UNESCO’s Draft Recommendation on ethics and AI: our thoughts
There is no shortage of forums and processes around the world looking to develop standards on the governance, use and regulation of AI. One of the furthest along is the United Nations Educational, Scientific and Cultural Organization (UNESCO), which is aiming to adopt a Recommendation on the Ethics of AI by the end of the year. The latest draft of that Recommendation was recently published ahead of two intergovernmental meetings to finalise its text.
Although the final Recommendation will not be binding in any way upon UNESCO member states, it’s still an important process for a number of connected reasons.
- First, in the absence of any global treaty on AI (or any prospects of one being developed any time soon), many states will turn to soft law and other authoritative sources, including UNESCO’s Recommendation, for guidance on how to govern AI, so it may still influence national policy.
- Second, UNESCO has already developed a wide range of Declarations and Recommendations on other issues, such as bioethics and human genetic data; and, bolstered by its status as a UN specialized agency, UNESCO’s outputs have a particular degree of legitimacy. So, even if not translated into national policy, the Recommendation will likely be referenced by states in other spaces where AI standards are being discussed.
- Third, those other forums looking at AI which are also at an advanced stage, particularly the Council of Europe and the European Union, are regional rather than global. While their outputs may be influential outside of Europe, it is only at global forums such as UNESCO that all states can participate, adding to the legitimacy of their outcomes.
The likely significance of this process makes UNESCO’s decision to take an “ethical” rather than human-rights based approach even more disappointing. While ethical approaches to issues are not problematic in and of themselves, when it comes to developing normative standards on an issue such as AI, there is a risk of undermining the existing international human rights framework. This is because the challenges to individuals and societies posed by AI (such as discrimination or impacts upon privacy) are already largely addressed by the human rights framework, with the UN Guiding Principles on Business and Human Rights clarifying the role of the state and the responsibilities of the private sector when it comes to businesses’ impacts on human rights. In addition, while it may be interpreted and implemented in different ways around the world, the international human rights framework has a level of geopolitical recognition and status that any alternative ethical framework would be unlikely to match.
Developing an alternative “ethical” approach to AI suggests that the international human rights framework is inappropriate or insufficient, presenting an opportunity for the development of standards which are weaker or even inconsistent with that framework. While some look to ethics in good faith, there are certainly others who see it as an opportunity to avoid their obligations and responsibilities under international human rights law, or even to discredit its continued relevance. It is therefore critical that the final Recommendation is firmly rooted in the international human rights framework—and our assessment is undertaken, and recommendations, made on that basis.
BACKGROUND TO THE RECOMMENDATION
The work on developing a Recommendation started in early 2020, when UNESCO appointed an international group of experts to develop a draft text. The goal is for the Recommendation to provide guidance to UNESCO member states, through the elaboration of “ethical values, principles and policy recommendations for the research, design, development, deployment and usage of AI, to make AI systems work for the good of humanity, individuals, societies, and the environment”.
The initial draft of the Recommendation was published in the summer of 2020 alongside an open consultation which sought comments and feedback on the text. In our response, GPD highlighted a number of ways that the draft Recommendation could be further strengthened to ensure that human rights are mainstreamed in the language.
A revised version of the draft was developed on the basis of the responses to the consultation as well as comments from member states, and is now being considered at two intergovernmental meetings. The first took place at the end of April, with the second planned for 21—25 June 2021. The final Recommendation will then be adopted at the 41st session of UNESCO’s General Conference at the end of 2021.
While the latest draft of the Recommendation is certainly an improvement on the text that was subject to consultation last year, there are still a number of elements which are concerning from a human rights perspective. These include:
- A lack of clarity on the relationship between “ethics” and “human rights” in the text. While the draft Recommendation is clear that any measures taken must not breach international human rights law, there are a number of parts of the text which suggest that undefined ethical values should supplement human rights considerations. We believe that the Recommendation should be clear that it builds upon the existing international human rights framework, rather than supplements or provides an alternative to it.
- The inclusion of unclear and undefined language. Not only are the terms “ethics” and “ethical values” never comprehensively defined, the draft also makes reference to the need for AI to enhance “quality of life” and to ensure “just” societies, terms which are also not defined. Without a clear understanding of what these terms mean, there is a risk that they could be misunderstood, and subjectively and inconsistently applied in different states in a way which is inconsistent with the international human rights framework. We believe that this language should therefore be removed.
- A limited approach towards the need for human determinations and oversight of AI. Of particular concern in this regard is the provision that “in scenarios that involve life and death decisions, final human determination should apply”. This is a very narrow set of circumstances when human determination should be required; instead, final human determination should be required (or available) in all circumstances when AI systems make decisions which constitute serious impacts upon an individual’s human rights, such as decisions around an individual’s sentencing following conviction of a criminal offence. Where impacts are less serious, and where human determination may not always be feasible, human rights oversight and accountability for the decisions will still be necessary. We believe that the language should be amended on that basis.
We also have concerns over the breadth and level of prescription of the draft Recommendation, which contains around 100 detailed policy recommendations to governments. In many areas, the policy areas where recommendations are made overlap with the mandates of other global forums and processes, including on specific policy areas such as taxation, employment and healthcare. With multiple forums and processes developing standards and guidance on AI, consistency is important to ensure that there are no gaps or fragmentation.
Further, with such a large number of specific recommendations and no indication of prioritisation, the most critical recommendations from a human rights perspective are not being given the necessary prominence. Important recommendations on areas such as impact assessments, human oversight, transparency and multistakeholder collaboration are lost among those on more peripheral issues, such as guidance on human-robot interactions, encouraging museums and galleries to use AI systems, and promoting AI education and digital training for artists and creative professionals.
We urge the UNESCO member states participating in the intergovernmental meeting to review and revise the text to ensure that it is clearly grounded in the existing international human rights frameworks, provides clear guidance for governments on the measures they should prioritise, and ensures consistency with other human rights-based frameworks, standards and guidance that are being developed elsewhere.
The second intergovernmental meeting will take place from 21 to 25 June 2021. The final Recommendation will be adopted at the 41st session of UNESCO’s General Conference at the end of 2021. We’ll be following developments at UNESCO closely—take a look at our AI Policy Hub and Forums Guide, and sign up to our monthly Digest for regular updates and analysis.