Position Paper: An Inner Interpretability Framework for AI Inspired by Lessons from Cognitive Neuroscience