​Applying human rights and ethics in responsible data governance and AI

This session will bring ethical and human rights' perspectives of AI as a technology that holds many promises and raises many concerns. Like any other technology, it all depends on how we use it. AI can contribute to addressing some of the world's most pressing problems, but it can also lead to more inequality. AI can make people's lives easier, but it can also generate discrimination and bias. So how do we develop and use AI in a human-centric and trustworthy manner? How to make sure that the data used by AI is reliable, accurate, and complete enough as not to generate discrimination? How to avoid privacy and data protection breaches in accessing and processing the large amounts of data that are at the core of AI? And how do we make sure that there is transparency and accountability in how algorithms function and AI is used?

At IGF 2018, we started a discussion on the role of ethical frameworks and human rights legal instruments in addressing these issues. One conclusion was that ethics differ across cultures and geographies, and, as such, we might be more efficient and effective in building trustworthy AI if we focus on applying existing human rights principles. Meanwhile, several new ethical frameworks have been developed by different entities (OECD, ITU, EU Commission, IEEE), and renewed calls have been made to ensure that human rights are always preserved in AI contexts, including in design, use and deployment. So where are we now, and what next? We have multiple human rights and ethical frameworks, but is this enough? How to apply them consistently in governing data, developing algorithms and actually using AI systems? Who bares this responsibility? And are there (or should there be) mechanisms for enforcement and monitoring in place?

More specifically, the session will address three policy questions:

a) What is, in fact, a trustworthy and responsible AI, especially with regard to data governance?

b) What is the role of human rights legal instruments and ethical frameworks in ensuring trustworthy and responsible data governance and AI? Are there any lessons learnt from existing frameworks?

c) How to cross the bridge between defining human rights and ethical frameworks and implementing them in AI systems and SDGs? What is the role of different stakeholders and how can they work together to achieve the best results?

Messages from last year's Main Session on Emerging technologies

As a basis for the discussion we will be using messages from last year's Main Session on Emerging technologies and main principles from actual Ethical and human rights concerned AI initiatives.

• Code is a reflexion of the people who have coded and influenced it. Ethics changes over time and across cultures, so is there some global ethics? Nevertheless ethics might be needed to guide a responsible approach to AI when regulation and laws can not keep up with the speed of technology developments.
• When we talk about ethics, we do not talk about what is the regulation in different countries, but we talk about what are the common values shared in a society, and that is something previous to law, that is something that runs parallel to law, and that is something that is additional to law. But if we build technologies with a global approach, developers may need a single guidance.
• Professional ethics is very much needed -- we have ethics for law, we have for psychologies, for medicine, but we don't have professional ethics in marketing, we don't have a professional ethics in engineering, and this is something that we need to start thinking more and more about.
• Fairness is the first success of building ethical approach. Most people agree algorithms should be fair, that means they should treat different groups of people equally (they should not be discriminating). We need to be more pro people in the development itself.
• Interpretability of algorithms is needed. Is there human responsibility and accountability for any decision taken by AI?
• Instead of using ethics as something we need to enshrine in technologies, would it be better to use principles defined in the UN Declaration on Human Rights that is less dependent on national, cultural and other differences and has a broader consensus within countries worldwide

AI main principles clusters

Using actual AI policies and initiatives we tried to cluster the main principles into the following categories

Principles for a human-centred AI for social good

• Serving people, society and the planet
• Supporting social and economic growth that benefit society
• Trustworthy, safe and secure AI
• Fair, unbiased and nondiscriminatory AI
• Respect for human dignity and choice
• Respect for ethical values and principles
• Respect for fundamental freedoms and rights (privacy, freedom of expression, etc)
• Lawfulness 

Principles for responsible AI

• AI as a solution to existing problems, not a solution in search of a problem
• Transparency in how AI is developed and works
• Algorithmic explainability
• Accountability and responsibility in the development and use of AI

Mechanisms

• Ensuring humans are always in control over AI 
• Embedding privacy, safety, and security in the development stage of AI
• Supporting the development of AI that benefits society, by making relevant data and infrastructures available 
• Developing policies, guidelines, standards, principles and best practices regarding the development and use of AI for social good
• Advancing the public’s understanding of AI and its implications
• Preparing the workforce for an increasingly-driven AI society
• Facilitating multidisciplinary dialogues on AI, its implications, and future
• Supporting synergies and cooperation between existings initiatives focused on human-centred AI

Reference:

• https://www.oecd-ilibrary.org/sites/eedfee77-en/1/2/5/index.html?itemId=/content/publication/eedfee77-en&mimeType=text/html&_csp_=5c39a73676a331d76fa56f36ff0d4aca&itemIGO=oecd&itemContentType=book#wrapper

• https://www.intgovforum.org/multilingual/content/igf-2018-emerging-technologies