· 5. Non Discrimination

Discrimination concerns the variability of AI results between individuals or groups of people based on the exploitation of differences in their characteristics that can be considered either intentionally or unintentionally (such as ethnicity, gender, sexual orientation or age), which may negatively impact such individuals or groups. Direct or indirect discrimination through the use of AI can serve to exploit prejudice and marginalise certain groups. Those in control of algorithms may intentionally try to achieve unfair, discriminatory, or biased outcomes in order to exclude certain groups of persons. Intentional harm can, for instance, be achieved by explicit manipulation of the data to exclude certain groups. Harm may also result from exploitation of consumer biases or unfair competition, such as homogenisation of prices by means of collusion or non transparent market. Discrimination in an AI context can occur unintentionally due to, for example, problems with data such as bias, incompleteness and bad governance models. Machine learning algorithms identify patterns or regularities in data, and will therefore also follow the patterns resulting from biased and or incomplete data sets. An incomplete data set may not reflect the target group it is intended to represent. While it might be possible to remove clearly identifiable and unwanted bias when collecting data, data always carries some kind of bias. Therefore, the upstream identification of possible bias, which later can be rectified, is important to build in to the development of AI. Moreover, it is important to acknowledge that AI technology can be employed to identify this inherent bias, and hence to support awareness training on our own inherent bias. Accordingly, it can also assist us in making less biased decisions.
Principle: Draft Ethics Guidelines for Trustworthy AI, Dec 18, 2018

Published by The European Commission’s High-Level Expert Group on Artificial Intelligence

Related Principles

V. Diversity, non discrimination and fairness

Data sets used by AI systems (both for training and operation) may suffer from the inclusion of inadvertent historic bias, incompleteness and bad governance models. The continuation of such biases could lead to (in)direct discrimination. Harm can also result from the intentional exploitation of (consumer) biases or by engaging in unfair competition. Moreover, the way in which AI systems are developed (e.g. the way in which the programming code of an algorithm is written) may also suffer from bias. Such concerns should be tackled from the beginning of the system’ development. Establishing diverse design teams and setting up mechanisms ensuring participation, in particular of citizens, in AI development can also help to address these concerns. It is advisable to consult stakeholders who may directly or indirectly be affected by the system throughout its life cycle. AI systems should consider the whole range of human abilities, skills and requirements, and ensure accessibility through a universal design approach to strive to achieve equal access for persons with disabilities.

Published by European Commission in Key requirements for trustworthy AI, Apr 8, 2019

· 2. The Principle of Non maleficence: “Do no Harm”

AI systems should not harm human beings. By design, AI systems should protect the dignity, integrity, liberty, privacy, safety, and security of human beings in society and at work. AI systems should not threaten the democratic process, freedom of expression, freedoms of identify, or the possibility to refuse AI services. At the very least, AI systems should not be designed in a way that enhances existing harms or creates new harms for individuals. Harms can be physical, psychological, financial or social. AI specific harms may stem from the treatment of data on individuals (i.e. how it is collected, stored, used, etc.). To avoid harm, data collected and used for training of AI algorithms must be done in a way that avoids discrimination, manipulation, or negative profiling. Of equal importance, AI systems should be developed and implemented in a way that protects societies from ideological polarization and algorithmic determinism. Vulnerable demographics (e.g. children, minorities, disabled persons, elderly persons, or immigrants) should receive greater attention to the prevention of harm, given their unique status in society. Inclusion and diversity are key ingredients for the prevention of harm to ensure suitability of these systems across cultures, genders, ages, life choices, etc. Therefore not only should AI be designed with the impact on various vulnerable demographics in mind but the above mentioned demographics should have a place in the design process (rather through testing, validating, or other). Avoiding harm may also be viewed in terms of harm to the environment and animals, thus the development of environmentally friendly AI may be considered part of the principle of avoiding harm. The Earth’s resources can be valued in and of themselves or as a resource for humans to consume. In either case it is necessary to ensure that the research, development, and use of AI are done with an eye towards environmental awareness.

Published by The European Commission’s High-Level Expert Group on Artificial Intelligence in Draft Ethics Guidelines for Trustworthy AI, Dec 18, 2018

· 8. Robustness

Trustworthy AI requires that algorithms are secure, reliable as well as robust enough to deal with errors or inconsistencies during the design, development, execution, deployment and use phase of the AI system, and to adequately cope with erroneous outcomes. Reliability & Reproducibility. Trustworthiness requires that the accuracy of results can be confirmed and reproduced by independent evaluation. However, the complexity, non determinism and opacity of many AI systems, together with sensitivity to training model building conditions, can make it difficult to reproduce results. Currently there is an increased awareness within the AI research community that reproducibility is a critical requirement in the field. Reproducibility is essential to guarantee that results are consistent across different situations, computational frameworks and input data. The lack of reproducibility can lead to unintended discrimination in AI decisions. Accuracy. Accuracy pertains to an AI’s confidence and ability to correctly classify information into the correct categories, or its ability to make correct predictions, recommendations, or decisions based on data or models. An explicit and well formed development and evaluation process can support, mitigate and correct unintended risks. Resilience to Attack. AI systems, like all software systems, can include vulnerabilities that can allow them to be exploited by adversaries. Hacking is an important case of intentional harm, by which the system will purposefully follow a different course of action than its original purpose. If an AI system is attacked, the data as well as system behaviour can be changed, leading the system to make different decisions, or causing the system to shut down altogether. Systems and or data can also become corrupted, by malicious intention or by exposure to unexpected situations. Poor governance, by which it becomes possible to intentionally or unintentionally tamper with the data, or grant access to the algorithms to unauthorised entities, can also result in discrimination, erroneous decisions, or even physical harm. Fall back plan. A secure AI has safeguards that enable a fall back plan in case of problems with the AI system. In some cases this can mean that the AI system switches from statistical to rule based procedure, in other cases it means that the system asks for a human operator before continuing the action.

