Op-ed: AI will fail, so we must study everyday failures to build resilience

This is an opinion piece from Ingeniøren, written by a group of Danish researchers including Naja Holten Møller, Henriette Langstrup and Mogens Jin Pedersen (University of Copenhagen), Laust Hvas Mortensen (ROCKWOOL Fonden), Nicki Skafte Detlefsen (DTU), Thomas Moeslund (AAU) and Andreas Bjerre-Nielsen (CAISA/KU). Its title argues that AI will fail and that resilience cannot be reached without studying everyday failures. Only the opening of the piece is available, so the full argument is not covered here.

The authors take the CPR case, in which millions of personal records became accessible to unauthorised people, as their starting point. They say it is being treated as an isolated incident, but that it illustrates a challenge that will grow as AI spreads. When public digital systems fail, they write, the cause is rarely one technical or human error, and the problem is often much broader.

Their central call is for society to create the frameworks that let practitioners learn from the mistakes and experiences of daily work, especially as technology plays a larger role in society and AI arrives.

Two readers commented. One agreed that an effective regime of rules and oversight is needed because self-regulation has failed. He argued that AI providers are running an influence campaign built on fear, uncertainty and myths, and that tools open to anyone tempt inexperienced people into uncontrolled experiments. His proposal is that those commissioning AI projects must be able to answer for the consequences and scope of what they do, and that projects stay within existing law rather than law being adapted to the technology.

The other commenter welcomed the researchers' support for investigating failures to prevent similar ones, and noted that a commentator at Ingeniøren and Version2 calls this a crash commission. He said the first step is getting the public sector to share information voluntarily, which need not mean sharing with the whole country or world.