Speaker
Description
This work introduces a comprehensive framework and discussion on the measurement of scientific understanding in agents, encompassing both humans and machine learning models. The focus is on artificial understanding, particularly investigating the extent to which machines, such as Large Language Models (LLMs), can exhibit scientific understanding. The presentation centers around fundamental physics, specifically particle physics, providing illustrative examples within this domain.
The study builds upon a philosophy of science perspective on scientific understanding, which is expanded to encompass a framework for assessing understanding in agents more broadly. The framework emphasizes three fundamental aspects of understanding: knowledge acquisition, explanatory capacity, and the ability to draw counterfactual inferences. Furthermore, the capabilities of LLMs to comprehend the intricacies of particle physics are examined and discussed.
Through this interdisciplinary exploration, the talk sheds light on the nature of scientific understanding in agents, bridging the gap between philosophical accounts and the potential of advanced machine learning models. The insights gained contribute to the ongoing dialogue on the boundaries of artificial understanding and its relevance in scientific research, particularly in the context of particle physics.
The work is based on https://arxiv.org/abs/2304.10327 and subsequent work.
| Collaboration / Activity | no collaboration, see comments |
|---|
