Inference to the best explanation and epistemic circularity