Self-generated goals in the age of foundation models: artificial autotelic agents and their implications for cognitive science

Goals have traditionally been treated as inputs to intelligent systems, biological and artificial, rather than as products of intelligence itself. A growing body of work in cognitive science and artificial intelligence (AI) challenges this framing through autotelic agents — self-improving systems motivated to represent, generate, select, and pursue their own goals. Recent advances in foundation models (large-scale generative AI models trained on broad datasets) expand this research program by extending the languages and tools through which artificial agents can express, evaluate, and pursue goals. By operating over formats such as natural language and code, foundation models allow agents to generate more abstract and compositional goals, approximate human judgments of interestingness and learnability, and support richer loops of problem generation and problem solving. These advances create new opportunities for cognitive science to model human goal dynamics, for AI to build more adaptive autotelic systems, and for human–AI systems to support shared goal exploration without narrowing human agency.