South Korea's Ministry of Science and ICT and the National Information Society Agency (NIA) will start a call from April 1 for a 2026 project to build 10 types of inference data to support artificial intelligence (AI) model development.

The project aims to build high-quality inference data that includes logical thinking processes, such as Chain of Thought (CoT), and causal relationships, to improve the reliability of AI models and their industrial applicability.

It will pursue 10 tasks with a total budget of 6.6 billion won. Focusing on large language models (LLMs) and physical AI (manufacturing and robotics), it will build advanced inference-type AI data that can be used in real industrial settings.

In the LLM area, it will build inference data that includes step-by-step reasoning and the basis for judgments in problem-solving processes centred on LLMs, such as complex document understanding, logical decision-making and tool use. It will focus in particular on securing foundational data to implement reasoning AI that reflects South Korea's socio-cultural context and language characteristics.

The detailed tasks are five projects: knowledge inference data based on complex documents; data to support research processes; Korean-language tool-calling inference data; web and GUI-based action inference data; and error-augmented and correction inference data.

In physical AI, it will build inference-type data based on problem situations in actual industrial sites, enabling AI to analyse causes and derive solutions. The goal is to secure data for autonomous manufacturing that can perceive situations, infer causal relationships and correct and carry out tasks across diverse manufacturing environments.

The detailed tasks are five projects: multi-sensor anomaly diagnosis and cause inference data for manufacturing facilities; surface defect cause analysis and quality assessment inference data; robot task failure cause analysis and recovery action data; physical simulation inference data for humanoid action generation; and asynchronous process causality analysis and inference data.

The data built through the project will later be released on 'AI Hub' so companies, research institutions and startups can use it freely.

Choi Dong-won (최동원), director general for AI Infrastructure Policy at the ministry, said: "Through this project, we will secure tailored inference data needed in real industrial sites and actively support a qualitative leap for South Korea's AI industry."