Published by The European Commission’s High-Level Expert Group on Artificial Intelligence in Draft Ethics Guidelines for Trustworthy AI, Dec 18, 2018

8. Principle of fairness

AI service providers, business users, and data providers should take into consideration that individuals will not be discriminated unfairly by the judgments of AI systems or AI services. [Main points to discuss] A) Attention to the representativeness of data used for learning or other methods of AI AI service providers, business users, and data providers may be expected to pay attention to the representativeness of data used for learning or other methods of AI and the social bias inherent in the data so that individuals should not be unfairly discriminated against due to their race, religion, gender, etc. as a result of the judgment of AI. In light of the characteristics of the technologies to be used and their usage, in what cases and to what extent is attention expected to be paid to the representativeness of data used for learning or other methods and the social bias inherent in the data? Note: The representativeness of data refers to the fact that data sampled and used do not distort the propensity of the population of data. B) Attention to unfair discrimination by algorithm AI service providers and business users may be expected to pay attention to the possibility that individuals may be unfairly discriminated against due to their race, religion, gender, etc. by the algorithm of AI. C) Human intervention Regarding the judgment made by AI, AI service providers and business users may be expected to make judgments as to whether to use the judgments of AI, how to use it, or other matters, with consideration of social contexts and reasonable expectations of people in the utilization of AI, so that individuals should not be unfairly discriminated against due to their race, religion, gender, etc. In light of the characteristics of the technologies to be used and their usage, in what cases and to what extent is human intervention expected?

Published by Ministry of Internal Affairs and Communications (MIC), the Government of Japan in Draft AI Utilization Principles, Jul 17, 2018

5 Ensure inclusiveness and equity

Inclusiveness requires that AI used in health care is designed to encourage the widest possible appropriate, equitable use and access, irrespective of age, gender, income, ability or other characteristics. Institutions (e.g. companies, regulatory agencies, health systems) should hire employees from diverse backgrounds, cultures and disciplines to develop, monitor and deploy AI. AI technologies should be designed by and evaluated with the active participation of those who are required to use the system or will be affected by it, including providers and patients, and such participants should be sufficiently diverse. Participation can also be improved by adopting open source software or making source codes publicly available. AI technology – like any other technology – should be shared as widely as possible. AI technologies should be available not only in HIC and for use in contexts and for needs that apply to high income settings but they should also be adaptable to the types of devices, telecommunications infrastructure and data transfer capacity in LMIC. AI developers and vendors should also consider the diversity of languages, ability and forms of communication around the world to avoid barriers to use. Industry and governments should strive to ensure that the “digital divide” within and between countries is not widened and ensure equitable access to novel AI technologies. AI technologies should not be biased. Bias is a threat to inclusiveness and equity because it represents a departure, often arbitrary, from equal treatment. For example, a system designed to diagnose cancerous skin lesions that is trained with data on one skin colour may not generate accurate results for patients with a different skin colour, increasing the risk to their health. Unintended biases that may emerge with AI should be avoided or identified and mitigated. AI developers should be aware of the possible biases in their design, implementation and use and the potential harm that biases can cause to individuals and society. These parties also have a duty to address potential bias and avoid introducing or exacerbating health care disparities, including when testing or deploying new AI technologies in vulnerable populations. AI developers should ensure that AI data, and especially training data, do not include sampling bias and are therefore accurate, complete and diverse. If a particular racial or ethnic minority (or other group) is underrepresented in a dataset, oversampling of that group relative to its population size may be necessary to ensure that an AI technology achieves the same quality of results in that population as in better represented groups. AI technologies should minimize inevitable power disparities between providers and patients or between companies that create and deploy AI technologies and those that use or rely on them. Public sector agencies should have control over the data collectedby private health care providers, and their shared responsibilities should be defined and respected. Everyone – patients, health care providers and health care systems – should be able to benefit from an AI technology and not just the technology providers. AI technologies should be accompanied by means to provide patients with knowledge and skills to better understand their health status and to communicate effectively with health care providers. Future health literacy should include an element of information technology literacy. The effects of use of AI technologies must be monitored and evaluated, including disproportionate effects on specific groups of people when they mirror or exacerbate existing forms of bias and discrimination. Special provision should be made to protect the rights and welfare of vulnerable persons, with mechanisms for redress if such bias and discrimination emerges or is alleged.

Published by World Health Organization (WHO) in Key ethical principles for use of artificial intelligence for health, Jun 28, 2021