LG’s Expert AI Strategy: From Answering Questions to Solving Industrial Problems

LG AI Research used its September 14 AI Talk Concert at LG Sciencepark in Seoul to make a practical case for “Expert AI”: AI tailored to the data, decisions and constraints of particular industries. The institute presented applications in manufacturing, science and finance, alongside plans for systems that could take a more active role in physical operations.814

The shift: judge results in the field, not just answers

In his keynote, LG AI Research co-head Lim Woo-hyung said the mission was not merely to build a strong model, but to solve problems industries had struggled with for years. He argued that AI working in real industrial settings must account for many variables—including rare exceptions—and demonstrate tangible results.36 That is the distinction behind LG’s strategy: a plausible response is less valuable than a prediction or recommendation that holds up in a real workflow.

LG is not abandoning general-purpose models. Its stated approach is to keep improving their performance while developing deeper expertise for selected domains.35

Manufacturing: predict quality and adapt inspections

LG introduced EXAONE Tabular to analyze production data and predict process conditions and product quality. EXAONE Omni-Inspect is designed to inspect products even when a process or product’s appearance changes, without requiring retraining, according to LG.414 Reporting on the event also described the use of agents to select and classify data and train models, with the aim of identifying production issues earlier.4

The longer-term ambition goes beyond individual tools. LG described plans for robot-foundation-model-based factories that operate as more integrated intelligent systems.9 That is a proposed direction, not evidence that fully autonomous factories are already operating.

Science: a reported discovery and a laboratory plan

LG presented EXAONE Discovery as a system for predicting material properties and designing new materials.4 For a concrete example, LG said work with affiliate LG Household & Health Care screened 420,000 candidate substances in one day and identified Rhamsydil for potential use in Dr.Groot hair-care products.214 Identifying a candidate ingredient does not establish that it is an approved or clinically proven hair-loss treatment.

LG also outlined an autonomous-laboratory plan in which AI-guided experiments and robotics would support continuous research and learning.3536 That proposed laboratory should be distinguished from the completed screening result.

Finance: analysis that explains its forecasts

EXAONE Business Intelligence applies the same domain-specific approach to finance. LG describes it as a system of collaborating agents for data analysis, reasoning, prediction and explanation.3950 In July, LG AI Research and Koscom signed an agreement to combine EXAONE BI with financial-market data for a Korea-focused analysis service; LG said a related product would provide share-price trend scores with commentary explaining their basis.4650 Those are descriptions of the service and partnership, not proof that its forecasts are reliable.

What the record shows—and what comes next

LG AI Research said it had addressed more than 100 industry-specific challenges since its founding in December 2020. Event reporting also cited 368 conference papers and 1,080 patent filings.1333 Those figures describe the institute’s reported work; they do not, by themselves, measure the performance of every Expert AI application.

The event’s message was therefore a dual-track one: put specialized AI to work on problems that can be tested in industry now, while continuing to develop broader intelligence. Guest speaker George Cameron, co-founder of Artificial Analysis, argued that South Korea should pursue artificial general intelligence as part of its AI vision because stronger baseline intelligence would help AI work in the field.17 For LG, the immediate test remains whether its systems deliver dependable results beyond a demonstration.