Abstract
Over the past two decades, foundational work has provided key insights into the cognitive and neural basis of hallucinations. This progress has led to the development of several families of theories, each of which propose that hallucinations arise from distinct cognitive mechanisms. Since these cognitive mechanisms likely map onto separate circuit-level implementations, arbitrating between them is critical to advance our understanding of hallucination pathophysiology and guide the development of novel targeted therapeutics. However, several obstacles have hindered this progress, including the under-specification of theories and inadequate comparative testing. To overcome these challenges, and following best practices in cognitive computational neuroscience, theories should 1) articulate computational and biological details at a level that allows the generation of precise, testable predictions, and 2) be evaluated using experiments designed to emphasize their unique signatures to facilitate falsification. To illustrate this general approach, we demonstrate how theory-driven computational models constrained by well-replicated findings across basic, preclinical, and clinical neuroscience can provide a principled means to prioritize falsifiable mechanistic theories of auditory hallucinations. We then discuss how these models can be used to inform the development of richer behavioral paradigms and analytical approaches that enable direct comparisons between competing theories. Overall, we propose a general strategy for the specification and falsification of candidate mechanisms underlying (auditory) hallucinations - a critical prerequisite for refining hallucination theories and advancing their translational potential.