Kim Jae-il:I lead the AI Strategy Office at Netmarble. Previously, I was involved with trading agents at LINE INVESTMENT TECHNOLOGY, under LINE. Our AI Strategy Office is comprised of three teams: the Platform Development Team, the Agent Development Team, and the Media Development Team.

The Platform Development Team focuses on developing technologies related to game development and localisation, such as AI-based QA automation, AI translation, and AI voice. The Agent Development Team is tasked with building the internal AI agent development environment and infrastructure, with the goal of establishing the '1 Employee, 1 Agent' ecosystem. The Media Development Team primarily handles the creation of marketing assets using AI.

Hong Yun-seok:I belong to the Platform Development Team within Netmarble's AI Strategy Office. I've worked in the voice sector for a long time, and recently, I've been researching and developing AI technologies for trending topics in game development, such as translation and QA automation.

Kim Jae-il:It's a technology used for generating in-game character voices and translations. Previously, due to cost constraints, we could only support regions with large user bases. Now, with AI translation and voice capabilities, we can explore opportunities to introduce our games to markets we couldn't previously access. We're confident that our technology offers much richer emotional expression compared to publicly available external models.

Kim Jae-il:For instance, marketing artwork is distributed across various platforms, each with different size requirements. Our AI automatically handles these conversions. By automating such processes, artists in the marketing department can dedicate more time to content creation.

Kim Jae-il:One of our top priorities is QA automation. Previously, testers would manually play the game to design and verify test cases. Now, AI autonomously plays the game and tests various scenarios. This allows for a much higher level of testing, as it can uncover unexpected situations that humans might miss, ultimately contributing to improved game quality.

Hong Yun-seok:Games aren't just finished upon release; they undergo continuous updates and content additions. Recently, bugs have been emerging not from new content but from existing parts of the game. AI can help us check a larger volume of these areas more quickly, making it quite effective.

Kim Jae-il:Our biggest advantage is our openness to AI adoption. When facing challenges in development or operations that can be solved with AI, we have a very wide array of tools and options available.

Furthermore, unlike many other game companies, Netmarble operates a diverse portfolio of games across various genres, including MMORPGs, open-world, action, and casual titles. This extensive experience in applying AI and the accumulated know-how from those real-world applications are, I believe, Netmarble's core strengths.

Kim Jae-il:We're approaching this from two directions. Firstly, we are creating an environment where employees can build their own agents for detailed, personalised tasks. Secondly, we are developing and deploying common agents at an enterprise level that are essential for all employees.

For individual employee agents, we haven't designated a specific platform; instead, we provide the means for each department to use the tools they prefer for development. Fundamentally, all employees are provided with Gemini. If ChatGPT or other tools are needed, they can be accessed upon application and approval. Rather than imposing individual limits, we grant access based on work relevance.

Hong Yun-seok:A single agent cannot solve everything; it's a process of creating the agents each person needs. Ultimately, our goal is to integrate work into a single pipeline, with agents working collaboratively within that process.

Kim Jae-il:In the common domain, we have 'NDIS', an internal knowledge search service that is currently in service. It's a system that allows for chat-based searching of the vast knowledge accumulated within the company. This eliminates the need to ask colleagues or open separate documents to find desired information. It has significantly reduced the process of obtaining information by contacting various departments.

It also greatly reduces time spent on communication. Communication between individuals in planning, art, and development, each handling their respective tasks, is very costly. Accurately conveying the planner's intent is crucial, but the sheer volume and fragmentation of planning documents presented practical difficulties. Our AI agents have neatly organised this, making inter-departmental communication exceptionally accurate and swift.

Beyond development, agents enhance efficiency by producing well-organised documentation across all pipelines. In the operational phase, when a new build is released, clear operational notices are required, and there was a lot of time-consuming communication involved. This has also led to significant time savings.

We have multi-agent systems for development and QA, and we are also developing and providing agents to streamline operations in the business and finance departments. For the finance department, agents collect information on changes in relevant laws and regulations, rapidly relaying information on potential company impacts.

