Can AI succeed in disrupting games this time?
Text: "Chinese Entrepreneur" reporter Li Xiaotian
Edited by Zhang Hao
Header image source: Visual China
Ma Mingtao has a "mysterious confidence" in Cai Haoyu.
Although his AI game startup project "Aperture Editor" did not go as smoothly as expected, he was really ignited again. A month ago, he saw a sentence posted by Cai Haoyu on the social platform X, the former chairman of miHoYo and the genius game boss who created "Genshin Impact" is the "spiritual idol" of many game entrepreneurs.
"99% of developers who create games on their own terms just to satisfy their ideas can consider switching careers, and the game world of the future will be made up of 0.0001% talented teams to create something that has never been done before."
AI is once again subverting the game, which has been described by the media as "tyrannical". Cai Haoyu, who has not updated his news for a long time, suddenly spoke out, and the focus of the industry directly turned to Anuttacon, which broke out earlier, which is said to be a new entrepreneurial project he launched in United States. According to the official website, it aims to "create new, innovative, intelligent, in-depth, and interesting virtual world experiences, and AGI (Artificial General Intelligence) products that explore the full potential of artificial intelligence."
However, AI games have somewhat "died down" in China. Zhuang Minghao, a heavy game enthusiast and former vice president of Matrix Partners Partners, is very cautious about Cai Haoyu's projects and views, and no one dares to make a definitive conclusion about AI games until a definitive product comes out. In his opinion, several AI game teams that received financing in the first two years are now dead, and major game manufacturers have been slow to make decisions, and the investment in AI seems to be far more significant than practical.
"Whether it's Google's AI game engine or what Cai Haoyu said, it looks subversive, but it's hard to say when it will really land." Zhuang Minghao said.
Cai Haoyu does not have many "followers", and his views have indeed detonated the game circle, but the vast majority of them "can't understand". Even if Ma Mingtao is like him, he only subjectively inferred: "Their resources and technology are not on the same order of magnitude as ours, if you want to study hard, you should be able to come out with something creative." ”
He doesn't think that he is on the same track as Cai Haoyu, "Cai Haoyu should be making large-scale high-standard games, and we are focusing on making game production for novices who don't understand AI technology." The original intention of his business was to realize that ordinary people could automatically generate a graphic game by "feeding" an idea to AI.
Xiao Mo also feels that Cai Haoyu's new project is a bit "virtual", he once joined a team of AI comics. The underlying logic of comics and games is the same, but static comics require a lot of manual assistance to repair, and the game is much more complex, how to subvert it?
"For Cai Haoyu, his first big dream as a 'secondary two teenager' ten years ago has been realized, and now he wants to chase another big dream, but in the current situation, it is indeed not easy to complete." Xiao Mo said.
The "not easy" he mentioned refers to the fact that AI technology is not yet at the level of making high-standard games at this stage, and what it can complete is only simple graphic games.
In 2020, NVIDIA's AI game studio GameGAN used AI to fully recreate the classic game "Pac-Man", a two-dimensional, uncomplicated game that was considered to represent the highest level at the time.
This year, Google's first AI game engine, GameNGen, generated a small shooter game "Doom" with 0 code, and the graphics look very rough, even behind the game 20 years ago, and there is not much innovation in gameplay. In Zhuang Minghao's view, there is no feeling of "shining in front of the eyes".
Source: Screenshot of the game Doom
And Anuttacon, which seems to be ambitious, doesn't seem to care about the "wall" that AI games currently hit. Cai Haoyu's "all-star" team includes Tong Xin, a former Microsoft global partner, Wu Xiaojian, a former senior research scientist at Microsoft, and Wang Yuyang, a former vice president of Bilibili.
"Chinese Entrepreneur" contacted Cai Haoyu himself several times based on the new project plan, but did not receive a reply as of press time.
It's struggling
Ma Mingtao started his own business in February last year, and the trigger point was the explosive spread of ChatGPT around the world. He originally ran his own script killing shop and bar, and was completely attracted to the ideal virtual world built by ChatGPT. "Only AI can accomplish the world I want in my head." Ma Mingtao said.
