The World Health Organization (WHO) has released comprehensive guidance on the ethics and governance of Large Multi-Modal Models (LMMs), a rapidly advancing form of generative artificial intelligence (AI) with diverse applications in healthcare. With over 40 recommendations, the guidance is directed at governments, technology companies, and healthcare providers to ensure responsible and beneficial use of LMMs for the health of populations.
LMMs, capable of processing various data inputs and generating diverse outputs, have witnessed rapid adoption, including platforms like ChatGPT, Bard, and Bert. The WHO emphasizes the potential benefits of LMMs in healthcare, such as diagnosis, patient-guided use, administrative tasks, medical education, and scientific research. However, it also acknowledges risks, including the generation of inaccurate or biased information and broader challenges like accessibility, affordability, and cybersecurity risks.
Key Recommendations from WHO:
- Investment in Infrastructure: Governments should invest in or provide not-for-profit/public infrastructure, including computing power and data sets, accessible to developers. Access should be contingent on adherence to ethical principles.
- Regulatory Oversight: Implement laws, policies, and regulations to ensure ethical obligations and human rights standards for LMMs in healthcare. Assign regulatory agencies for assessment and approval of LMMs.
- Post-Release Auditing: Introduce mandatory post-release auditing and impact assessments by independent third parties. These assessments should include outcomes and impacts disaggregated by user type.
- Inclusive Design: Engage potential users, stakeholders, and experts from various fields in the early stages of LMM development. Encourage transparent, inclusive, and structured design to address ethical concerns.
- Task Performance: Ensure LMMs are designed for well-defined tasks with necessary accuracy and reliability. Developers should anticipate and understand potential secondary outcomes.
The WHO underscores the need for collaboration among stakeholders—governments, technology companies, healthcare providers, patients, and civil society—to facilitate the safe and effective development and deployment of LMMs, with a focus on ethical considerations, human rights, and the betterment of health outcomes.