인간을 모방한 딥러닝 시스템으로
배우고 성장하는 인공지능

Although limited, artificial intelligence technology is already part of our daily lives. It is linked to various smart devices such as smartphones and computers, helping us search and utilize them more easily. The growth of AI is the synergetic result of advances in hardware, capable of processing large amount of information, such as GPUs (highperformance devices for processing graphics), and the development of software, such as deep learning algorithm. Deep learning, a type of machine learning, mimics the way the human brain works, which is complexly composed of brain neural networks. Professor Choi is also conducting research into developing high-performance deep learning models:

“In the past, machine learning modeled objects in the real world and the relationships among them based on mathematics and logic (symbolic techniques), and then made prediction, diagnosis, and deduction. These include knowledge (rule)-based systems, expert systems, ontology systems, and Bayesian networks. These days, deep learning models that mimic the human brain in software (sub-symbolic techniques) are being actively used. This is a method that anchors real world object characteristics data to the input and output of a deep learning model and trains it by repeatedly adjusting the anchored weight within the deep learning model, similar to the human brain. Through deep learning, artificial intelligence can learn large amounts of information more efficiently.”

When a deep learning system is used to learn enough high-quality data available for a specific task, artificial intelligence shows results that are similar to, or sometimes exceeding, those of humans. Professor Choi explained that artificial intelligence can easily assist or replace standardized, repetitive, and detailed tasks such as radiology analysis, analysis of large-scale legal precedents, document summarization, writing reports on specific topics, actuarial work, medication dispensing, and production process optimization and automation. However, this will take time. There are many fields in which practical, full-scale use of deep learning models to date is difficult because the logical explanations of the prediction, diagnosis, and deduction process can be abstruse; and in rare cases, it can produce fatal errors or disastrous results in important tasks. It is also difficult or too costly to obtain large-scale, high-quality learning data in many cases. For all these reasons, would be true to say that AI that completely learn all tasks on their own, like humans, are still a long way off.

Today, the whole world is paying attention to artificial intelligence. AI research is being conducted in various fields, including multimodal deep learning models (CLIP, DALL-E, etc.), large language models (BERT, BART, GPT, etc.), generative AI, and eXplainable AI (XAI).

Research is also in progress on the issues of AI Hallucination (a phenomenon in which an AI generates false information) and societal AI bias. Currently, the number of research results published at top-tier conferences, and by internationally renowned academic journals in the field is increasing rapidly in Korea. However, it is also true that there is a gap between the world’s best in terms of original theory and technology that can serve as the basis for future artificial intelligence technology, as well as in unique AI development. It is also unfortunate that much of the domestic research is still focused on deep learning applications and improving performances through tuning of existing models.

“Technology is developing so quickly in the field of artificial intelligence that papers pile up for you to read every morning. ChatGPT was not introduced to the general public until November 2022. To lead the field of artificial intelligence, we need insight into the future and conduct innovative research for it.”

Professor Yong Suk Choi plans to continue to advance his research results and strive to develop a “universal deep learning” model. Artificial intelligence has the ability to solve countless and diverse problems with consistent performance, and demonstrates plausible problem-solving ability for things that had never been heard of before. Therefore, developing AI comparable to human general intelligence is the ultimate goal for Professor Yong Suk Choi and many other researchers like him.

It has been a long time since a forklift replaced 100 people with shovels, but the world did not become unhappier as a result. People who worked with shovels now make forklifts or smartphones, or work in new industries with higher added value, such as entertainment. It has only been about 300 years since people began to understand the concept of “electromagnetic force,” which is an important foundation for modern human civilization. Professor Yong Suk Choi emphasized that it is impossible to guess how much we humans still do not know, living in a space that is merely a dot in this vast space and time of the universe:

“One day, if we can scientifically identify the secrets to human intelligence more clearly, future artificial intelligence will be able to achieve another level of innovation. To prepare for this, it is important to have a solid foundation in basic scientific theory, and to thoroughly understand and analyze the characteristics of not only classical AI models, but also the latest deep learning models optimized for individual tasks. In addition, I believe that conducting research on more general-purpose AI models by combining their outstanding points and achieving synergy effects is one of the ways to lead the AI era proactively and innovatively in the future.”