Kim Jae-il:It's challenging to provide a single agent that satisfies all development studios simultaneously. However, for several ongoing projects, we provide agent prototypes that development studios can then take, tune at the code level, and integrate into their workflows.

Hong Yun-seok:This requires collaboration between game development studios and teams like ours, who possess deep AI expertise. We are gradually working on co-creating agents with the development studios. Some teams are actively pursuing the goal of 'making everything an agent' with us.

Kim Jae-il:Variations (creating multiple versions of the same image, etc.) are heavily utilised in marketing asset creation. For internal use within development studios, such techniques fall under the unique domain of artists, so we haven't provided separate guidelines.

Kim Jae-il:We've integrated AI into operational announcements and customer support (CS). It's crucial to quickly gauge user trends and communicate immediately after a launch or update, and to have a process for promptly reflecting user feedback. It's impossible for humans to monitor user opinions by browsing every community. Especially since Netmarble operates games globally, not just in Korea, we need to collect and analyse data from numerous communities across various countries and languages. In this regard, we've implemented AI agents to better understand public sentiment and improve our operations.

Hong Yun-seok:We are also operating a bot detection AI. This AI actively patrols to detect and respond to bot activities. A new model for detection is released weekly, based on the latest information, to reflect changes in the user base and improve detection rates. Being deep learning-based, it can sensitively pick up on shifts in player behaviour. We also monitor reports from the community to further enhance its capabilities.

Kim Jae-il:We are striving to provide maximum support for AI tools that employees can utilise. To minimise discrepancies in usage among employees, we are conducting company-wide training and providing relevant information. We hold internal seminars to share related knowledge and discuss areas where we can provide assistance. Specifically, we differentiate between technical and non-technical roles, sharing how AI is being used, the extent of its applicability, and how collaboration is possible on company-wide projects with the AI Strategy Office.

Kim Jae-il:The company's stance is that by introducing AI, we aim to secure more time for developers, planners, and artists to concentrate on their core responsibilities. In that sense, we believe (AI adoption) will contribute to creating more polished games.

Hong Yun-seok:It seems users are experiencing disappointment and frustration with unverified content. Personally, I believe that if AI-generated content doesn't surpass human-created quality, it shouldn't be used. Therefore, we must advance by using generative AI to enhance both development speed and quality.

Kim Jae-il:Our agents are processed within our internal private cloud, ensuring safety from leakage and other security threats.

Kim Jae-il:We do not train foundation models ourselves; we utilise external models. Therefore, we don't foresee any related issues on that front.

Hong Yun-seok:For AI voice, we use data that has been verified as problem-free for model training. Legally, training data and data shared externally must be segregated. Therefore, internal data training is managed in a separate storage, and we ensure that externally shared data is free of copyright issues.

Kim Jae-il:It's challenging to provide numerical figures. When AI enables developers to focus on their core areas, even if development time isn't reduced, the outcome is increased quality. Therefore, quantifying this precisely is difficult.

However, we have observed that the initial concept development and prototyping phases have become significantly faster. For instance, agents are adept at analysing reference games to identify commonalities, differences, and factors contributing to success or failure. Yet, areas requiring unique creativity from planners remain an inherent domain that is not yet fully automatable by AI.

Kim Jae-il:Netmarble's direction is not to reduce headcount through AI. The core objective is to have AI support repetitive and mundane tasks, allowing developers, planners, and artists to focus more on their creative work.

Hong Yun-seok:I view AI not as a replacement for humans, but as a tool for enhancing productivity to create higher-quality games.

Kim Jae-il:The AI Strategy Office's goal is to support the development of high-quality games and establish a system that rapidly reflects public sentiment in operations, ultimately creating games that users will appreciate even more. Our objective is to build this foundation through AI.

Hong Yun-seok:At its core, a game's essence is fun, and AI is a tool to better realise that fun. We will continue to actively utilise AI technology to deliver better games, without compromising the fundamental nature of gaming.