By Uzman Unis Bah
Geneva —The World Health Organization (WHO) has unveiled a comprehensive publication today, highlighting crucial regulatory considerations in the realm of artificial intelligence (AI) for health. The document underscores the significance of ensuring the safety and efficacy of AI systems, expeditiously providing these systems to those who require them, and fostering dialogue among various stakeholders, including developers, regulators, manufacturers, healthcare professionals, and patients.
In a landscape where health data has become increasingly available, and analytical techniques, including machine learning, logic-based approaches, and statistical methods, are rapidly evolving, AI tools hold the potential to revolutionize the healthcare sector. WHO acknowledges the transformative possibilities AI offers, from bolstering clinical trials to enhancing medical diagnosis, treatment, self-care, and person-centered care. Moreover, AI can complement the knowledge, skills, and competencies of healthcare professionals, proving particularly valuable in regions with limited access to medical specialists, such as in interpreting retinal scans and radiology images.
However, the rapid deployment of AI technologies, including large language models, sometimes occurs without a full understanding of their performance implications, potentially benefiting or harming end-users, including healthcare providers and patients. When dealing with health data, AI systems often have access to sensitive personal information, necessitating robust legal and regulatory frameworks to safeguard privacy, security, and data integrity—an objective at the heart of this newly released publication.
Dr. Tedros Adhanom Ghebreyesus, WHO Director-General, expressed, “Artificial intelligence holds great promise for health, but also comes with serious challenges, including unethical data collection, cybersecurity threats, and the risk of amplifying biases or misinformation. This new guidance will support countries in regulating AI effectively to harness its potential, whether in treating cancer or detecting tuberculosis, while minimizing the risks.”
Responding to the escalating need among nations to responsibly manage the rapid proliferation of AI health technologies, the publication outlines six essential areas for the regulation of AI in healthcare:
- Transparency and Documentation: The publication underscores the importance of transparency and documentation throughout the entire product lifecycle, including tracking development processes to foster trust.
- Risk Management: To manage risks effectively, issues like the “intended use,” “continuous learning,” human interventions, training models, and cybersecurity threats must be comprehensively addressed, with an emphasis on simplifying models whenever possible.
- External Data Validation: Clear articulation of the intended use of AI and external validation of data are vital for ensuring safety and facilitating regulation.
- Data Quality Commitment: Rigorous evaluation of systems pre-release is crucial to ensure that AI systems do not amplify biases and errors.
- Complex Regulation Challenges: The publication addresses the challenges posed by intricate regulations, such as the General Data Protection Regulation (GDPR) in Europe and the Health Insurance Portability and Accountability Act (HIPAA) in the United States, with a focus on understanding the scope of jurisdiction and consent requirements to protect privacy and data.
- Fostering Collaboration: Encouraging collaboration among regulatory bodies, patients, healthcare professionals, industry representatives, and government partners is essential to ensure that products and services remain compliant with regulations throughout their life cycles.
AI systems are intricate, relying not just on the code that powers them but also on the data they are trained on, often sourced from clinical settings and user interactions. One of the critical concerns addressed by these regulations is the risk of AI amplifying biases present in training data. For example, AI models may struggle to accurately represent the diversity of populations, leading to biases, inaccuracies, or even system failure. Regulations can mitigate these risks by ensuring that attributes such as gender, race, and ethnicity in the training data are reported and that datasets are intentionally made representative.
The new WHO publication aims to delineate key principles that governments and regulatory authorities can adopt for developing new guidance or adapting existing guidance concerning AI at national or regional levels.
For more information, please visit the official WHO website: https://www.who.int/news/item/19-10-2023-who-outlines-considerations-for-regulation-of-artificial-intelligence-for-health
