The Toronto Declaration: Protecting the rights to equality and non-discrimination in machine learning systems was launched on May 16, 2018 at RightsCon Toronto.
The preamble of the Declaration can be found below and the full text is available here.
At the time of the launch, the Declaration prepared by Amnesty International and Access Now and it has been endorsed by Human Rights Watch and Wikimedia Foundation.
As machine learning systems advance in capability and increase in use, we must examine the positive and negative implications of these technologies. We acknowledge the potential for these technologies to be used for good and to promote human rights but also the potential to intentionally or inadvertently discriminate against individuals or groups of people. We must keep our focus on how these technologies will affect individual human beings and human rights. In a world of machine learning systems, who will bear accountability for harming human rights?
As the “ethics” discourse gains ground, this Declaration aims to underline the centrality of the universal, binding and actionable body of human rights law and standards, which protect rights and provide a well-developed framework for remedies. They protect individuals against discrimination, promote inclusion, diversity and equity, and safeguards equality. Human rights are “universal, indivisible and interdependent and interrelated.”
This Declaration aims to build on existing discussions, principles and papers exploring the harms arising from this technology. The significant work done in this area by many experts has helped raise awareness about and inform discussions about the discriminatory risks of machine learning systems. We wish to complement this work by reaffirming the role of human rights law and standards in protecting individuals and groups from discrimination and non-equality in any context. The human rights law and standards outlined in this Declaration provide a solid grounding for the developing ethical frameworks for machine learning.
From policing, to welfare systems, online discourse, and healthcare – to name a few examples – systems employing machine learning technologies can vastly and rapidly change or reinforce power structures or inequalities on an unprecedented scale and with significant harm to human rights. There is a substantive and growing body of evidence to show that machine learning systems, which can be opaque and include unexplainable processes, can easily contribute to discriminatory or otherwise repressive practices if adopted without necessary safeguards.
States and private actors should promote the development and use of these technologies to help people more easily exercise and enjoy their human rights. For example, in healthcare, machine learning systems could bring advances in diagnostics and treatments, while potentially making health services more widely available and accessible. States and private actors should further, in relation to machine learning and artificial intelligence more broadly, promote the positive right to the enjoyment of the benefits of scientific progress and its applications as an affirmation of economic, social and cultural rights.
The rights to equality and non-discrimination are only two of the human rights that may be adversely affected through the use of machine learning systems: privacy, data protection, freedom of expression, participation in cultural life, equality before the law, and meaningful access to remedy are just some of the other rights that may be harmed with the misuse of this technology. Systems that make decisions and process data can also implicate economic, social, and cultural rights; for example, they can impact the provision of services and opportunities such as healthcare and education, and access to opportunities, such as labour and employment.
Whilst this Declaration is focused on machine learning technologies, many of the norms and principles included are equally applicable to artificial intelligence more widely, as well as to related data systems. The declaration focuses on the rights to equality and non-discrimination. Machine learning, and artificial intelligence more broadly, impact a wider array of human rights, such as the right to privacy, the right to freedom of expression, participation in cultural life, the right to remedy, and the right to life.
Using the framework of international human rights law
- States have obligations to promote, protect and respect human rights; private sector, including companies, has a responsibility to respect human rights at all times. We put forward this Declaration to affirm these obligations and responsibilities.
There are many discussions taking place now at supranational, state and regional level, in technology companies, at academic institutions, in civil society and beyond, focussing on how to make AI human-centric and the “ethics” of artificial intelligence. There is need to consider current and future potential human rights infringements, and how best to address them with better thinking about harm to rights, and regulatory and legal regimes.
Human rights law is a universally ascribed system of values based on the rule of law which provides established means to ensure that rights, including the rights to equality and non-discrimination, are upheld. Its nature as a universally binding, actionable set of standards is particularly well-suited for borderless technologies such as machine learning. Human rights law provides both standards and mechanisms to hold the public and private sectors accountable where they fail to fulfil their respective obligations and responsibilities to protect and respect rights. It also requires that everyone must be able to obtain an effective remedy and redress where their rights have been denied or violated.The risks machine learning systems pose must be urgently examined and addressed at governmental level and by the private sector conceiving, developing and, deploying these systems. Government measures should be binding and adequate to protect and promote rights. Academic, legal and civil society experts should be able to meaningfully participate in these discussions, critique and advise on the use of these technologies. It is also critical that potential harms are identified and addressed and that mechanisms are put in place to hold accountable those responsible for harms.
The rights to equality and non-discrimination
- This Declaration focuses on the rights to equality and non-discrimination, critical principles underpinning all human rights.
Discrimination is defined under international law as “any distinction, exclusion, restriction or preference which is based on any ground such as race, colour, sex, language, religion, political or other opinion, national or social origin, property, birth or other status, and which has the purpose or effect of nullifying or impairing the recognition, enjoyment or exercise by all persons, on an equal footing, of all rights and freedoms.” This list is non-exhaustive as the United Nations High Commissioner for Human Rights has recognized the necessity of preventing discrimination against additional classes.
- The public and the private sector have obligations and responsibilities under human rights law to proactively prevent discrimination. When prevention is not sufficient or satisfactory, discrimination should be mitigated.
In employing new technologies, both the public and the private sector will likely need to find new ways to protect human rights, as new challenges to equality and representation of diverse individuals and groups arise. These types of technologies can exacerbate discrimination at scale.
Existing patterns of structural discrimination may be reproduced and aggravated in situations that are particular to these technologies – for example, machine learning system goals that create self-fulfilling markers of success and reinforce patterns of inequality, or issues arising from using non-representative or “biased” datasets.
All actors, public and private, must prevent and mitigate discrimination risks in the design, development and, application of machine learning technologies and that ensure that effective remedies are in place before deployment and throughout the lifecycle of these systems.
Protecting the rights of all individuals and groups and promoting diversity and inclusion diversity
This Declaration underlines that inclusion, diversity, and equity are key components to ensuring that machine learning systems do not create or perpetuate discrimination, particularly against marginalised groups. There are some groups for whom collecting data on discrimination poses particular difficulty, however, protections must extend to those groups as well.
Intentional and inadvertent discriminatory inputs throughout the design, development and, use of machine learning systems create serious risks for human rights; systems are for the most part developed, applied and reviewed by actors which are largely based in particular countries and regions, with limited input from diverse groups in terms of race, culture, gender, and socio-economic backgrounds. This can produce discriminatory results.
Inclusion, diversity and equity entails the active participation of, and meaningful consultation with, a diverse community to ensure that machine learning systems are designed and used in ways that respect non-discrimination, equality and other human rights.