Problem Learning: Towards the Free Will of Machines

Problem definition remains human-controlled: The paper identifies that while researchers have automated most components of machine learning pipelines (representation, model, loss, optimizer, metric), the critical first step—defining what problem to solve—still requires extensive human expert effort, representing a significant gap in automation.

Machines could discover hidden problems: The authors argue that automating problem discovery could help identify "valid and potentially important problems hidden in data that are unknown to domain experts," suggesting AI systems might uncover research opportunities or application areas that human experts have overlooked.

Ethical problem formulation is central: The paper emphasizes that problem learning should focus on discovering "valid and ethical problems," indicating that responsible AI development requires building ethics considerations into the problem definition stage itself, not just downstream in model development.

Current AI lacks autonomous problem-seeking: The paper frames problem learning as a step toward "free will of intelligent machines," noting that today's systems are "limited to solving the problems defined by humans, without the ability or flexibility to freely explore various possible problems"—highlighting a fundamental constraint in current AI autonomy.

Broader implications for AI independence: By automating problem discovery, machines could potentially expand their scope of tasks and inspire new findings autonomously, raising important questions about how much independent problem-seeking capability we want AI systems to have.