Recent research suggests that working with generative artificial intelligence (AI), such as ChatGPT, can produce more creative outcomes than humans alone. However, does AI retain its creative edge when humans have access to alternative information sources, such as another human or the internet. We explored this question in a between-group experiment with 202 German participants across four conditions (human-human dyads, human-Internet, and two human-AI groups with basic or specific instructions) and four creativity tasks (two alternate uses tasks, a consequences task, and a problem-solving task). Results showed that the human-human condition obtained higher creativity scores in the divergent thinking tasks than the remaining groups. No significant between-group differences were observed in the problem-solving task. Moreover, interacting in human dyads made people more creatively confident, an effect not observed in the other groups. In addition, we compared human-rated outcomes with AI-based automated scoring (Ocsai). Interestingly, notable discrepancies emerged between the AI assessment and the human-judged results, raising concerns about AI's susceptibility to "elaboration bias." These findings highlight the benefits of human collaboration for creativity and call for further studies about the reliability and potential biases of AI in evaluating creative performance.