[Reporter’s Notebook] For AI Autonomous Manufacturing, the Key Is Not Technology but "Perception"

When preparing the special series "AI Autonomous Manufacturing: Opening the Future," the first image that came to mind was a factory where robots manufacture products and control facilities entirely on their own, without any human intervention. However, in reality, there was no 'magical' artificial intelligence (AI) on site. The essence of AI autonomous manufacturing was not rapid and flashy automation, but rather a transformation that begins with the slow and meticulous work of human hands.

Companies encountered at industrial sites in Ulsan and Yeosu have spent years pursuing "hand-built innovation" to introduce applied AI. Manually written inspection sheets, equipment records scattered across Excel files, and defect judgment criteria that varied by person were all too unstructured for computers to recognize and learn from. The starting point of AI innovation was the painstaking task of reorganizing this tangled data into a language that machines could read.

Companies such as SK Innovation, GS Caltex, and HD Hyundai Mipo have, since the 2010s-before the AI boom-conducted multiple reviews and pilot projects to define precise requirements and design site-specific models. They independently defined why AI was needed in their processes and carefully proceeded with data refinement and modeling. Rather than simply adopting technology, they worked together to ensure AI could function properly in the field. This was, in a sense, a process of teaching the world to AI, requiring a high degree of patience.

The companies also cited "people’s perceptions," rather than technology, as the biggest obstacle to the transition to AI autonomous manufacturing. Among some employees, deep-seated distrust persisted that AI would replace humans. Conversely, some government officials and company executives misunderstood AI as a universal technology that could be instantly deployed across entire operations. Caught between distrust and blind faith, those responsible had to explain and persuade each step of the way. AI autonomous manufacturing was not a product of technology alone, but of coordination and convergence based on accurate understanding.

Government-led AI projects aim to build "universal platforms" that can be widely used across industries. However, on the ground, most sites are still at the stage of refining their data. Practitioners say this is a time to build up from the basics, rather than to produce packaged results. For AI to demonstrate autonomy, incremental innovations must occur across all areas, including initial modeling, anomaly detection, predictive maintenance, and quality assessment.

According to the theory of technology diffusion, there comes a point when awareness reaches a critical threshold and technology spreads explosively. The arduous accumulation of corporate efforts, hands-on fieldwork, and data refinement is the process of preparing for that inflection point. At every moment of crisis, South Korea has leapt forward through innovation; now, in the new domain of AI autonomous manufacturing, it must once again prove its growth potential. AI must move beyond experimentation to real-world application. And unless it is infused with the sweat of those on the ground, it will never truly function.