This internship consists of two parts. In the first part the theoretical limits of artificial intelligence are explored. These limits are related to foundational computer theorems, such as Gödel’s theorems, and the halting problem. These theoretical limits are contrasted with capabilities of humans, that do not seem to suffer from these limits. The second part of the internship translates these theoretical considerations into an empirical research line.

The aim of this research line is to uncover tasks that humans can easily perform accurately, but large language models struggle with. This research line builds on recently conducted investigations that indeed found that large language models fail at seemingly easy tasks, while humans perform these tasks nearly flawlessly. The overall goal of this research is to challenge the influential idea that human cognition is entirely computational in nature.