ETRI WEBZINE

It is evaluated that the Korean artificial intelligence R&D project by domestic researchers has revitalized the artificial intelligence(AI) ecosystem in Korea and has greatly accelerated the settlement of Korean AI. As a result, it is said that it is preventing the market erosion by foreign AI solutions and establishing itself as a world-class artificial intelligence leader.

Electronics and Telecommunications Research Institute(ETRI) announced that the Korean artificial intelligence “Exobrain” project greatly contributed to the creation of a domestic artificial intelligence ecosystem in Korea with ▲692 domestic and foreign papers ▲463.5 domestic and foreign patent applications ▲KRW 6.2 billion in technology transfer and KRW 14.5 billion in commercialization ▲2 KOSDAQ listed companies, etc.

Exobrain^* project was conducted for 10 years in 3 phases from May 2013 to February this year, with ▲general and 1 detailed task by ETRI ▲2 detailed task by Saltlux ▲3 detailed task by KAIST^**.

* ExoBrain: The name of the Korean artificial intelligence project, which means an artificial brain outside of my body. http://exobrain.kr/
** (Overall/1 detailed task) Development of intelligent evolutionary WiseQA platform technology for human knowledge augmentation service (ETRI)
(2 detailed task) WiseKB: Development of self-learning knowledge base and reasoning technology based on big data understanding (Saltlux)
(3 detailed task) Symbolic Approach-based human-simulating self-learning intelligence source technology (KAIST). From step 3, it was integrated into 1 detailed task

Exobrain researchers developed world-class Korean artificial intelligence technologies, such as deep question-answering technology that finds answers to natural language questions in searched documents, Korean language analysis technology, deep learning language model-based application technology, and knowledge based reasoning technology.

In November 2016, it won the victory in <Contest! Exobrain>, scholarship quiz contest with human quiz kings, proving the possibility of securing the sovereignty of domestic AI.

In addition, ETRI has distributed language intelligence technology and machine training data developed through the project through the open API and data service portal(http://aiopen.etri.re.kr) since 2017 for the first time as a domestic R&D project.

The research team has so far disclosed 18 types of language processing technologies, 14 types of learning data, and 2 types of language models. As of the end of 2022, the published data had been used more than 61 million times by 2,349 institutions.

In 2019, the Korean language model KorBERT^*, which has 4.5 % better performance than the language model BERT developed by Google, was developed and released for the first time in Korea.

In addition, Exobrain researchers achieved KRW 6.2 billion in license fees and KRW 14.5 billion in commercialization through 171 technology transfers over 10 years. Through this, it has greatly contributed to preventing foreign AI solutions such as Google and IBM from encroaching on the domestic market.

* KorBERT: Korean Bidirectional Encoder Representations from Transformers. A language model optimized for Korean learned at ETRI. Many companies are utilizing deep learning language model technology based on morphological analysis, such as the method proposed by ETRI.

The achievements of the organizations participating in the research are also notable. In July 2020, Saltlux, 2 detailed task organizer, and Mindslab, 1 detailed task co-organizer, were listed on the KOSDAQ in November 2021, respectively.

In addition, in the Exobrain project, efforts were made to secure original technology through standardization^* related to language intelligence data and technology, and submission of numerous patents and papers, etc.^**

Since April of last year, ETRI researchers have been conducting research on the development of artificial intelligence SW that can support decision-making by experts in various fields by expanding the explanatory attribute of natural language processing technology as a follow-up project to the Exobrain project.^***

Unlike ChatGPT, which is easy to use and versatile for the general public, it is expected to show differentiated results by presenting a natural language-based explanation(correct answer) specialized in a specific domain such as law and finance together with a reliable ground for the explanation.

Lee Hyun-kyu, a PM in the Institute for Information and Communications Technology Planning and Evaluation(IITP) said, “Exobrain is a representative technology that is most widely used in the commercialization market as a result of national projects, and has made a great contribution to the development of domestic artificial intelligence technology. In the future, it would be desirable to promote it as a mission-oriented project that enhances the current performance or insufficient functions.”

Dr. Bae Kyeong-man in ETRI, who is in charge of the Exobrain Project, said, “We will make efforts so that the achievements of the Exobrain project accumulated over the past 10 years will not be lost and will be continuously utilized, and through follow-up projects with the project members, we will work hard to develop the technology to advance the Exobrain technology one step further.”

This project was promoted as an innovative growth engine project by the Ministry of Science and ICT and the Institute of Information and Communication Planning and Evaluation(IITP), and in February 2023, the 10-year grand project was completed.

** Adopted 8 domestic and 7 international standards, adopted 88 from international contributions, adopted 19 from domestic contributions
** Registered 12 international patents and 106 domestic patents, applied 69.5 international patents, 276 domestic patents
Submitted 87.1 cases of SCI/E level, 605.6 cases of non-SCI (international+domestic)
*** (4 detailed task) Development of artificial intelligence technology to support decision-making by experts who can explain the reasons/grounds for the judgment results of the subject of expertise (2022.4.1.~2026.12.31. (4 years and 9 months))