Some people earn five - figure sums a day, while others are chased and scolded by players. So, how exactly should one combine AI with games?
Text | Feng Ruiyang
Editor | Liu Shiwu
The implementation of AI in the gaming industry has been incredibly rapid. Game companies of all sizes have integrated AI into their games.
Tencent Games has transformed the virtual character "Jili" in "Peacekeeper Elite" into an AI - driven tactical advisor. NetEase Games has introduced CopilotAI as a combat partner in the mobile game "Naraka: Bladepoint", and deployed the intelligent NPC "Shen Qiusuo" based on DeepSeek in another game "The Chinese Paladin: Swords of Legends".
Everything seems to be moving in a positive direction, but there have also been quite a few AI blunders in the gaming industry.
Besides some so - called "air projects" that suddenly emerged and are controversial for "AI fraud", a series of actions by the well - known game company "Activision" have also sparked dissatisfaction among players on overseas gaming forums.
In fact, while AI is creating the prosperity of "N new games per day" and also giving rise to the curse of homogenization like "cyber canned goods", developers have to recalibrate the balance between efficiency and craftsmanship.
In this article, we will discuss several AI - integrated games from different perspectives to find the answer to the question "How exactly should AI + gaming be done?"
When AI + Gaming Becomes the Focus
Recently, developer @levelsio's flight simulation game developed using AI tools has attracted tens of millions of onlookers and earned tens of thousands of dollars in just a few days. The game's code was generated using Cursor and Gork, and the developer only invested a few hours of manual work.
Image source: Internet
This news has received extensive attention and discussion in both the gaming and technology fields. For the technology field, it is a good proof of the current powerful capabilities of artificial intelligence. For the gaming field, some game makers are amazed that the originally cumbersome game development can be so conveniently solved directly through AI, indicating that the gaming industry is about to change.
In fact, using large language models to generate code for game production has become a trend, and there have been many similar projects both at home and abroad recently.
Elon Musk has retweeted four X posts about simple games made with Grok within a month. These games have only a small number of pages and levels, but almost all of their code and textures are directly generated by large models and are somewhat playable.
Image source: Internet
In contrast, Tencent Games' "Peacekeeper Elite" has integrated DeepSeek R1 into the game, bringing a more user - centric experience. The character Jili in the game can have conversations with players outside the game. These conversations can be about in - game information such as the effects of in - game equipment or other information like local weather.
NetEase Games' "The Chinese Paladin: Swords of Legends" has integrated AI into the game's NPCs. Players can interact with NPCs on the map. According to the official preview, the NPCs will be able to understand players' emotions and make more relevant responses in the future.
Image source: "Peacekeeper Elite"
Meanwhile, the well - known game company Activision is being criticized by players for using AI.
In several recent posts on Activision's Instagram account, there were multiple AI - generated preview images for games such as "Call of Duty: Zombies Defense" and "Guitar Hero (Mobile)". These images have a very obvious AI style in terms of painting and have many problems in details, such as meaningless shapes in the musical notes.
Image source: X
In the comment section, a large number of players expressed their strong dissatisfaction. Players believe that a well - known and established game company should not use AI to make posters. Moreover, since these AI posters involve classic IPs, long - time players with an emotional attachment to these games feel strongly offended.
As the incident has developed, players have expressed their anger on various social media platforms, and several media outlets on foreign platforms have followed up.
Players' comments on this incident on Reddit
However, since these are just promotional images for unlaunched games, although they have attracted extensive discussion and criticism from players, they have not affected the actual products or their revenues. In contrast, when "White Night Aurora" used AI to draw a game holiday image before, it aroused more anger from players.
In 2023, Tencent's second - generation game "White Night Aurora" posted a Valentine's Day holiday image on foreign social media, which was suspected by players to have traces of AI generation.
The image posted by the painter himself, which has more obvious AI traces than the official game version
In the image, there are obvious sticking phenomena in the characters' hair, arms, and feet during AI drawing, and even the left and right feet are drawn in reverse. Some players questioned, "We spent 648 yuan, and this is what you show us?" However, the painter did not directly admit or apologize but instead refuted, and the incident once made it onto the hot search.
To this day, whenever an image of a second - generation game is suspected of using AI, "White Night Aurora" is still often brought up as a topic of discussion.
It can be seen that different ways of using AI will lead to extremely different feedback from players. Therefore, how to make good use of AI has become the key.
