Few people today are unfamiliar with artificial intelligence (AI). People ask questions of ChatGPT and create text and images with generative AI. Hardly a day passes without the news introducing a new AI model. Yet the more information about AI proliferates, the harder it becomes to understand its essence. This is because semiconductors, the cloud, communication networks, data centers, and even energy all now appear under the name of the "AI industry."
Following only fragmentary news makes it easy to fall into the error of touching an elephant's trunk and thinking it represents the whole elephant. Non-experts in particular tend to accept a specific technology or sensational piece of information that frequently appears in the news as the entirety of AI. But AI no longer refers to a single algorithm or large language model alone.
AI is a vast ecosystem in which semiconductors that process data, the cloud and data centers that provide computing resources, energy infrastructure that supplies enormous amounts of power and cools the heat, and the communication networks that link all of these elements operate together. This is precisely why semiconductor, power, cooling, and telecommunications companies have recently drawn attention as AI-related firms. To understand the big picture of the AI era, one must first examine the structure that connects these elements, rather than individual technologies.
If the internet connected information and the cloud connected computing resources, the network of the future will connect AI to AI. In an era where multiple AI agents divide roles and collaborate to process a single request, the network must decide in real time which AI to assign a task to, where inference should take place for the fastest and most efficient result, and which information should be delivered first.
The goal of communication is changing as well. Computers of the past recognized information as entirely different even if only a single bit changed, but today's AI can infer core meaning from incomplete information and decide on its next action. Therefore, in an era where AI becomes a major agent of communication, the important task moves beyond "how many bits can be delivered accurately" to "how efficiently the necessary meaning can be shared." Future communication will expand from a technology that transmits bits to a technology that connects intelligence and shares meaning.
This shift is also leading in a direction that complements the limitations of centralized AI. Gathering all data and computing resources in one place can improve management efficiency, but it brings with it problems such as privacy protection, data sovereignty, communication costs, and the risk of system failure.
Federated learning is a representative technology that can reduce these limitations. This is because it leaves data with each institution or device while sharing only the learning results and knowledge. What matters here is not simply building a single model with superior performance. It is more important to design how participants with different data, computing capabilities, and network environments can collaborate safely and fairly.
The core question of the AI industry is also expanding from "how large a model should be built" to "who participates in the learning, how trust is maintained, and how accumulated knowledge is shared." Ultimately, the competitiveness of AI is likely to be determined not only by the performance of individual models but also by the connection structures that allow diverse actors to collaborate.
The combination of AI and networks does not remain only on the ground. Low-orbit satellites can develop beyond mere data relay equipment into distributed computing nodes that perform AI learning and inference. If ground-based clouds and satellite networks collaborate organically, a new infrastructure will be built to provide AI services beyond the constraints of location.
If communication was one of the foundations that made the growth of AI possible, communication will also be the key foundation that further advances AI going forward. This is why it is difficult to fully understand the structure of future industry if AI and networks are viewed only as separate technologies.
South Korea, on the basis of high-speed internet and world-class mobile communication infrastructure, led the smartphone era and drove the growth of the mobile service and platform industries. Now, in the AI era, we must ask the same question. Can we secure future competitiveness simply by developing a better AI model?
He is:
- Ph.D. from Korea Advanced Institute of Science and Technology (KAIST)
- Former Visiting Researcher at City University of Hong Kong
- Former Principal Researcher at Samsung Electronics DMC R&D Center
- Former Visiting Fellow at Princeton University, U.S.
- Ph.D. from Korea Advanced Institute of Science and Technology (KAIST)
- Former Visiting Researcher at City University of Hong Kong
- Former Principal Researcher at Samsung Electronics DMC R&D Center
- Former Visiting Fellow at Princeton University, U.S.
