Modern Machine Learning (ML) models show impressive performance, especially when it comes to natural language processing. Should we thus believe that present-day Artificial Intelligence (AI) literally thinks and understands? To answer this question, I renovate Searle's famous Chinese Room Argument (CRA) in the context of modern ML. I will first argue that Searle's own extension of the CRA to the context of ML with Artificial Neural Networks (ANNs) is flawed. By offering a more careful take on basic ML theory, I shall then suggest an alternative extension, called the Chinese Library, in which almost all of the original intuitions are preserved. Second, I will highlight some well known issues with Searle's original argument, and formulate a more cautious argument, to the effect that the total available behavioral evidence for machine understanding is defeated by the applicability of the Chinese Library to modern ML. In essence, I will thus argue that we are still short of evidence that puts us in a position to reasonably believe in machine understanding.
Alle Termine des Institutskolloquiums im SoSe 2026 finden Sie hier.