"Securing the strongest artificial intelligence (AI) model is not the key; rather, obtaining high-quality data is crucial for the success of 'agentic AI' (an autonomous AI system that sets its own goals, makes judgments, and acts)."
On the 27th, Chu Min-sik, head of AI Advanced Technology Research Institute at LG CNS, noted this at the largest domestic tech conference 'SMARTCLOUD SHOW 2025' held in the Westin Josun Hotel in Jung-gu, Seoul, under the theme 'AI and the Future.'
Chu pointed out that "90% of performance issues with agentic AI are related to data quality," stating that "corporations do not invest in infrastructure to acquire high-quality data." He added, "We need to increase investment in infrastructure focused on data quality and invest in building robust data pipelines and performance monitoring systems," warning that "without securing high-quality data, 80% of projects may fail, leading to a potential AI catastrophe wasting billions of dollars."
Chu indicated that blind expectations about AI are a hindrance to its practical application. He said, "The blind application of trends by company executives, misunderstandings about the purpose of use, and poor communication with technicians are the causes of failure," urging that "there needs to be sharing of opinions between executives and technical practitioners from the outset of terminology construction."
Chu also mentioned that accuracy can be improved through data learning. He stated, "General AI agents have a response accuracy level of 50-65%, but learning 'meta-knowledge and domain-specific terminology in management and accounting can increase the accuracy to 80-90%,' emphasizing that the establishment of an ontology model, which structures and expresses knowledge, is key."
Meanwhile, LG CNS recently unveiled its enterprise agentic AI platform 'AgenticWorks.' AgenticWorks is an agentic AI platform that supports corporations in designing, building, operating, and managing AI agents directly. LG CNS provides specialized AI models across different industries and value chains, secured through over 500 AX (AI transformation) projects.