At the 2026 Taipei Game Show, Japanese companies Ubitus, Sin Chem, and Kneron outlined the current industry map of the transition from "generative AI" to "mobile AI" from the application layer, enterprise logic layer, and hardware computing power layer, respectively.
UbiOne: Giving NPCs a "soul," instead of just talking teleprompters.
"In the past, the most common problem we encountered with virtual idols or game NPCs was 'amnesia.' You would chat with them for ten minutes, and the next time you met them, they would ask you who you were." Hsu Chao-chin, Business Director of Ubita Taiwan, pointed out the pain point of current AI characters in his sharing.
UbiOne, proposed by Ubisoft, was created specifically to solve this problem. Hsu Chao-chin emphasized that UbiOne is not merely an "outer skin" connected to an LLM (Large Language Model); it's more like a complete "AI virtual human generation platform." By integrating vision, automatic speech recognition (ASR), and text-to-speech (TTS), UbiOne enables AI characters to possess the following three core capabilities:
• Long-term memory:AI can remember fans' preferences and past conversations, which is crucial for managing a VTuber character. It not only makes the audience feel valued and fosters emotional connection, but also makes the game content more immersive and interactive for players.
• Inner OS:This is the key to giving AI a "soul." Traditional chatbots are designed for a "you ask, I answer" approach, but UbiOne's AI-driven character has its own emotional judgment, sense of time, and can even initiate conversations. For example, when there's an awkward silence during a live stream, the AI will find a joke to join the conversation, or it will proactively tease or encourage players when they're stuck in a game.
• Multimodal interaction:Beyond just typing, the AI character can move in 3D space, make facial expressions, and even "see" the gifts or game screens in the live stream and react instantly.
Hsu Chao-chin showcased "Ubi-chan," an AI VTuber, at the event. This AI can stream continuously for 24 hours without fatigue and even adjust its tone and actions based on viewer comments. This means that future game marketing and social media management will be able to utilize such AI agents for round-the-clock support, significantly reducing labor costs while creating more stable interaction quality than real people.
Jingcheng Information: From chatbots to agents, AI is beginning to take over the "digital workforce".
If Ubita makes AI "fun," then Lin Ganwei, associate of Jingcheng Information, points out the "practical" value of AI agents in enterprise and game operations.
Lin Ganwei pointed out that we are now at a critical juncture, shifting from generative AI (generating content) to agent AI (taking action). While traditional large-scale language models can write poetry and create art, they often lack the ability to connect with external systems, leading to serious "knowledge silos" within enterprises.
The solution proposed by Jingcheng Information treats AI as a "digital employee." Through RAG (Retrieval Enhanced Generation) technology and API integration, AI agents can go beyond simply answering questions and execute complex task processes. For example, in customer service scenarios at game companies, AI can not only soothe players' emotions but also directly query the backend database to confirm server status and even automatically generate repair orders.
"Future AI will not only need to understand knowledge, but also process handling." Lin Ganwei emphasized that this kind of actionable AI will be able to transform the tacit knowledge scattered in Wikis, Line conversation records, or the minds of senior employees into executable standard operating procedures (SOPs), completely changing the efficiency of game development and operation.
Kneron Intelligence: Edge computing solves the fundamental problems of "latency" and "privacy".
Achieving such highly interactive and intelligent AI agents presents significant challenges due to the computational costs and latency issues. This is precisely the key point emphasized by Chen Yuchun, Assistant Vice President of Product Marketing at Kneron Intelligent Systems—Edge Computing.
Chen Yuchun analyzed that if all AI inferences rely on the cloud, not only will there be network latency leading to a poor gaming experience (for example, your AI teammates will be a step behind in healing), but there will also be high token costs and data privacy risks.
Kneron's NPU (Neural Processing Unit) technology boasts the ability to bring AI computing power down to end devices (such as PCs, in-vehicle systems, or AIoT devices). Through its Hybrid AI architecture, simple real-time inference calculations can be performed locally on the NPU, while complex logical inferences are uploaded to the cloud for execution. This not only achieves millisecond-level response speeds but also ensures that not all player dialogue data is uploaded to the cloud, thus balancing privacy and performance.
Analysis: The next stage for AI is "warm collaboration"
Based on the views of the three speakers, I believe that the role of AI in the gaming and technology industries has changed significantly since 2024.
UbiOne from Ubisoft demonstrates that the market is no longer satisfied with impersonal chatbots, but rather craves virtual companions with personality, memory, and initiative. This aligns perfectly with the "active AI" concept advocated by JINXIAN Info—AI must be able to perceive its environment and take proactive action. Kneron Intelligence's edge computing power provides the foundational infrastructure for this.
In the future game world, NPCs may no longer be just backdrops standing at the village entrance giving out quests, but AI agents that can remember your last injury, proactively plan tactics for you, and even "live" on their own after you log off.
