Professor Kyoung Mu Lee says, “Large AI models needs to take in visual data to be complete.”(Yonhap News, 2023.06.19)
“LG’s large AI model, ‘EXAONE’, has advantages that language models such as ChatGPT do not possess.”
Chair Professor Kyoung Mu Lee, SNU
“When humans engage in cognitive activities, 90% of the information the brain receives comes from visual inputs. This implies that without the ability to interpret and comprehend visual information, creating true AI is challenging.”
On the 18th (local time), Professor Lee Kyung-mu, Chair Professor of the Department of Electrical and Computer Engineering at Seoul National University, explained the importance of "image captioning" AI, which distinguishes and describes images, during an interview with accompanying journalists two ahead of the opening of the Conference on Computer Vision and Pattern Recognition (CVPR) 2023 conference held at the Vancouver Convention Centre in Canada.
Professor Lee is a distinguished scholar in the field of computer vision and is a Fellow of the Institute of Electrical and Electronics Engineers (IEEE), which hosts the CVPR conference. Last year, Seoul National University appointed him as a Chair Professor in recognition of his outstanding international achievements.
As a professor at Seoul National University's AI Graduate School, he has been actively collaborating with LG AI Research since last year. Together with LG, he established a joint research center at the university, and he has been dedicated to the development of image captioning technology using LG's large AI model, ‘EXAONE’.
He explained, "Exaone has a unique advantage compared to models created by other big tech companies, which focus on language. Exaone specializes in visual information, offering capabilities that AI models such ChatGPT cannot perform." He also mentioned that they are working on challenging tasks, such as video captioning technology.
Professor Lee added that "Through visual information, the model can carry out all types of inferences, including the determining the object, the state of objects, relationships between objects, and predictions on the objects." He further added, "Ultimately, for robots to act and interpret like humans, the most critical factor is whether the robot can express visual information in language that humans can understand."
On this day, where he participated in LG's Image Captioning AI Workshop held the CVPR conference, Lee said that "organizing workshops on specialized fields in the competitive CVPR conference is a recognition of being the leader of a field." He further explained that there is growing interest in visual AI in the academic community.
He also emphasized that actively researching specialized fields of AI could be a survival strategy in the competitive environment of the era of large AI models.
He further explained, "There are even evaluations that even AI models involving tremendous amount of resources like ChatGPT may find it difficult to compete with companies like Google or Microsoft within the United States," and added, "In the field of artificial general intelligence, given the uncertainty of the benefits even with substantial investment, there is a need to adopt a specialized strategy."
In particular, Professor Lee suggested that through such strategies, in the long term, it is necessary to expand research towards 'Multimodal AI' capable of processing various information such as vision and language simultaneously, as well as 'Embedded AI technology'.
He stated, "To develop comprehensive AI technology that makes judgments based on various types of information, the 'Multimodal Layer' technology, which allows one system to handle everything like our brain, will be a crucial aspect in the future." He also predicted, "Since AI has to be used in real physical systems such as robots, the technology has to be embedded such systems to become a practical technology."
Source: https://ee.snu.ac.kr/community/news?bm=v&bbsidx=53792
Translated by: Do-Hyung Kim, English Editor of the Department of Electrical and Computer Engineering, kimdohyung@snu.ac.kr
