Intrinsic Geometric Vulnerability of High-Dimensional Artificial Intelligence
This identifies a fundamental vulnerability in modern AI systems, which is critical for improving robustness in applications like security and safety.
The paper argues that high predictive performance in AI systems is intrinsically linked to vulnerability to adversarial attacks due to the geometry of high-dimensional data sets, highlighting a major pitfall and suggesting research directions to address it.
The success of modern Artificial Intelligence (AI) technologies depends critically on the ability to learn non-linear functional dependencies from large, high dimensional data sets. Despite recent high-profile successes, empirical evidence indicates that the high predictive performance is often paired with low robustness, making AI systems potentially vulnerable to adversarial attacks. In this report, we provide a simple intuitive argument suggesting that high performance and vulnerability are intrinsically coupled, and largely dependent on the geometry of typical, high-dimensional data sets. Our work highlights a major potential pitfall of modern AI systems, and suggests practical research directions to ameliorate the problem.