Paper #9: The Missing Drive: Functional Analogs of Intrinsic Motivation in Large Language Models for Creative Tasks

Amabile’s Componential Theory of Creativity identifies three prerequisites for creative output: domain-relevant skills, creativity-relevant processes, and intrinsic task motivation. For large language models (LLMs), the first two components are extensively studied, with training data providing domain expertise and divergent thinking benchmarks assessing creative processes. The third component, intrinsic motivation, has been dismissed as inapplicable to machines that lack consciousness and intentionality. We challenge this view by adopting a functional account: rather than asking whether LLMs experience motivation, we ask what mechanisms serve the same functional role that intrinsic motivation serves in human creativity. We propose a framework mapping five functional dimensions of motivation to concrete LLM mechanisms: (1) creative intent via persona and instruction prompting, (2) persistence via structured creativity methods, (3) social motivation via multi-agent deliberation, (4) intrinsic reward signals via creativity-aware training objectives, and (5) freedom from extrinsic constraint, where we argue that reinforcement learning from human feedback (RLHF) alignment acts as a functional analog of the extrinsic pressure that Amabile’s theory predicts will suppress creativity. We provide experimental evidence for three dimensions and discuss implications for developing LLMs that are not merely capable of creativity, but driven toward it.