Abstract
Research on human and artificial intelligence has largely focused on problem solving: given a predefined problem, can an agent produce an effective solution? Yet many real-world challenges cannot be captured by a fixed problem representation. They require agents not only to solve problems, but also to recognize when the original problem representation has become inadequate and must be revised to reflect new evidence or changes in the problem itself. We argue that this capacity, termed adaptive reframing, constitutes a fundamental but under-researched dimension of intelligence. We propose a multidisciplinary framework for studying adaptive reframing across neuroscience, cognitive science, robotics, experimental economics and psychology, and complex systems. Adaptive reframing is defined as the capacity to revise the construction of a problem in response to changing evidence, causal structure, or social context. To support a systematic science of adaptive reframing, we introduce three complementary paradigms: the Dynamic Scenario Framing Task, the Perturbed Causal Model Task, and the Social-Affective Problem Setting Task. These paradigms provide a common framework for comparing humans and artificial agents in situations requiring problem reframing, and for building architecture-enhanced AI systems. We propose a shift from problem solving to adaptive problem framing as a central focus of intelligence research.