He sold his shop, raised a sum of money, and found 2 like-minded partners to set up Aperture Future Technology together, and finally formed a team of 5 people.
However, the road to AI games was far more complicated than they imagined, and there was a Ph.D. specializing in AI technology in the initial team, and after working together for nearly 3 months, he quit. "He assessed that the development cycle of our products was too long, but there were other reasons as well. As a result, our entire front-end work was suspended, and we couldn't continue development for 2 months. Ma Mingtao was very helpless, the simulator product was originally planned to be launched in February this year, but there were too many bugs.
Meow Jito Studio, which was also established because of the belief that AI can change games, is also a scale of 5 people. Founded in Shanghai in 2023, the initial members of the team include Alibaba's NPL algorithm engineer and NetEase Games' planner. But the development cycle did not meet expectations, and the self-developed game "Speech Project" was originally scheduled to be officially launched in June this year, but now the release date of the update on Steam (the world's largest game publishing platform) has been changed to the fourth quarter of 2024.
Li Chi, the founder of Meow Jito, posted on Xiaohongshu in March this year, hoping to find financing smoothly. By June, Meow Jito was on the industry resource docking platform of the self-media "Game Teahouse", and the demand was still "looking for distribution and investment".
Compared with the Meow Jito team with mature industry experience, Ma Mingtao is more like a "reckless man": "We have seen a lot of investors during the period of entrepreneurship, but on the one hand, they want to see mature works, and we have not yet; On the other hand, the general evaluation of the project is that 'the idea is top, but the team is not up to the mark'. He can't help it, this is a "chicken lays eggs, eggs lay chickens" problem, AI games are still too new, the talent pool is far from enough, investors don't invest money, and they will definitely not be able to form an excellent team.
The beta test product of Aperture Editor has attracted some novel creators to use it spontaneously, but in Ma Mingtao's words, it is difficult for the generated graphic game works to reach the industrial level, and the creators need to spend more time polishing and modifying the content of the game itself.
Aperture Editor interface Source: Screenshot of Xiaohongshu
In March last year, Xiao Mo had great expectations for the fact that "AI can easily generate comic works that can be delivered". The beginning was not bad, the team got the investment intention of Instant and Baidu, but in less than half a year, he discovered the shortcomings of AI comics: "It can achieve a passing score for a single picture, but several pictures with only a passing score are combined to fail the whole work." ”
Therefore, he quit early, "From the perspective of AI, art and games are actually the same, and I am quite pessimistic about the future of AI games as a whole." ”
Without investment, Ma Mingtao could only continue to maintain a small team, and he had to force himself to change from a complete AI technology layman to a semi-expert who could read code and fix bugs.
But he still has a passion for AI, "Even if I am the only one left in the team in the future, I will insist on realizing what I want through AI little by little." ”
Every game company is labeling itself as "AI", but until last year, "AI games" failed to become an independent statistical unit in major investment and financing media. Most of the companies that can get financing are AI technology companies, and there are very few game production companies.
The main battlefield of AI games
At the moment, it seems that there is only one way to go - AI NPCs (in the game, non-player characters controlled by AI).
So much so that from 2021 to the present, InWorld, the star company of this track, has received more than 1 billion yuan in financing, including Microsoft, Intel, Disney, and Samsung. Embarrassingly, the inflection point of InWorld is to shift from the very beginning of AI game production to the production of AI NPCs. It has established cooperation with game companies including NetEase Games and R-Star Game Studio (the developer of the "Grand Theft Auto" series), and has completely changed from "Party A" to "Party B".
In the eyes of many practitioners in the game industry, AI NPCs exist more on the user experience side and do not affect the production process of the game too much. And this is quite controversial, some players say, if you team up to fight monsters are all "robots", then what's the fun of playing the game? However, this has still become the main battlefield of various game companies, and it is the first "position" occupied by AI.
In 2017, NetEase established the Fuxi AI Artificial Intelligence Lab, the main direction of which is AI NPC, and the products "Against the Cold" and "Naraka: Infinite" have become classic cases that the industry will cite when it comes to AI games.
Wu Feng, who used to work for NetEase Fuxi, mentioned that there are more than 200 AI NPC characters in "Against the Cold" that can interact with players.
