□ In the wake of Chat GPT’s phenomenal influence worldwide, vague expectations of AI have been proven to be highly effective, setting off ‘the daily application’ of AI used by everyone with ease.
□ Global big tech companies are in a fierce, fast-paced competition* to preempt hyper-scale AI platforms backed by enormous computing power and vast capital.
* OpenAI (Chat GPT, Nov 2022)→Google (Bard, Feb 2023), Meta (LLAMA, Feb 2023)→OpenAI (GPT4, Mar 2023)
□ South Korea will also participate in the global competition in the leading group based on our own hyper-scale AI platforms ⇨ Support needs to be strengthened to related companies to secure the competitiveness of hyper-scale AIs and cultivate them into future strategic exports
* South Korea is included in four countries (USA, China, Israel, and South Korea) that have their own hyper-scale AI.
| Progress of Competitiveness Enhancement Strategy for hyper-scale AIs |
| “Conversation with Top AI Scholars” (Sep 2022) → “Korea’s Digital Strategy”(Sep 2022) → “Announcement of Plan for Everyday Use of AI and Industrial Sophistication” (Jan 2023) → Private-sector discussion series in response to hyper-scale AI, including “AI Strategy Summit” (Jan-Mar 2023)
|
| Ⅱ |
Influence of hyper-scale AIs |
✅ Hyper-scale AIs are expected to lead the destructive innovation of the digital industry itself and the internalization of AI across all sectors of the economy and society based on human-level language and visual abilities.
※ (Stanford HAI) Generative AI technology has reached an inflection point, and its social influence will be expanded all over the world. |
□ (Technology) AI performance is dramatically improved with transformers, reinforcement learning, etc., and AI is combined with search and word processors to evolve into extremely efficient tools.*
* (MS) Application of Copilot to Office software such as Word, Excel, and PPT (Mar 2023) → Upgraded to professional assistant level
□ (Industry) Hyper-scale AI has emerged as an essential factor in revolutionizing the productivity of all industries by assisting human knowledge services and helping to solve industrial challenges.
○ By making hyper-scale AIs learn additional expert data (fine-tuning), it is possible to develop AI services in various specialized fields.
□ (Society) As hyper-scale AIs are fully utilized, it is necessary to resolve AI risk factors, strengthen AI literacy, and prepare for the improvement of existing systems in terms of ➊reliability and ethics issues and ➋acceptance of AIs.
* Copyright debates over AI outputs, privacy issues in the data learning process, etc.
* Some concerns also raised about the rapid development of AI → American non-profit organization “Future of Life Institute” calls for a pause in the development of powerful AIs beyond GPT4, citing risks to humanity.
| Ⅲ |
Analysis of South Korea’s Competitiveness |
□ To date, the focus has been on creating a foundation for artificial intelligence by accumulating data (691 types of learning data), improving AI technology (89.1% compared to the US), and launching domestic AI semiconductors (NPU, PIM).
□ With these policy achievements and early developments in the private sector, South Korean companies have secured their own hyper-scale AI platforms* and started aiming for the global market.
* Released hyper-scale AIs: Naver (Clover), LG (Exaone), Kakao (KoGPT), SKT (A.), and KT (Mideum)
○ Companies that apply these platforms are competing with “specialized AI” that further learns company-specific data based on APIs provided by hyper-scale AI platforms.
* (Wrtn) released a writing and content creation platform, and (Upstage) released an AI chatbot “AskUp” capable of answering inquiries with photos.
| ⇨ There is enough potential to compete in the global market and achieve high performance based on the AI industry infrastructure that has been accumulated and domestic hyper-scale AI platforms. |
| Vision and Goals |
Acceleration of the transformation into a digital economy with smart artificial intelligence
|
| hyper-scale AI Platforms |
Korean platform to become No. 1 in the world |
| + Preemptively dominate the non-English global market |
|
| Application Services |
To aim for the world’s No. 1 in specialized sectors |
| by fostering collaborative ecosystems between companies |
|
| Strategy |
1. Expand technical and industrial infrastructure to support the development and sophistication of hyper-scale AI
2. Create an ecosystem for hyper-scale AI innovation, including the convergence of hyper-scale AIs in private and public sectors.
3. Establish nationwide institutional and cultural foundations for AI innovation
|
| ✅ Identify and focus on tasks that need to be added or improved to strengthen the competitiveness of hyper-scale AIs and foster the industry ecosystem based on the AI policy achievements to date. |
| Strategy 1. Expand technical and industrial infrastructure for hyper-scale AI |
□ (Data) Augment high-quality text data for the development of hyper-scale AIs (2023 and onward) New
○ Acquire 200 types of specialized text data (equivalent to 150,000 books) by field (34 types in 2023)
* Obtain text data with a focus on Korean, but gradually expand into non-English data (targeting Southeast Asia, Middle East, etc.)
