Interview
[Interview] NAIRL Director: “Physical AI Must Prove Itself One Task at a Time”
Kee-Eung Kim, Director of National AI Research Lab and Chair Professor at KAIST Kim Jaechul Graduate School of AI, spoke with Edaily about where physical AI and AI agents actually stand. Director Kim said expectations of a “ChatGPT moment” for physical AI have run ahead of the technology, and that general-purpose robots will not arrive all at once. Physical AI will spread as it proves useful in defined tasks, one at a time, in fields such as agriculture and manufacturing.
The reason is data. Language models can scrape the internet, but robots generate data only by moving through the physical world, and a robot that performs well in training can still falter when lighting or object placement changes. Simulators and world models are starting to close that gap with synthetic data, which is why the realistic path is to prove usefulness in specific settings first. The same holds for AI agents: rules for what cooperating agents may share are not yet written, and in tasks that run for days, small errors at each step compound into failure at the end.
With this understanding, National AI Research Lab pursues research that makes AI work reliably in the world as it is, from robots in greenhouses to agents running multi-day tasks, and serves as a hub for Korea’s AI research.
Link: https://www.edaily.co.kr/News/Read?newsId=01223446645646376&mediaCodeNo=257
