(Seoul = Yonhap News) Reporter Sang-yong Han = As generative artificial intelligence (AI) spreads throughout daily life and industry, not only companies but also the government are paying attention to specialized AI tailored to specific fields, so-called 'vertical AI'.
In a situation where American big tech has already taken the lead in the general-purpose AI model market, some analysis suggests that latecomers are trying to differentiate themselves by combining industry-specific data and each company's own field know-how.
◇ General-purpose AI is ‘generalist’… ‘Specialist’ delving into industry and work
The English word 'vertical' in the dictionary means vertical, and in the industrial field, it refers to a strategy that focuses on a specific industry or market.
When AI is added to this word, it usually refers to 'specialized AI' designed to support or process actual tasks in a specific industry or job, unlike general-purpose AI that covers a wide range of topics.
It is a method of combining specialized data, work rules, and field systems of a specific industry.
In Korea, it is mainly discussed as industry- and task-specific AI that is applied to medicine, finance, manufacturing, and security.
If general-purpose AI is a 'generalist' that is used in general areas, specialized AI can be seen as a 'specialist' that demonstrates specialized capabilities in a specific field.
While general-purpose AI provides the basic ability to answer questions and summarize and write documents, vertical AI is closer to a form that reflects industry-specific terminology, regulations, document formats, and decision-making procedures so that it can be used immediately in the field.
For example, manufacturing AI refers to AI that specializes in analyzing abnormal signs based on facility data and production process information.
Medical AI goes beyond searching medical papers to summarize medical records or helps medical staff write documents, and financial AI supports consultation and report writing by reflecting internal regulations and investor protection principles.
The fierce competition in the global general-purpose model market is behind the growing interest in specialized AI from both companies and governments in Korea.
Google, OpenAI, Antropic, etc. are releasing the latest large language models (LLM) one after another, leveraging their enormous capital, large-scale data, and graphics processing unit (GPU) infrastructure.
For latecomers to compete head-to-head with these companies based solely on model size or general performance, the starting line itself is bound to be different.
Accordingly, domestic AI companies are focusing on increasing the utility of specific fields by utilizing the service operation experience, industrial data, regulatory response capabilities, and business system interconnection capabilities they have accumulated.
◇ Instead of going head-to-head with the general-purpose model, ‘industrial data’… Government also specializes in security AI
The government's cybersecurity-specific AI foundation model development project is also an example of being evaluated as a form of vertical AI.
The government recently selected the Naver Cloud Consortium as a business operator, and the key to this project is to create an AI model that supports cyber threat analysis and response by reflecting malware and vulnerability information, security control records, domestic security systems and operating environments, etc.
The Naver Cloud Consortium plans to develop an AI model tailored to the domestic security environment and demonstrate it at key national facilities and major industrial sites.
However, just because specialized AI stands out in a specific field does not mean that it always outperforms general-purpose AI.
The characteristic of vertical AI is to increase accuracy and practicality in specific tasks by adding expert knowledge, work flow, and verification devices to the foundation of general-purpose AI.
In order to improve the performance of specialized AI, the language understanding, reasoning, and information processing capabilities of general-purpose AI, which are its foundation, must also be supported.
This is also the background for the government to promote 'AI for all', which secures a general-purpose AI-based model, and the development of cyber security-specific AI.
In fields such as medicine, finance, public affairs, and security where AI mistakes can lead to major losses, protecting personal and confidential information, managing access rights, verifying the basis for answers, and final human review are important.
Therefore, future AI competitiveness is expected to depend not only on the size of the LLM, but also on whether reliable data and work know-how for each industry, ability to link on-site systems, and responsible verification system have been secured.
