The European Union’s Artificial Intelligence Act, conceived as a traditional _ex-ante_ product safety regulation, is fundamentally flawed because AI systems operate in unknown environments and take unforeseen actions, making an _ex-ante_ regime ineffective against unpredictable harm. This approach risks replicating the market concentration outcomes of the 2016 EU General Data Protection Regulation by disproportionately burdening smaller firms.
The Act's reliance on a false dichotomy between high- and low-risk AI, with only 5-15% classified as high-risk, fails to account for AI's dynamic nature and evolving risks, as seen with entertainment chatbots leading to self-harm. To address this, the AI Act should be revised, moving towards a balanced mix of _ex-ante_ and _ex-post_ measures. This involves reducing the _ex-ante_ compliance burden for most AI suppliers in exchange for robust _ex-post_ judicial review based on an ad-hoc AI liability framework, alongside new _ex-post_ learning, monitoring, and enforcement tools. Such a recalibration would more effectively prevent AI-related harm, reduce market distortion, and is expected to result in a negative net effect on AI companies’ compliance costs, garnering political support.
No comments:
Post a Comment