'Everyone’s AI' Project Turns Into a 6-Way Race Among SKT, KT, Kakao, and Others

The government’s public AI service project, titled "Everyone’s AI," has entered a competitive phase after 6 operators and consortiums submitted proposals. The Ministry of Science and ICT announced the status of proposal submissions for participation in the "Everyone’s AI" project on the 19th.

A total of 6 operators or consortiums submitted proposals. The competition is a 6-way race among the 6 consortiums (operators), with SKT, MKD, ETRI, Xenon, Kakao, and KT listed as lead organizations.

SK Telecom, KT, Kakao, ETRI, MKD, and Xenon were named as lead operators. Including participating organizations, the total number of companies and institutions taking part in the public call is 62.

This project is not simply about building a general-purpose chatbot. It is being viewed as a large-scale AI ecosystem project that combines domestic AI models, computing infrastructure, data, and specialized services.

The Ministry of Science and ICT will select the operators by the end of this month after reviewing the suitability of submitted documents, conducting a written evaluation, and holding presentation reviews. After selection, the ministry plans to sign agreements and run beta services, with a public service launch targeted within the year.

Among the contenders, SK Telecom has the largest number of participating organizations, with 19 in its consortium. The participants are Goodoc, Nota, the Korea Single-Parent Family Support Center for the Elderly, Liner, Soundable Health Korea, SelectStar, Shinhan Card, Alfred, SK Broadband, AtoSys, Aim Intelligence, ELICE Group, Channel Corporation, T Map Mobility, Persona AI, Hana Card, the National Tax Accountants Association Computerization Corporation, Habit Factory, and Hashed Ventures.

SK Telecom’s strategy is to combine a general-purpose AI service with medical, financial, tax, education, mobility, and customer service functions. To this end, SK Telecom has structured the consortium to bring together participating organizations from a range of fields.

SK Telecom’s strengths include its own AI model and experience in consumer services based on A. It also has the potential to link with SK Broadband’s fixed-line subscriber base and T Map Mobility’s mobility data. With the participation of SelectStar and Aim Intelligence, the consortium also includes components for AI reliability and safety.

KT formed a consortium with 16 participating organizations. KT’s consortium includes Daum, DeepSearch, Lablup, Rebellions, Mathpresso, Motif Technologies, Musinsa, Vessel AI Korea, BC Card, Saltlux, ActionPower, Upstage, NC AI, Zigbang, Connectwave, and the Korean Educational Broadcasting System (EBS). The consortium’s scope covers AI models, semiconductors, cloud operations tools, financial services, education services, real estate services, fashion services, and e-commerce services.

Within the KT consortium, Rebellions is participating as an AI semiconductor company, while Upstage, NC AI, and Motif Technologies are model companies, and Lablup and Vessel AI Korea are AI infrastructure operations companies. As a result, the KT consortium has a structure linking model development with inference infrastructure operations.

Meanwhile, Kakao has formed a separate camp with LG affiliates, medical, financial, and AI semiconductor companies. Kakao’s side has 14 participating organizations: Day1 Company, Lunit, Seoul National University Hospital, Shinhan Bank, LG AI Research, LG CNS, LG Uplus, LG Electronics, Worksphere, Wonderful Platform, Jarvis & Villains, Kakao Enterprise, Crowdworks, and FuriosaAI.

Based on Kakao’s public platform touchpoints, the proposed structure combines LG AI Research’s AI models, LG CNS and Kakao Enterprise’s enterprise AI implementation capabilities, LG Uplus’s telecommunications services, and LG Electronics’ home appliance ecosystem. In addition, Lunit and Seoul National University Hospital may handle medical expertise, Shinhan Bank and Jarvis & Villains may handle financial and tax services, and AI semiconductors for inference from FuriosaAI are included.

This structure can be summarized as a vertically connected setup spanning services, models, and infrastructure. It adds the roles of LG affiliates and several specialized companies and institutions to Kakao’s platform touchpoints, linking services, models, and infrastructure into a single axis.

ETRI has chosen to form a consortium with Megazone, Bidraft, SureSoftTech, Wisenut, Eastaid, Polaris Office, and the Korea Productivity Center. Although the number of participating organizations is smaller than those in the major telecom and platform companies, the consortium is structured around cloud services, AI search and chatbot services, office software, AI human services, and software verification capabilities, and is being analyzed as a practical productivity-service framework.

ETRI also has plans to bring in additional companies in areas such as commerce and travel. It has said it plans to expand the scope of services and broader application areas going forward.

MKD and Xenon submitted standalone proposals without forming separate consortiums. This standalone strategy is interpreted as aiming to quickly deliver a consistent service centered on its own AI models and platform. By contrast, large consortiums combine services from dozens of companies.

The advantage of a large consortium lies in its ability to accommodate a wide range of everyday services on a single platform. However, the complexity of data linkage and service integration among participating organizations could become a burden. Standalone operators may be disadvantaged in terms of service breadth and public touchpoints, but they have the strengths of making development decisions quickly and implementing a unified user experience more easily. Future evaluations are expected to focus not so much on consortium size as on how specifically the roles of participating organizations are designed.

The goal of the Everyone’s AI project is to make AI services accessible to anyone in the country with low cost and low usage burden. This year, the government plans to unveil a general-purpose AI chatbot and a public AI agent. After that, it is envisioning personalized services such as asset management, learning coaching, and retirement planning, with the long-term goal of creating "1 AI agent per person for the entire public."

The government plans to distribute up to 512 of the secured NVIDIA B200 GPUs to the 2 or 3 selected operators and to begin supporting the costs required to run public services starting next year. The government’s plan is to select 2 or 3 operators and provide computing resources on a split basis.

A key factor in evaluating the project is the composition of the AI models. Participating operators must build a certain percentage of the system with domestic AI models, and they must use models developed by other Korean companies as well as their own lead operator models. This follows the project’s aim of avoiding the spread of a single company’s model and creating a market in which multiple domestic models are actually used in public services.

However, the government believes the final measure of success cannot be limited to parameter scale or benchmark scores alone. The project must secure functions that ordinary people can use in daily life, and the criteria for success also include error control in sensitive areas such as finance, healthcare, and education, privacy leak prevention, and the ability to bear substantial inference costs.

Because free or low-cost services are structured so that GPU, power, and network costs rise along with user growth, concerns have been raised that, without a sustainable revenue and operating model even after government support is reduced, the project could end up as a one-off proof of concept.

The evaluation scheduled for the end of this month will reflect these considerations, viewing not only the strength of the technology itself but also the extent of domestic AI model usage, service composition that users can feel in their daily lives, the ability to handle large numbers of users, privacy and AI safety management, role sharing among participating organizations, and sustainability after government support as key variables. The competition in this public call is also centered less on who has the strongest AI model and more on who can connect AI to people’s everyday lives most broadly and reliably.

In the end, the future of Everyone’s AI depends on whether it remains a showcase for domestic models or becomes a daily-life platform that can compete with foreign AI services, and the deciding factors will be the quality of the actual service and the ability to attract users after selection.

Source: IT DAILY · Lee Jae-young
Original: https://www.itdaily.kr/news/articleView.html?idxno=241064

References

• https://www.itdaily.kr/news/articleView.html?idxno=241064