems medical records civil registration health financing libre health system Open licensing is a foundation of every digital public good; for software, this means an OSI-approved open source license. Reviewed against the DPG Standard (9 indicators); DPGs can serve as a base layer for Digital Public Infrastracture.

AeHIN Hour on FOSS The IMIA Open Source Working Group - A neutral home for the four freedoms in health The IMIA Open Source Working Group, since 2002 — IMIA is the International Medical Informatics Association. Disseminate knowledge A neutral platform benefits of FLOSS in health for every stakeholder, equally Foster collaboration Lower the barriers connect projects to each other to real-world adoption Not a vendor, not a product — a meeting place.

AeHIN Hour on FOSS A trajectory: Open Source, Open Data, and Open Science The same four freedoms, aimed at a widening field — and at how open source, open data, open science, and AI interconnect. 2017 2018 2019 2020–21 2022 2024 Open collaboration Open data Open science & AI Pandemic & infodemic Resilience & equity Case series FLOSS as a model Privacy, security, and data governance open data + open tools + open research against black-box AI open source + open data + civic tech whole health systems field evidence Open Science is the frame; Open Data and Open Source are the two operational foundations of AI. Throughout: Medfloss.org — one of the largest catalogues of health open source.

AeHIN Hour on FOSS Why this matters for Asia & AeHIN Four stakes this region has raised for years. Equity Sovereignty run, adapt, and afford it on your own terms control of your data and your stack Digital public goods Capacity shared infrastructure, not rented dependence local engineers who study and modify, not just consume From consumers of AI to governors of AI Open source is one of the few answers that scales to all four.

AeHIN Hour on FOSS Does the yardstick survive AI? The same four freedoms, now aimed at systems that are part code, part weights, part data. What ‘open’ must mean for AI Where it breaks The four freedoms + Data Information + Code + Open weights ≠ Open Source AI Parameters Weights alone are not enough. Open Source AI Definition (OSAID) 1.0 — OSI, 2024 · a useful starting point, not a medical safety standard Risk: open-washing We saw it coming, in 2019, Paton & Kobayashi argued for an open-science approach to AI, against black-box systems.

https://www.unopensource.org/

• https://www.unopensource.org/