As the generative AI boom slows, China's AAA title Black Myth: Wukong has reignited discussions on how the gaming industry can harness AI.
According to AWS representatives, gaming has long served as a testing ground for AI innovations, with clients—including Chinese companies—using AI to create NPCs, build virtual societies, analyze public sentiment, and detect abusive language. However, for companies with user bases in the hundreds of millions, careful planning is essential before deploying AI on such a large scale.
AI: game changer or just enhancement?
At a recent forum, AWS Senior Business Development Manager Yi-Ying Lin shared that even before the rise of generative AI, game developers were experimenting with AI for tasks such as recommending in-game items, testing player reactions, and creating NPC companions.
Techniques such as neural networks and reinforcement learning have been key to these advancements.
Lin emphasized that understanding player preferences is critical. For example, while some players may easily navigate the challenging levels of Black Myth: Wukong, others may struggle. AI can help by simulating the behavior of 100 different types of players, allowing developers to detect potential issues before launch.
The goal isn't necessarily to create smarter NPCs, but rather to diversify NPC personalities to reflect the wide range of players' experiences.
While NPC companions can enhance the gaming experience, some players find AI-driven companions too predictable and lacking in challenge. Game developers are now working towards creating NPCs that mimic human behavior as much as possible.
Parametrix.ai, for example, launched "Living Chang'an City" in March 2023, a virtual AI society with diverse NPC personalities.
Lin pointed out that since the rise of generative AI in 2023, many gaming companies have been exploring its potential. The big question is whether GenAI will simply enhance gaming or fundamentally change how games are played.
Currently, GenAI can be used for text-to-image generation, text-to-speech conversations, task scene creation, improving internal knowledge management systems, and public sentiment summary analysis.
AI-enabled marketing and community management
Moonton Games, a Shanghai-based developer known for competitive games is a client of AWS. The studio has a vast player base in Southeast Asia, with an estimated 100 million monthly active users.
For a company of this size, careful planning and strong commitment are essential before implementing AI, as even minor errors could affect millions of players.
Since mid-2023, AWS has assisted Moonton in exploring application scenarios, such as leveraging large language models (LLM) for accelerated public sentiment analysis, recognizing offensive posts, and localizing marketing efforts.
Popular games often build their influence through community-driven engagement, and regular competitions help strengthen user loyalty. To stay relevant, it's crucial to recognize cultural trends and regional slang, something LLMs are key to identifying.
LLMs can also assist in translating languages and identifying offensive language. For risk management, a reporting mechanism is in place for filtering out abusive language during competitive matches, which is then reviewed by customer service for further action.
Sentiment analysis is also critical to business strategy, such as designing characters' appearances and personalities in response to regional cultural preferences. For instance, Moonton has released characters based on Indonesia's Gatotkaca and the Philippines' hero Lapu-Lapu.
Lin highlighted that with thousands of daily player comments on social media, manual sampling and analysis are not feasible. Using LLMs allows every comment to be tokenized and analyzed in greater depth. This is especially important for annual e-sports events, where sentiment analysis is critical.
She concluded by noting that generative AI is still in its very early stages. "The race has just begun," Lin said, adding that how the marathon of long-term competition unfolds remains to be seen.
