[arXiv22] EST: Evaluating Scientific Thinking in Artificial Agents