1.6조 들인 공공부문 AI 데이터, 중복·미흡에 ‘쓸 만한 자료’ 적다
The government spent ₩1.6 trillion to build artificial intelligence (AI) learning data, but the content turned out to be similar or unsuitable for AI learning, according to a government audit. This highlights shortcomings in the government's approach to transitioning to AI, with data management being the first weak point. According to the audit released on the 12th, the Ministry of Science, ICT and Future Planning (MOTP) spent ₩1.6328 trillion from 2017 to November last year to build 908 types of AI learning data.
However, 26 individual organizations, such as Seoul City and the Korea National Highway & Transportation Safety Agency, individually built learning data without prior sharing or coordination, resulting in many cases of building data similar to existing data. For example, the Sejong City Office invested ₩130 million to build 9,000 photos of waste, which turned out to be similar to data prepared by the MOTP with a 2020 budget of ₩16 billion.
The MOTP also invested ₩233 billion in 2021-2023 to build data for drugs, oral, and autonomous driving, but three organizations spent ₩84 million on similar data during the same period. Even the dental photos data set of ₩1,000 photos, built by the Ministry of Health and Welfare, was identical to MOTP data. The audit recommended that individual public institutions first examine the possibility of using existing data when building new data, and establish a pre-review system for sharing and coordinating each organization's construction plans.
Additionally, data quality management was inadequate during the individual institutions' learning data construction process. For instance, the 'CCTV Learning Data Set' built by the Korea National Highway & Transportation Safety Agency was found to be missing vehicle-specific location information, making it unsuitable for autonomous driving development.
Furthermore, four out of the 20 institutions with the highest achievement rates in building learning data had no quality management standards at all.
Written by urgent.news from Hankyoreh's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