An AI NPC in Against the Cold. Source: "Against the Cold" official Weibo
"It eats a large part of the dividends of AI games, on the one hand, it creates a new game experience, where players can communicate freely with NPCs that are not scripted in a 'small society' of agents, and can also have a long-term emotional connection with these intelligent NPCs. At the same time, on the market side, it also plays a good marketing role, when the product can provide an experience that breaks through the established limitations, it will have a good marketing effect. Wu Feng said.
In NetEase's "Naraka: Bladepoint" mobile game launched this summer, it can already support "AI teammates". Unlike their sluggish robot teammates, AI teammates listen to commands, maintain voice communication with players in real time, and even customize their voices.
Tencent's AI LAB studio is also in the same direction. The AI product "Jue Yi" in the Go game has achieved four consecutive championships in professional competitions, and the online game function is currently only open to professional chess players, and ordinary chess players can use its "review" function.
The AI product "Juewu", which is based on "Honor of Kings", plays more of a role as a novice guide and trial. In 2019, Juewu made its debut at ChinaJoy, playing 504 games against human players and losing only 1 game. In the entertainment session before the Glory of Kings World Championship that year, Juewu played against 5 professional players without a single defeat.
Similarly, miHoYo laid out its own AI sector in 2018 and established a "reverse entropy" research department, and the "Luming" created by it has become a well-known virtual anchor at Station B. In the two-dimensional mobile game "Star Dome Railway", there are also a large number of AI-assisted generated plot dialogues.
The entry points chosen by major game manufacturers for AI are highly similar, and in the eyes of many practitioners, this status quo is very "skinny". Using AI to develop a game from start to finish may be a fruitless and thankless task at this stage.
The hard-to-replicate Genshin Impact
The problem that everyone sees is very unified, that is, AI is not enough to create works above the level of industrial red line.
"It's okay to have AI generate a single image, but it can't handle multiple portraits and large scenes. If you need to deliver a four-square comic, each one can only do 60%, the combination is 60%× 60%× 60%×60%, and there will be a lot of places that need to be manually repaired in the later stage. Xiao Mo said.
Ma Mingtao encountered the same problem: "Whether it is character painting, background or text, AI cannot achieve perfect presentation. If the creator still needs more time to modify and polish in the later stage, the efficiency will be lowered. ”
In the process of AI creation, sometimes the problems that arise are "inexplicable". Li Xinglei, CEO of Haiyue Technology, which makes the translation of Haiwang text, once mentioned to reporters that where a text was originally an ellipsis, AI will play its own role and interpret the content of hundreds of words according to the context, and it will have to proofread and review it later.
In the view of game investor Ming Hao, even if major manufacturers are pursuing AI NPCs that are closer to real people, in fact, no matter how real AI teammates are, they can't provide the emotional value that real teammates can bring, "If I know that your teammates are AI, then I don't even bother to spray them." ”
"In general, AI is a test of the ability of the back, and it will also require you to have a high level of professionalism. It's impossible to get a newcomer with zero foundation to do it. "There used to be legends that the murals in the popular "Black Myth: Wukong" game were made with AI." If they had this technical strength, they would have stopped making games and specialized in AI companies a long time ago. Xiao Mo said.
Lu Hongyan, chairman and general manager of Gigabit Games, said in response to a question about whether AI will change the game industry at an extraordinary shareholders' meeting: "If AI reaches that point, it may not only be games, but many industries will be changed. We still use AI as a tool, and AI native games have not seen scientific research results that are particularly close to practicality. ”
As for the imagination space of AI games in the next 2 to 3 years, Zhuang Minghao is still very cautious: "Maybe various functions will enter the product pipeline, and AI penetration and AI functions in existing game projects will be seen more." And the pure AI game that I started today will be officially available to users at that point in time, but as the first batch of attempts, it is difficult to have much expectation. ”
"Our vision is to create a virtual world in which one billion people around the world want to live by 2030." miHoYo wrote on its official website.
Not surprisingly, Cai Haoyu will continue the underlying logic of this vision in the new project. They are planning a big move to disrupt the industry, but can the concept of AI games alone replicate the new situation that Genshin Impact created on the whole network in 2020?
No one dares to answer this question.