The Korean Intellectual Property Office (KIPO) has announced its proposed budget for 2025, revealing the office’s key investment priorities for the coming year. Among other objectives, the budget proposal highlights KIPO’s continuing efforts to effectively integrate artificial intelligence (AI) into its work processes.

 ▒  Key Focus Areas for 2025

The total budget of KRW 705.8 billion (approx. USD 511 million) represents a 0.6% increase from the previous year, and is set to be finalized by the National Assembly in December. Through the proposed budget, KIPO outlines four key focus areas for 2025:

• Leveraging patent big data: Expanding the use of global patent big data to better support national R&D planning, advance the development of emerging technologies, and prevent technology leaks.
• IP financing for startups and SMEs: Increasing support for IP-based financing options to help small and medium-sized enterprises (SMEs) and startups leverage their IP for growth.
• Strengthening export companies’ IP dispute resolution: Enhancing the capacity of Korean export companies to prevent and resolve IP disputes in other countries, and expanding efforts to curb the distribution of counterfeit goods online by harnessing AI.
• Developing a high-quality, AI-based examination support service: Building on KIPO’s partnership with LG AI Research to develop improved AI-based search capabilities for providing faster and more efficient examination services.

The latter AI project is in keeping with KIPO’s ’23-’27 roadmap for AI technology implementation, adopted in early 2023, which maps out the office’s ambition of “building the world’s best AI-based examination and trial system”.

 ▒  Partnership with LG AI Research

KIPO initially partnered with LG AI Research in 2023, with the goal of developing a ‘patent expert AI’ for use in prior art searches, classification, and summarization of patent documents. This lead to the development later that year of what is, according to KIPO, the world’s first hyperscale AI language model tailored specifically for patent administration.

The model was trained with 1.78 terabytes of patent-related data provided by KIPO—including patent gazette publications, office actions, machine-translated documents, CPC classifications, machine reading comprehension information, and collections of consultation case studies—based on LG AI Research’s hyperscale multimodal AI, EXAONE. Designed with 8.8 billion parameters, tailored to the environment and usage needs of KIPO, the model has already been successful in handling tasks such as patent document summarization, translation, and classification.

 ▒ Applying the new model

2025 will see the beginning in earnest of the next phase of the model’s development and application: building KIPO’s AI-based examination support system, through which the office ultimately plans to introduce new AI-based prior art search and trademark/design image search capabilities into its examination processes.

Given the increasing volume and complexity of patent applications, prior art search in particular is an area that stands to benefit greatly from the use of AI tools. As the first such tool based on an AI language model tailored for patent administration tasks, and boosted by the new investment, the model developed by KIPO and LG AI Research promises a step forward in this regard. Although 2025 is only the beginning of the next phase, patent applicants in Korea, as well as PCT applicants using KIPO as International Searching Authority, can look forward to more efficient, speedier, and more reliable services in the near future.

Written by Simon Voget