○ Establish high-quality corpus* and applied corpus** in Korean (25 types in 2023, Ministry of Culture, Sports and Tourism) Reinforced
* Korean corpus with analytical information (named entities, sentiment, etc.)
** Translation of Korean to foreign languages, sign languages, braille, etc.
| As is |
+ |
To be |
| Projects regarding AI training data are centered on building labeled data |
Building (unlabeled) textual data for pre-training of hyper-scale AI |
□ (Technology) In addition to the technology development for deep learning improvement (2022-2026, KRW 265.5 billion), undertake new R&D to break through the limits of the current hyper-scale AIs (2024 onwards) New
| As is |
+ |
To be |
| Basic research to strengthen AI’s learning and communication skills |
Development of technology that breaks through the limitations of hyper-scale AIs, such as failure to reflect the latest information, false answers, etc. |
□ (Infrastructure) Develop core HW-SW to enable hyper-scale AI to utilize high-performance and low-power K-cloud based on domestic AI semiconductors (2023 onwards) New
○ Expand support of computing resources for hyper-scale AI to SMEs and universities (2023 onwards) Reinforced
| Strategy 2. Create an ecosystem for hyper-scale AI innovation |
□ (Super AI Flagship Project) Develop application services that innovate productivity, such as expert assistance and support, by applying hyper-scale AI to the top five professional fields in the private sector (2024 onwards) New
Super Giant AI Flagship Project for Top 5 Professional Fields in the Private Sector (Example)
| Law |
Medical |
Psychological counseling |
Culture and arts |
Academia and research |
| Supports filing legal complaints Suggests precedents and decisions |
Supports writing of medical opinions Recommends treatment methods |
Provides child care consultation Generates psychological analysis reports |
Provides Korean language education for foreigners interested in the Korean wave |
Suggests references and statistics Assists abstract writing |
□ (Public and administrative innovation) Implement hyper-scale AI-related public initiatives such as streamlining internal work and complaint responses (Ministry of Science and ICT, Ministry of the Interior and Safety) and hyper-scale AI patent search (Korea Intellectual Property Office) (2023 onwards) Reinforced
□ (Innovative Ecosystem of hyper-scale AI Industry) Operate “hyper-scale AI Council” to strengthen cooperation of digital companies, such as private-level investment and new service creation (2023 onwards) New
○ Comprehensively support the development of hyper-scale AI-based cloud services for small and medium-sized enterprises(2023 onwards), New, and implement a project to visualize hyper-scale AI in the metaverse as an intelligent assistant (2024 onwards) New
* (Intelligent teacher) Support for teaching and learning support; (Personal assistant) Schedule management and reservation; (Shopping assistant) Product recommendation, payment, and delivery
□ (Talent development) Along with the existing AI-SW workforce, further develop global-level talents specialized in the development and utilization of hyper-scale AI (2024-2027). New
○ Provide education programs centered on hyper-scale AI projects for job seekers with MS, etc. (2023 onwards, Ministry of Employment and Labor) Reinforced
○ Offer training for SW developers, teachers, and students to strengthen their ability to utilize hyper-scale AI and basic knowledge and ethics training for the general public (2023-2027, Ministry of Education and Ministry of Science and ICT) Reinforced
| As is |
+ |
To be |
| Nurturing of SW and AI talent at different levels (2023-2027) |
Capacity building for hyper-scale AI development and utilization (2023-2027) |
|
Official programs such as MA and Ph.D. in AI (65,000 participants) Non-official programs for job seekers and professionals (132,000 participants)
|
Training of hyper-scale AI professional talent and improvement of hyper-scale AI utilization training, AI capacity training for the general public (1 million participants) |
| Strategy 3. Establish nationwide institutional and cultural foundations for AI innovation |
□ (Social acceptance) Derive directions for improving regulations related to hyper-scale AI, discuss social issues such as education and security caused by the spread of hyper-scale AI, and seek countermeasures (2023 onwards) Reinforced
* Allowing ChatGPT for academic purposes, security issues and the risk of privacy leakage in the process of utilizing ChatGPT
○ Supplement the “Ethical Principles for AI in Education” by reviewing the use of hyper-scale AI (2023, Ministry of Education) and prepare guidelines for the use of hyper-scale AI in the public sector for government officials (2023, Ministry of the Interior and Safety)
□ (Reliability evaluation) Evaluate risk factors and performance of hyper-scale AI services developed by companies through authorized third-party organizations (2023 onwards, operated through private voluntary applications) New
○ Construct an implementation system for responding to social changes and improving acceptance (Apr 2023 onwards)
○ Detailed planning of new projects and reflection on budget (Apr 2023 onwards)
For further information, please contact the Public Relations Division (E-mail: msitpress@korea.kr) of the Ministry of Science and ICT.
Please refer to the attached PDF.