Abstract: Artificial intelligence has been ballooning as a promise to accelerate progress in scientific research and as a new tool for teaching science. The lessons from the use of social networks, however, warn us that reliance on these tools may increase biased opinions and reinforce prior beliefs in discussions within communities that become increasingly isolated and intolerant. I will discuss the example of asking Deepseek about the origin of the difference between the proton and neutron masses according to QCD, where the answers of AI to the physics questions seem clear, elegant and compelling, but may simply follow from the way the user is asking questions and may end up reinforcing wrong ideas and beliefs that may be incorrect as physical explanations. The talk will discuss both on the QCD physics of the proton and neutron mass difference and how AI may or may not be useful to learn on questions like this.