'Tacit Knowledge AI' extracts manufacturing masters' senses: Whose labor is this?

Lee Young-jin, a master steelmaker at POSCO, describes how he relies on subtle sensory cues, the sound of molten metal, the angle of flames, and ambient heat, to make critical judgments in the steelmaking process. These intuitive skills, honed over decades, are difficult to teach through manuals alone. POSCO has been working since the 2010s to create a 'smart factory' by digitizing these expert insights, using deep learning to mimic the know-how of workers with over 30 years of experience to achieve optimal results. This effort has led to 'one-touch operation' for furnaces, eliminating quality variations between workers.

“

I used to judge more than 70 subtle signs, such as the pitch of the sound of molten iron in the converter, the angle of the flames, and the subtle heat inside the workplace, relying on my five senses.

South Korea's manufacturing sector, a key pillar of its economy, relies heavily on the expertise of its master craftspeople. However, with the retirement of the 'second baby boom' generation of skilled workers, there's a growing concern about the loss of this invaluable 'tacit knowledge', skills and intuition that are not easily codified. While large corporations are actively seeking ways to preserve this knowledge, many small and medium-sized manufacturers lack the resources to do so, facing the imminent loss of their core workforce.

“

Skills can be taught through manuals, but the intuition to grasp the state of molten iron requires a long accumulation of time. I think it is more important than ever to instill a sense of craftsmanship that goes beyond technique, making junior colleagues observe the 현장 (workplace) as if conversing with molten iron.

Recognizing this challenge, the South Korean government has launched a project to secure manufacturing tacit knowledge, designating it a 'national core asset.' The Ministry of Trade, Industry and Energy aims to develop AI models based on the non-standardized data of master craftspeople across key sectors like automotive, shipbuilding, steel, machinery, electronics, and more. The project plans to select around 30 tasks, supporting each with up to 1.6 billion won.

“

The tacit knowledge of manufacturing, such as the work tips, judgment criteria, and sensory know-how of artisans and skilled workers that cannot be quantified, is difficult to transfer and is in danger of being cut off and disappearing with retirement.

Capturing and digitizing these subtle, intuitive judgments is a complex process. The Korea Institute for Advancement of Technology (KIAT) is overseeing the project, focusing on how to acquire and train AI on non-quantifiable data like movements, sounds, and judgments. This involves extracting and refining various data types, video, audio, images, that are contextually linked to the tacit knowledge. For instance, the AI might learn to correlate weather conditions, material properties, and a master's adjusted working methods to achieve consistent results, verifying the AI model's performance against a non-expert's output.

“

We agree with the government's policy direction to utilize AI for manufacturing innovation, but it is being pushed unilaterally without social discussion on structural changes like job reduction.