LG AI discovers hair-loss treatment material in a day, eyes manufacturing next

LG's science-focused AI model has discovered a new hair-care material by analyzing 420,000 candidate substances in a single day. LG Household & Health Care, the group's cosmetics and consumer goods affiliate, plans to apply the material to its scalp-care brand Dr.Groot.

The material, named Rhamsydil, helps ease female pattern hair loss by creating a scalp environment more conducive to hair growth.

"We have selected industrial challenges based on where AI can have the greatest effect and scalability, and have driven innovative change as a result," said Lim Woo-hyung, co-head of LG AI Research, in a keynote address. "That approach has allowed us to solve more than 100 industrial problems to date."

The model behind the discovery, Exaone Discovery, is tailored to the sciences, designing new materials and predicting synthesis outcomes. It is one of several models LG AI Research has built for specific industries. Exaone Tabular, aimed at manufacturing, reads and predicts process conditions and quality data. Exaone Omni-Inspect is an autonomous visual inspection model that adapts to changes in a product's or process's appearance without needing to be retrained each time. Exaone Business Intelligence, built for finance, uses multiple AI agents working together to handle data analysis, inference, forecasting and explanation, drawing on Exaone Forecast, a foundation model that tracks how information changes over time. Exaone 4.5 is the company's general-purpose large language model. LG AI Research has also begun developing a robot foundation model that would let robots judge and act in physical environments, a step toward autonomous factories.

Working with affiliates including LG Innotek and LG Display, the research lab found that the harder challenge was not building a model or deploying it, but keeping it working afterward.

"Manufacturing conditions change constantly," said Lee Hwa-young, head of AI business development at LG AI Research. "The problem shifts, operating conditions shift — the on-site situation is so dynamic in ways that break models trained on a fixed snapshot of data."

LG's manufacturing models are designed to run on minimal GPU capacity, making it easier to deploy them on-site rather than through the cloud. A major source of hesitation around AI adoption in manufacturing is the risk of domain knowledge leaking overseas, since routing data through an external cloud provider raises that exposure. Running models on-premise, inside a factory's own infrastructure, avoids that risk while also cutting costs, since larger models require more GPUs and drive up expenses as they scale.

BY LEE JAE-LIM [[email protected]]