Health: WHO publishes guide for the use of AI

Artificial intelligence can transform medical treatments, but without an understanding of how it works, rapid deployment of digital health programs could harm patients, the WHO warned on Thursday (October 19).

“With the increased availability of health care data, and rapid advances in analytical techniques – machine learning capabilities of computers, based on logic or statistics – AI can transform the health sector,” he said. indicated the World Health Organization in this document devoted to the use of AI, intended to provide guidance to public authorities.

According to the WHO, AI can improve performance in clinical trials, diagnosis, development of a treatment protocol, and complement medical knowledge and skills. For example, AI is useful in the absence of specialists, in the field of radiology for the interpretation of medical imaging and retinal images. However, AI is being deployed rapidly, sometimes without adequate understanding of how these technologies work, "which can either benefit or harm users", whether patients or professionals, WHO warned .

» READ ALSO - Artificial intelligence: Google's promise to its users

Artificial intelligence systems applied to health allow access to personal data, which is why a solid legal framework is necessary to safeguard privacy, underlined the WHO. “Artificial intelligence holds great promise for health, but also presents serious challenges, including unethical data collection, threats to cybersecurity, and the amplification of bias and misinformation,” said the WHO Director-General Tedros Adhanom Ghebreyesus. “This new guide will help countries to effectively regulate AI, to exploit its potential, whether it is to treat cancer or detect tuberculosis, while minimizing the risks,” he said. he adds.

According to the WHO, artificial intelligence systems rely on the training data they use to learn, so better regulation can help manage the risks and dangers of bias present in existing data. amplified by AI. “For example, it can be difficult for AI models to accurately represent the diversity of populations, leading to bias, inaccuracies and even failures,” the WHO said.

“To help reduce these risks, regulations can be used to ensure that attributes, such as gender, race and ethnicity of individuals in training data are reported and that all data is intentionally designed to be representative,” the WHO added. The WHO recommends in its guide external validation of data, evaluation of systems to avoid errors and biases, consent for the collection of private data, and collaboration between regulators, patients, governments and health professionals .