Governments Keep Letting AI Make Decisions & It’s Already Going Wrong

 Governments worldwide are rushing to implement AI systems to save time and money. Invariably, pitches are centred around efficiency increases such as smarter policing, faster queues and cleaner fraud detection. But the reality is much messier. Automated systems have wrongly cut benefits, facial recognition is growing faster than its safeguards, and prediction tools keep recycling biases of the past. This global snapshot outlines the most serious failures in recent years and what to look out for next. 

Across countries and use cases, the same traits continue to reappear. First is opacity; vendors and agencies claim secrecy, but people are left guessing why a model flagged them with little room to appeal. Secondly, the scale of implementations lends itself to major errors. A mistake in code rolled out nationwide can harm thousands at record speed, but would’ve been caught with slower, human-managed systems. “Bias in, bias out” is a third common theme across the models, meaning that the training is based on yesterday’s prejudices in policing or welfare patterns and expected to make tomorrow’s decisions. Fourth is the political difficulty to “undo” systems regardless of the errors they produce. When a tool is live and wired into performance targets or key governmental systems, rolling back becomes almost impossible. ...<<<Read More>>>...