EAIRA: Establishing a Methodology for Evaluating AI Models as Scientific Research Assistants