The World Health Organization (WHO) releases new guidance on the ethics and governance of large-scale multimodal models (LMMs). LMM is a type of rapidly growing generative artificial intelligence (AI) technology that has applications across healthcare.
This guidance outlines more than 40 recommendations for governments, technology companies, and healthcare providers to consider to ensure the appropriate use of LMM to promote and protect the health of the population.
LMMs can accept one or more types of data input, such as text, video, or images, and produce a variety of outputs that are not limited to the type of data input. LMMs are unique in that they can mimic human communication and perform tasks for which they are not explicitly programmed. LMMs will be adopted faster than any consumer application in history, with several platforms including ChatGPT, Bard, and Bert entering the public consciousness in 2023.
“Generative AI technologies have the potential to improve health care, but only if those developing, regulating and using these technologies identify and fully consider the associated risks,” said the WHO chief. says scientist Dr. Jeremy Farrar. “To achieve better health outcomes and overcome persistent health inequities, we need transparent information and policies to govern the design, development, and use of LMMs.”
Potential benefits and risks
The WHO's new guidance outlines five broad applications of LMM for health.
- Diagnostic and clinical care, including answering written patient questions.
- Used under patient guidance, including investigation of symptoms and treatment.
- Clerical and administrative tasks such as documenting and summarizing patient visits within the electronic medical record.
- Medical and nursing education, including providing trainees with simulated experiences with patients.
- Scientific research and drug development, including the identification of new compounds.
Although LMMs are beginning to be used for specific health-related purposes, there is also a documented risk of producing false, inaccurate, biased, or incomplete statements, which should be avoided when making health decisions. may cause harm to those who use the information. Additionally, LMMs can be trained with low-quality or biased data, regardless of race, ethnicity, ancestry, gender, gender identity, or age.
The guidance also details broader risks to the health system, including access to and affordability of the best-performing LMMs. LMMSs can also encourage “automation bias” by health professionals and patients, whereby errors that would otherwise be identified are overlooked or difficult choices are inappropriately delegated to his LMM. There is a possibility that LMMs, like other forms of AI, are vulnerable to cybersecurity risks that can jeopardize patient information, the reliability of these algorithms, and broader healthcare delivery.
To build safe and effective LMMs, WHO will engage a wide range of stakeholders, including governments, technology companies, health care providers, patients, and civil society, at all stages of the development and deployment of such technologies, including monitoring and regulation. emphasizes the need for stakeholder engagement.
“Governments of all countries must jointly lead efforts to effectively regulate the development and use of AI technologies such as LMM,” said Dr. Alain Labric, Director of Digital Health and Innovation, WHO Scientific Division. says.
Main recommendations
WHO's new guidance includes recommendations for governments, which have primary responsibility for developing and deploying LMMs and setting standards for their integration and use for public health and medical purposes. For example, governments should:
- Invest in or provide nonprofit or public infrastructure, including computing power and public datasets, that is accessible to developers in the public, private, and nonprofit sectors. This requires users to comply with ethical principles and values in exchange. access.
- Leverage laws, policies, and regulations to ensure that LMMs and applications used in healthcare and medicine, regardless of the risks and benefits associated with AI technologies, are ethically and Ensure that obligations and human rights standards are met. privacy.
- As resources permit, allocate existing or new regulatory authorities to evaluate and approve LMMs and applications for use in healthcare or medicine.
- When LMM is deployed at scale, mandatory post-release audits and impact assessments by independent third parties, including data protection and human rights, are implemented. Audits and impact assessments should be publicly available and include outcomes and impacts disaggregated by user type, such as age, race, and disability.
This guidance also includes the following key recommendations for LMM developers to ensure that:
- LMMs aren't just designed by scientists and engineers. Potential users and all direct and indirect stakeholders, including healthcare providers, scientific researchers, health professionals, and patients, should be encouraged to engage in structured, comprehensive, and transparent AI development from the earliest stages. They should be given the opportunity to be involved in a design, raise ethical issues, express concerns, and express their opinions. Provide input to the AI application under consideration.
- LMMs are designed to perform well-defined tasks with the required accuracy and reliability to improve the capacity of health systems and promote patient benefits. Developers also need to be able to anticipate and understand potential side effects.
Editor's note
The new AI Ethics and Governance document for health guidance on large-scale multimodal models builds on WHO guidance published in June 2021. Access the publication here.
