IGF 2019 WS #365
Algorithmic decision-making for the benefit of all

Description: The use of AI in decision-making is all around us; whether we visibly see their uses or not, we are affected by them, and their impact on our lives is only set to grow. Treatment for patients in the public health system in Italy is allocated with the help of an automated system, the Danish state tries to identify children vulnerable to neglect using an algorithm, the SyRI system in the Netherlands is supposed to detect welfare fraud with the help of so-called AI, the Swedish municipality of Trelleborg has automated parts of its decision-making for the disbursement of social benefits, and the EU tests an automated lie detector at its borders in Greece, Hungary and Latvia.
At the same time, decisions concerning marginalized groups in the Global South are often based on non-representative data (irrespective of whether these are with the aid of AI or not). Countries in Africa are on their way towards a centralized, unified and biometric repository of their population, bringing with it opportunities for financial inclusion – as a result of systems to assess creditworthiness – but at the same time these raise the risk of Chinese-style citizen scoring.
Much of the debate in the field of Automated / Algorithmic Decision-Making (ADM) systems and so called Artificial Intelligence privileges discourses of lofty ethical norms. Too little is known about where and how these systems are used in practice, especially in the Global South. Too little is also discussed of the limits of technical solutions to bias predicated on a desire to optimize multiple (often conflicting) notions of fairness. The discourses also often disregard the opportunities for ADM to bring to light biases in human decisions making that hereto were difficult to reveal.
Within these disparate use-cases, it is very important to ask what constitutes context-specific, fit-for-purpose policies with regards to ADMs. To do this we need evidence that will both inform and shape the needed actions, so that ADMs could be used increase welfare and liberties, whilst also limiting their abuses through harmful surveillance, discrimination, and control.
We will share state-of-the art research about the limits of technical notions of fairness and discrimination, as well as the practice of algorithmic decision-making processes already in use: from our reports "Automating Society – Taking Stock of Automated Decision-Making in the EU", “Identity-management and citizen scoring in Africa”, the “Atlas of Automation”, the “Internet Health Report” and "Bias and the Global South: Care now? Care later? Or not at all?”.
In a first part, we will briefly present excerpts of our evidence-based research on issues of ADM from both the Global North as well as the Global South and lay out our recommendations for policies and government arrangements. In the second part we will discuss these with the participants (including government, the private sector, and the technical community) and the audience.

Expected Outcomes: - Better (shared) knowledge among stakeholders about the use of ADM/AI systems in practice
- Better (shared) knowledge among stakeholders about possible governance mechanisms for these systems
- Tangible ideas to further evolve the governance of ADM systems