What is the problem?
The term ‘AI’ is scientifically contested. Ethical and human rights issues are often not directly linked to AI in a narrow sense. Specific technologies (e.g. machine learning, deep neural networks; reinforcement learning) can have properties leading to concerns that are not relevant to other technologies. For example, machine learning using deep neural networks requires large training datasets which can raise issues of data protection and security, but may also perpetuate biases that are contained in the datasets. This would not apply to AI techniques that do not rely on the analysis of large data sets, such as symbolic AI or systems that use purely technical data
Who should act?
All bodies developing policies and guidelines for ethical or trustworthy AI, including the European Commission, national governments, standardization bodies.
The scope of AI needs to be clearly defined in each use context with regards to relevant issues.
Where appropriate, ‘AI’ should be replaced with more specific terms, such as ‘machine learning’.
- Some issues, such as changes to employment, or political and economic power redistribution are only peripherally linked to AI. When addressing them it may be more appropriate to use inclusive terms such as ’emerging digital technology’.
- Other issues (e.g. autonomy of machines) are already well understood and categorised (e.g. the taxonomy of levels of autonomy in vehicles)[BS2] , which may help develop similar categorisations in other application domains.
Use a concept of AI that points to the features of the technology that are ethically relevant, such as opacity (can hide bias) or automation (replaces jobs). Characteristics or examples may be more helpful than definitions.
The concepts used influence the scope of the technology in question but also the responses (e.g. scope of a risk or impact assessment).
Whichever concept of technology is used, the ethical and human rights implications depend heavily on the application area. Machine learning, for example, may have very different consequences in healthcare or in gaming.
The focus of this recommendation is on the definition of technology. If the focus is on ethical issues of AI, then the concept of ethics needs to be clearly understood. Human flourishing is a useful term to highlight ethical ideas, but there are many other ethical positions worthy of considerations.
Human rights are outlined in various legal documents including the Universal Declaration of Human Rights, Charter of Fundamental Rights of the European Union, European Convention of Human Rights and the OECD’s Framework for the Classification of AI Systems
SHERPA’s work on case studies and scenarios has informed the categorisation of AI in terms of narrow AI (machine learning), converging socio-technical systems and artificial general intelligence. This definition can help delimit ethical and human rights issues.
Ryan, M., Antoniou, J., Brooks, L., Jiya, T., Macnish, K., & Stahl, B. (2020). The Ethical Balance of Using Smart Information Systems for Promoting the United Nations’ Sustainable Development Goals. Sustainability, 12(12), 4826. https://doi.org/10.3390/su12124826