Special Issue on Algorithms and Ethics now available

As the entire world is preoccupied with the pandemic, Artificial Intelligence (AI) ethics may seem to have dropped in importance. Yet, as we adjust to new circumstances, the impact of algorithms on society remains a topic of considerable interest.

Just last week, concerns arose in the UK regarding the impact of algorithmically predicted grades. Thousands of students received their grades from an algorithm that predicted the results of their exams, which had been cancelled by the coronavirus pandemic. For many of UK students, access to higher education was affected by the algorithm, leading to wide-spread concern and the eventual withdrawal of the algorithmically predicted grades.

The impact of automatic decision-making in society has been treated before in the context of the justice system, as can be seen from the sheer number of articles in a special issue of the Royal Society published in August 2018 and now made freely available.

The special issue was compiled and edited by Professor Sofia Olhede, Chair of Statistical Data Science at EPFL and her colleague Patrick Wolfe at Purdue University following a meeting that they hosted at the Royal Society. Olhede and Wolfe have written an overview paper in the same issue covering the growing use of algorithms in society, e.g. intersections between legal and ethical frameworks, bias and fairness, accountability and agency, as well as data management.

“Automated decision-making based on the usage of algorithms stands to make many functions in public life considerably more cost effective, and in principle, transparent,” says Olhede. “Algorithms implement the instructions given to them, and a key challenge in their broad usage is public acceptance.”

The problems and fears raised in the articles of the special issue are all the more relevant today in the UK in the wake of the grading algorithm controversy. In the issue, the former president of the UK Law Society (Christina Blacklaws) discusses the question of transparency for legal decisions. Another article looks at algorithmically assisted decision-making in the public sector. Both aspects are direct related to the question of algorithmic grade adjustment.

“Many of us have been involved in different projects to develop the ethical framework for large-scale deployment of automated decision-making,” says Olhede. “Real large-scale data has a complex structure, providing new challenges; but even more so, technical innovation relies on public adoption.”

The special issue addresses several aspects required for the public acceptance of automated decision-making, but a core, persistent challenge is “adapting a framework into practical solutions in real applications”.

“Even so, careful work needs to be done to codify the trade-offs from principles into code, and to communicate implementation choices and trade-offs as algorithms are deployed, as the recent controversy in the UK shows,” says Olhede, who is also one of the co-organizers of the virtual CIS colloquium series that will host Kay Firth-Butterfield this autumn. A notable contributor to discussions on algorithms and ethics, Firth-Butterfield is head of AI and Machine Learning at the World Economic Forum, as well as one of the co-chairs on the IEEE project on Ethical Considerations in Artificial Intelligence and Autonomous Systems