Douwe Korff
This article looks at three main issues raised by the PNR scheme: (i) the base-rate fallacy and itseffect on false positives; (ii) built-in biases; and (iii) opacity and unchallengeability of the decisionsgenerated, and at whether the Court has properly addressed them. It concludes that the AG and theCourt failed to address the evidentiary issues including the base-rate fallacy—a lethal defect. It alsofinds that neither the Member States nor the Commission have even tried to assess whether theoperation of the PNR Directive has resulted in discriminatory outputs or outcomes; and that theCourt should have demanded that they produce serious, verifiable data on this, including onwhether the PNR system has led in practice to discrimination. But it also finds that the AG and theCourt provided important guidance on the third issue, in that they made clear that the use ofunexplainable and hence unreviewable and unchallengeable“black box”machine-learning artificialintelligence (ML/AI) systems violates the very essence of the right to an effective remedy. Thismeans that any EU Member State that still uses such opaque ML/AI systems in its PNR screeningwill be in violation of the law.