Tsinghua researchers unveil AI framework that breaks problems into smaller steps

Researchers at Tsinghua University have introduced Calculus of Intelligence, a mathematical framework designed to help AI systems solve complex tasks by dividing them into smaller parts. The work, published July 17 in iFuture, could shape more scalable and reliable agentic AI for software, cybersecurity, data analysis and scientific research.

Why it matters: - AI systems are being asked to do more complex work with less human oversight. - The new framework aims to make advanced AI easier to build, understand and verify. - The approach could support agentic AI systems that handle real-world tasks with minimal supervision.

What happened: - Researchers at Tsinghua University introduced a new mathematical framework called Calculus of Intelligence, or COIN. - The study was published July 17 in iFuture, an open-access journal published by Tsinghua University Press. - The work was co-authored by Yang Yuan, an associate professor at Tsinghua University's Institute for Interdisciplinary Information Sciences, and Andrew Chi-Chih Yao, professor and dean of IIIS. - Both researchers are also affiliated with the Shanghai QiZhi Institute.

The details: - COIN breaks large problems into smaller tasks that can be solved individually and then combined into a complete solution. - The framework is designed to organize advanced AI systems in a structured way. - The researchers compare the method to aircraft engineering, where separate teams handle wings, engines and control systems under shared design rules. - COIN uses advanced mathematics to define how smaller tasks connect within a larger system. - The researchers say the goal is to develop reliable AI systems that can handle increasingly complex real-world challenges. - The framework is intended to support specialized AI systems working together instead of relying on one all-purpose model. - The research points to possible uses in software writing, data analysis, cybersecurity, workflow management and scientific research.

Between the lines: - The framework reflects a broader shift in AI design away from single, large models and toward coordinated systems of specialized agents. - That structure could make future AI more scalable and easier to manage, especially as tasks become more complex. - The emphasis on verification suggests a response to growing concerns about trust, reliability and control in AI systems.

What's next: - The researchers believe COIN could help shape the next generation of agentic AI. - Further work will likely focus on how the framework performs in practical AI systems and whether the mathematical structure improves reliability at scale. - The full paper is available here.

The bottom line: - Tsinghua researchers are proposing a way to make AI development more modular, more understandable and potentially more dependable by solving big problems one piece at a time.