At Amsterdam’s Vrije Universiteit, assistant professor Filip Ilievski is exploring the intriguing intersection of AI and human reasoning. His research focuses on enhancing AI’s ability to solve logic puzzles and riddles, revealing a significant gap in AI’s common sense reasoning. While AI excels at recognising patterns, it struggles with abstract thinking and temporal reasoning, as demonstrated in a recent study where OpenAI’s GPT-4 failed to determine if a character named Mable was alive at noon based on provided details.

Ilievski’s work suggests that researchers can unlock secrets about our cognitive processes by examining how both AI and humans tackle complex tasks. For instance, humans often trust their intuition, which can lead to errors, while AI can sometimes excel at problems that trip us up, such as the classic bat and ball riddle. However, when faced with novel puzzles, AI models still lag behind human performance, indicating a unique advantage in human reasoning skills.

Recent advancements show promise, as newer AI models like GPT-o1 have improved significantly, successfully answering challenging questions and demonstrating a better grasp of logic. Despite these strides, researchers caution that understanding AI may not directly illuminate how human cognition works, highlighting that the two systems operate differently. Ultimately, the collaboration between AI and human intellect could improve cognitive understanding in both realms.