Vibe Research Drives Scientific Inquiry Shallower Unless Human Judgment Is Scaffolded

As AI systems grow more capable, researchers increasingly collaborate with them in scientific inquiry, a practice we term vibe research. This position paper argues that vibe research drives scientific inquiry shallower unless human judgment is scaffolded. We identify the root cause as Knowledge-Judgment (K-J) decoupling: at the research frontier, where no established ground truth exists, human judgment is the primary means of evaluating AI-generated knowledge, yet it is fallible, and its errors may go undetected. We propose the Asymmetric Coupling Framework (ACF), which models human–AI research collaboration as two nested loops. An exploratory study provides initial support for the framework's core implication: human judgment, not AI usage volume, determines the quality of AI-assisted research. To address K-J decoupling, we propose decision readiness scaffolding, through which AI helps researchers build the evaluative capacity needed at specific decision points. Scaffolding substantially reduces the expertise deficit: all 13 participants working outside their expertise produce ideas that meet or exceed the quality baseline they set within it. These results point to a broader principle: as AI takes a larger role in science, reliable research depends less on more capable AI than on deliberately cultivating the human judgment it cannot replace.