Assessing the Human Likeness of AI-Generated Counterspeech