A Stanford study found that while humans rated AI-generated research ideas highly, human-generated ideas ultimately performed better in practice