What does it mean to engage in a dialogue with AI?   How can Human-AI dialogic interactions transform our approach to scientific inquiry and collaboration?   As AI becomes integral to scientific research and broader societal applications, we investigate these pivotal questions through the design and evaluation of dialogic human-AI interfaces. By combining the theoretical frameworks of Dialogism and Bayesian inference we recognise AI's potential to support and empower rather than replace human participation. We enable its role as a 'complementary voice' within citizen science activities, specifically assessing its effect on the accuracy of bumblebee species identification.   Central to our research is the creation of a dialogic space that fosters real-time dialogue between humans and AI, focusing on dialectics (i.e. examining and discussing the validity of opposing views) to facilitate knowledge construction through collaborative decision-making. Our findings illustrate the benefits of this collaborative approach, showing a significant improvement in identification accuracy for both humans and AI when engaged in a dialogue. Moreover, in continuing these dialogic collaborations, a Bayesian model that combines human and AI inputs outperformed decisions made post-collaboration by either AI or humans.   Our work encourages a dialogic approach in scientific practice, leveraging the complementary strengths of humans and AI. The outcomes of this research highlight the potential of such methods to reshape the landscape of scientific inquiry and collaboration.