Detecting Dark Patterns Using Generative AI: Some Preliminary Results

Abstract

This article presents preliminary investigations of generative AI technologies as tools for detecting dark patterns. This work is motivated by growing regulatory interest in dark patterns and deceptive online choice environments, and the need to development meaningful behavioural auditing tools as a response to these practices. We propose and offer initial explorations into three approaches to using generative AI to simulate the behaviour of online users with differing levels of digital skill. Our initial testing shows the ‘AI Vision’ approach to hold promise at present; the ‘Choose your own adventure’ approach to hold potential; and the ‘Decision Network’ approach to raise interesting technical challenges.

Keywords: Dark Patterns, Generative AI, Behavioural Audits, Algorithmic Fidelity