'AI learns from our masters': Posco's vision for the AI-powered steel mill

"At Posco, whether or not someone holds the title of master craftsman, there are people throughout the company who have accumulated remarkable know-how and experienced all kinds of unexpected situations. When these experts teach AI their knowledge and expertise, the AI's capabilities rise as well."

The "veteran's know-how" that senior workers have passed down to juniors on the steel mill floor over decades is now being taught to AI. Work is underway to convert into AI knowledge the accumulated experience and judgment of skilled workers — from reading the internal state of a blast furnace to responding to sudden emergencies.

Im Chi-hyun, head of Posco Holdings' Group DX Strategy Division, identified "people" as the core competitive advantage behind the Posco Group's AI transformation, or AX, in an interview on Friday. "If the people who originally did the work have strong capabilities, AI can advance rapidly as well," Im said. "The fact that Posco has so many outstanding field experts with experience, know-how, knowledge and a sense of responsibility is a great strength."

Im is a professor of industrial and management engineering at Postech who was appointed Posco Holdings' chief digital officer last year — an unusual outside hire that drew attention for placing a young academic specialist in a key group executive role. He currently oversees the group's digital, AI and robot strategy and the execution of core AX initiatives.

The blast furnace is the clearest example. Workers cannot directly observe how iron ore, coke and other materials are reacting inside a massive blast furnace. That is precisely why the experience of field experts — who have spent years estimating internal conditions and responding based on equipment data — is so valuable.

Posco is systematically organizing the judgment criteria of these experts and feeding them into AI training. Records of decisions made during unexpected or abnormal situations on the floor — and the outcomes that followed — also serve as training material for the AI.

"A blast furnace is a kind of black box — you cannot directly see what is changing inside — but there is a deep reservoir of know-how built up over the years for estimating its condition," Im said. "We are organizing the experience and judgment of our experts and using it to train AI that predicts the state of the blast furnace."

In some areas of the production floor, AI has already taken over a significant share of decisions that humans once made.

The galvanizing process is a prime example. Because applying too little zinc can cause defects, human operators tended to err on the side of caution and apply more than the required amount. Quality was not compromised, but the excess use of expensive zinc drove up production costs.

Posco addressed this by having AI predict coating conditions in real time and control the process so that only the necessary amount of zinc is applied. When predictions go off due to changing conditions or raw material variations, the model is recalibrated based on subsequent inspection results.

In the steelmaking process, Posco is also combining computer vision with the know-how of experienced workers to automate tasks that once required workers to visually assess conditions and manually operate equipment. "We have incorporated know-how that pure automation alone could not capture, and we are now well along the path toward AI handling the process from start to finish with almost no human intervention," Im said.

Posco's next challenge is linking the AI systems scattered across its facilities into a single, unified whole — moving beyond optimizing individual processes to finding the best answers for the steel mill as an integrated operation.

Energy management across the steel mill is the primary testing ground. Steel mills generate their own energy not only from purchased electricity and LNG but also from byproduct gases produced during the steelmaking process, including from blast furnaces. When equipment operation and production plans change, so do the required energy volumes and the most economical way to supply them.

The Posco Group has developed an AI agent that understands the energy flows across the entire production process and the physical characteristics of each facility. It serves as an "energy brain" for the steel mill, capable of assessing how a change in one process affects other processes and energy costs across the operation.

For instance, the AI judges whether to increase self-generation, purchase electricity from Korea Electric Power Corp., or prioritize the allocation of byproduct gas to a particular process — all calibrated to production plans and equipment conditions. The plan is to eventually expand this to identifying energy waste in the production process — such as unnecessary equipment operation or slab reheating — and optimizing the overall production schedule.

Posco has not yet reached the stage of a fully autonomous steel mill that eliminates human judgment entirely, however. Im said there are candidate cases where AI can plan and execute tasks on its own, but given the need for thorough verification in a manufacturing environment, human approval and intervention are still required at this stage.

The guiding principle running through Posco Group's AX strategy is a refusal to pursue "AI for AI's sake."

Posco Group Chairman Chang In-hwa has repeatedly stressed that "the company that transforms through AX fastest will win" — and his "mission-oriented AX" strategy reflects the same thinking. The goal is not to increase the number of AI deployments but to use AI as a tool to help each business — steel, secondary battery materials, energy and others — do its core work better.

"AI should not be used more for its own sake — it should be used as a tool to help the company do what it needs to do better," Im said. "If you approach it simply as 'let's try using AI,' you may end up automating specific tasks but fail to translate that into overall productivity or profitability gains."

Posco's vision of physical AI extends well beyond robots.

"Physical AI at Posco Group does not mean only humanoid robots," Im said. "Attaching sensors and intelligence to massive physical equipment so it can perform tasks more optimally is also a major pillar."

Replacing dangerous tasks with robots is another parallel effort. Robots are used for blast furnace equipment inspections, and in environments that place heavy burdens on workers — such as extreme heat or hazardous gases — specially modified robots take over tasks that humans once performed.

The ultimate vision of AX that Im has in mind is not a factory where individual AI systems replace humans one by one. It is a structure in which AI's computational power is added to the knowledge humans have built up over decades, and the judgments dispersed across processes and departments are connected so that the steel mill as a whole makes better decisions.

"I think of AI as intelligence and knowledge," he said. "Just as a person with high intelligence and deep knowledge does their work well, the same is true for a company. Receiving from AI the intelligence and knowledge needed to do what we originally do — better — that, I believe, is AX."

Im is scheduled to speak at the Herald Business Forum 2026, to be held Sept. 29 at the Shilla Hotel in Jung-gu, Seoul, where he will outline the group's AX strategy and his vision for the future of manufacturing AI in greater detail. Registration is available on the Herald Business Forum website.