When machines offer wisdom, what happens to our own?

KATHMANDU: When we were children, elders would jokingly tell us, “Here’s Rs 0.5 (five paisa). Go and buy some wisdom!” Today, that childhood joke has become a reality: intelligence can, in a sense, be bought. That purchasable intelligence is artificial intelligence (AI), a form of intelligence created by machines rather than naturally developed by humans.

An “AI factory” is a special computing infrastructure that processes massive amounts of data to develop, train, and operate artificial intelligence models, in the exact same way a traditional factory transforms raw materials into finished goods. This massive factory of data, algorithms, and powerful machines is pushing human civilization into an unprecedented era. As this storm-like new era shakes established structures and assumptions, questions about who will stand where, what status everyone will hold, and whose existence will be threatened are becoming increasingly intense. To grasp the core of this mysterious subject, it is imperative to first understand how artificial intelligence is built, how it works, and how it reaches us.

The overall system of artificial intelligence can be understood primarily in three layers: infrastructure, model, and application. The mutual coordination of all three layers makes the development, operation, and utilization of AI possible, and that “purchasable wisdom” ultimately reaches our mobile phones, computers, or other devices for consumption.

Infrastructure layer

Just as a traditional factory needs buildings and machines, an “AI factory” also requires massive physical infrastructure, at the center of which are data centers. To build such infrastructure, a reliable power supply, sufficient land, and a high-capacity cooling system are required first. Next comes “compute,” meaning the ability to process massive amounts of data at high speed and perform complex mathematical calculations. High-capacity special chips such as GPUs and TPUs are used for complex AI-related calculations. The capability to design such cutting-edge chips is currently concentrated in a few major technology companies such as Nvidia and Google. These chips are extremely expensive, and their demand is also high.

Servers equipped with thousands of such chips are connected to each other via high-speed networks to build a system that functions like a single massive supercomputer. These servers, the software operating and coordinating them, data storage systems, and communication networks connecting with the outside world together form the main technical structure of the infrastructure layer of the intelligence factory.

When operating such powerful machines continuously, AI data centers consume an extremely large amount of electricity. Cutting-edge cooling systems are mandatory to control the heat generated by the machines and prevent them from overheating.

Billions of dollars can be invested in building massive AI infrastructure. Today, massive investments are being made to build and expand such infrastructure in countries including the United States, China, and India, alongside major tech companies such as Google, Microsoft, and Amazon. Thus, artificial intelligence infrastructure is becoming a strategic issue linked not only to technology but also to energy, capital, and national capacity.

Model layer

With the help of this massive infrastructure, the “model,” considered the brain of AI, is prepared by training it on large volumes of data. Large Language Models (LLMs), which answer questions, write, and solve various problems according to human prompts, are prime examples of this. Besides these, there are also models that process and create various types of content, including images, audio, and video. Such versatile AI models are broadly called “Foundation Models.” Companies such as OpenAI, Anthropic, Moonshot AI, and Google DeepMind are the major developers of this layer.

Four primary resources are required to build an AI model, meaning an “artificial brain”:

Data: Massive, diverse, and high-quality data used to train the model.

Algorithms and architecture: Mathematical methods and structures capable of learning by identifying patterns hidden within the data.

Skilled workforce: High technical experts such as AI scientists, mathematicians, and machine learning engineers.

Access to infrastructure: Access to powerful supercomputing systems for model training.

Training a state-of-the-art frontier model requires continuously running supercomputing systems integrated with tens of thousands of GPUs and TPUs for a long period. The training and development of such models can cost anywhere from tens of millions to billions of dollars.

The core intellectual capability of artificial intelligence lies within this model itself. Based on patterns learned from data during training, it processes user prompts and generates responses or results accordingly. The process by which a trained model works in this manner is called “inference.” To provide AI services to users worldwide around the clock, the computing systems operating these models must be continuously available. This also requires massive computational infrastructure and energy. Therefore, not only the development of advanced AI models but also their operation is costly.

Application layer

The application is the medium that connects the artificial intelligence model to the user. From popular AI platforms like ChatGPT, Gemini, Kimi, and Claude to special apps and websites built for various fields such as education, law, and health, all fall under this layer. Users typically input their prompts into such applications via the internet, which the respective model processes to provide an answer or result. Ultimately, through these applications, that “purchasable wisdom” reaches our mobile phones, computers, or other devices.

Although building the infrastructure and model layers requires billions of dollars, cutting-edge hardware, and specialized scientists, the door to the “application layer” is comparatively open to everyone. At this layer, even with modest investment, solo entrepreneurs or small startups can commercialize their creative capacity and innovative thinking.