[Bbang Gubneun Tajaegi] "When AI Gains a Body, the World Changes"

From "Explaining AI" to "Moving AI"

VLA Ushers in a Major Transformation in Knowledge and Physical Labor

"Computers excel at tasks difficult for humans, but struggle with things that are simple and trivial for us." This is a quote from robotics engineer Hans Moravec. Even when we wake up groggy in the morning, we naturally brush our teeth and get dressed. Until now, however, computers have had to break down these actions into countless sentences and data points in order to understand them. Organizing the infinite variables of the real world into 'rules' has been virtually impossible. This is why, until just a few years ago, robots showcased to the public often moved awkwardly and clumsily.

That so-called 'Moravec's paradox' is beginning to be upended. The stars of 'CES 2026', the world's largest electronics and IT exhibition held last month, were robots that could box naturally, climb stairs, and carry goods in factories-so-called 'physical artificial intelligence (physical AI)'. In their book "Physical AI Mega Trend," Choi Hongseop, CEO of MindAI, and team leader Won Mireu identify the visual-language-action (VLA) model as the core engine of this trend. While large language models (LLMs) and visual language models (VLMs) were limited to language and visual understanding, VLA connects these to actual actions, according to the authors.

Just as humans learn about the world through senses and experience, AI is also evolving through a 'data-driven' approach by training on vast amounts of image, text, and behavioral data. The transformation of the autonomous driving industry illustrates this well. In 2022 and 2023, the industry faced a crisis due to the 'long-tail problem'-the need to handle an almost infinite number of exceptional situations. In response, Tesla, led by Elon Musk, shifted its strategy from explicitly coding rules to training on massive driving video datasets so that the system could make its own judgments. Camera-based autonomous driving has become symbolic of a software-centric strategy, in contrast to hardware-first approaches.

The book defines this as 'the next chapter after ChatGPT.' Just as generative AI has rapidly reshaped the knowledge work market over the past few years, physical labor is also expected to undergo structural change through physical AI. In a world facing labor shortages and aging populations, physical AI is not a choice but an inevitability. The authors emphasize that physical AI should be seen not just as a robotics technology but as a value chain. The entire ecosystem, from hardware like actuators and sensors, to AI models and semiconductors, operates in close coordination.

Industries expected to adopt physical AI rapidly include agriculture, defense, construction, and manufacturing. In addition to reducing labor costs, it can shorten construction periods and improve safety, further increasing its industrial impact. Of course, certain challenges remain. There are still technical barriers, such as the limitations of battery energy density, heat generation and power consumption, and on-device chip optimization. Ultimately, the book argues that the progress of physical AI should be evaluated not by "how human-like it appears," but by "how useful it is to humans."

The strategic approaches of the United States and China are also intriguing. The United States is expanding its ecosystem based on big tech-centered platforms, capital, and research infrastructure such as graphics processing units (GPUs). China, on the other hand, is catching up quickly with its state-led manufacturing base and vast domestic market. In this context, the book explores South Korea's potential as well. It suggests that Korea's capabilities in precision manufacturing and system integration could remain competitive in the physical AI era, and calls for bold deregulation and the establishment of standards. Just as ChatGPT suddenly became central to daily life, embodied AI could also experience explosive proliferation once it reaches a tipping point. However, the speed and direction of this change will depend on industrial structure, policy, and capital choices. The South Korean government also presented a vision of becoming the 'world's leading nation in physical AI' last year.

If an era is approaching where both knowledge work and physical labor are being redefined, where does human value remain? The book stresses the importance of "uniquely human qualities that machines cannot imitate." It is about having a personal standard for what makes a story emotionally resonant, and what constitutes an ethically right decision. In an age where technology quickly becomes the market, it is crucial not merely to acquire specific skills, but to cultivate a fundamental capacity for learning anything. This is the strength needed to discern the true value amid the endless output generated by AI.

Physical AI Mega Trend｜Written by Choi Hongseop and Won Mireu｜Wisdomhouse｜340 pages

This content was produced with the assistance of AI translation services.