What Are the Approaches to AI + Gaming?
When discussing AI + gaming, the most taboo thing is to regard "AI" as a label or a whole. In fact, different AIs have different principles and functions, corresponding to different usage scenarios and bringing different effects to games.
The most basic way of utilization is to use the general basic capabilities provided by large language models, such as providing inspiration, writing copy, and generating code.
A relatively simple way of use is to let large models assist in designing game content and writing game copy. When humans complete game content and copy, they need inspiration, while large language models have a much larger information reserve and faster information - processing speed than humans. Therefore, they can provide a large number of "seemingly reasonable" ideas in batches, greatly improving efficiency.
Image source: Feedback from large models
Let GPT - o1 design a game based on an idea, and a few hundred words of description can be quickly expanded into several pages of detailed content.
Some teams are also trying to integrate large language models into games so that players can have conversations with large models during the game. This is also the way of AI + gaming adopted by many teams at present.
However, if large language models are simply placed in games without adjustment, they will be separated from other parts of the game - even causing players to wonder, "Why don't I switch to use other more mature and convenient models?" To make large language models fully play a role as a certain character in the game, fine - tuning of the models is required, which requires a large amount of resource investment for training, and there are not many successful cases in this regard at present.
Another relatively intuitive way of use is to use AI to generate images or 3D models. Human painters take a long time and charge high fees for painting. A commercial painting may cost tens of thousands of yuan, while the average cost of AI - generated images is only a few cents to a few dollars.
Although the generated results are not always good, due to the extremely low generation cost, dozens or even hundreds of images can be generated at once and then selected. For games with low requirements for quality and consistency, the currently generated images and models can already meet the usage standards.
Anime character image generated by Midjourney
3D model generated by Tripo
In addition, some products are exploring the development of independent AI tools for games. Microsoft's MUSE, released in February this year, was trained using more than one billion images and controller actions and can generate coherent game images and actions.
Such independent new tools require a huge amount of technical investment. The paper related to the WHAM (World and Human Action Model) technology used behind this work was even published in the top - tier scientific journal "Nature".
Image source: Internet
However, due to the current immaturity of AI technology, there will still be many problems when using it directly.
Firstly, the generation results are not stable enough. For use during the game development stage, the problems caused by instability can still be offset by refreshing a few more times. But if AI is directly integrated into the game, a single output that does not match the game may cause great damage to the player's experience.
Secondly, it is difficult for the content generated by AI to enter the workflow and cooperate with other links. For example, the code written by AI is different from human - accustomed code, making it difficult to modify and maintain; the process of AI - generated images is also inconsistent with the human creative process. Most tools only support exporting single - layer bitmap formats without layer separation, resulting in the need for partial redrawing when modifying. Even adjusting the content generated by AI may consume as much time and energy as remaking it.
Moreover, high - quality generation, especially for images and modeling, requires a large amount of computing power. Currently, this computing power is often provided by game companies themselves or through third - party services. Therefore, the cost is huge. If this computing power is not provided by the user's device, each use is a cost for the game. If there is no guarantee that the generated results can be well utilized every time, it is difficult to make the revenue higher than the cost.
In addition, there are some problems that are not currently highly regarded but will surely be faced as AI is used more frequently in more aspects. For example, the copyright issues of AI - generated content and the homogenization of creativity caused by only piecing together existing content. These have become potential risks for AI + gaming and urgently need to be addressed.
What Other Ways Are There for AI + Gaming?
Of course, there are some promising new directions for AI + gaming. Some attempts are doing things that previous technologies couldn't do and things closely related to game content.
For example, upgrading the human - controlled characters in the game, which could only participate in the game in a fixed way, into AI teammates that can understand players' instructions and cooperate with them.
At the Gamescom 2024, "Break Point" demonstrated its voice - commanded FPS AI F.A.C.U.L. In the real - machine demonstration, the AI teammate can understand complex instructions from players, such as "Use the tree in front as cover", "Look for a green box", "Throw a smoke grenade after I open the door", and act accordingly.
Compared with early games like "Desert Storm 2: Back to Baghdad" or the more classic "Tom Clancy's Rainbow Six: Vegas", where players could only issue simple commands like "Attack", "Defend", "Take position", "Follow" to Bot teammates using buttons, F.A.C.U.L. is obviously more AI - enabled.
