As noted in the section on core human rights issues across public governance sectors, the ECHR and the ESC prohibit discrimination.[1] Under Article 3 of the Oviedo Convention, State Parties are required to take appropriate measures with a view to providing, within their jurisdiction, equitable access to health care of appropriate quality.[2]

Unwanted biases in the data used to develop AI systems may skew the assessment of health needs and treatments for patients and thereby perpetuate or exacerbate existing biases. It is notable that AI models trained predominantly on data from specific populations may misdiagnose conditions or underestimate illness severity in underrepresented groups such as women and girls, persons belonging to ethnic minorities, indigenous populations, the elderly or persons with disabilities.[3] Examples include prioritisation systems for kidney transplants, where biased historical data skewed outcomes against some patients.[4] Similarly, inadequate representation in training datasets has led to misdiagnoses of skin conditions.[5] In addition, there is concern that access to the benefits offered by AI in healthcare may not be equally available to all. The deployment of such care may be geographically uneven across a given country, or dependent on the financial means of the patients.[6] A lack of accessible design of AI applications may exclude older persons or persons with disabilities. States should adopt measures to ensure AI systems are developed and deployed equitably.

 

[2] See also Articles 15 §1(b) and 2§2 of the International Covenant on Economic, Social and Cultural Rights (ICESCR) on the right of everyone to enjoy the benefits of scientific progress and its applications, without discrimination of any kind.

[3] See, e.g., CDBIO Report p. 26; see also WHO, Ethics and governance of artificial intelligence for health (2021), pp. 54-57. Further on the underrepresentation and low quality of data of women, as well as gender diverse persons in scientific research, the GEC/CDADI Study, p. 25. Also (p. 26) on the structural discrimination embedded in AI systems with respect to systematically disadvantaged patients with ethnic minority backgrounds. Furthermore, see WHO, Ageism in artificial intelligence for health (2022), showing that algorithmic systems used in the healthcare sector are trained on the data of predominantly younger populations, leading to disproportionately lower performance of these systems for older patients, including incorrect diagnosis.

[6] CDBIO Report, p. 26. On the discussion on the possibility that the existing digital divide (including with respect to AI) and inequalities (within and between countries, as well as societal groups) will exacerbate the unequal distribution of healthcare and problems of effective access to healthcare, see PACE Recommendation 2185 (2020), Artificial intelligence in healthcare: medical, legal and ethical challenges ahead. An additional concern could be linked to the use of AI for resource allocation and case prioritisation.