Insights into Web3 Games: How AI Creates a New Genshin Impact

Author: GalaxyBlitz

The Thought Provoked by the Phenomenal "Genshin Impact"

Many people have noticed that the traditional patterns in the gaming industry have "failed" when faced with "Genshin Impact." In the past mobile game industry, the emergence of "blockbusters" in categories like card games, APRG, MMO, SLG, MOBA, and battle royale quickly guided the industry in a certain "direction," allowing many companies to reap substantial profits.

In the face of "Genshin Impact," the vast majority of people feel a tremendous sense of powerlessness. Most want to copy the homework, only to realize they have only an elementary school education, while "Genshin Impact" is already a member of the academic elite. miHoYo's open world, content packaging, large investments, industrialization, multi-platform support, global distribution, IP, and branding have left most people in the gaming industry in a position where they can't even understand the answers.

Investors are even more anxious. miHoYo has grown from a dormitory startup to a gaming giant with only a 1 million yuan investment from Sky Network's CEO, Song Tao. Today, this investment is valued at 340 billion yuan, a legendary return comparable to Masayoshi Son's investment in Alibaba. So where is the next miHoYo?

Can the myth of "Genshin Impact" be replicated in Web3? The GalaxyBlitz team believes that with the support of SLG + AI + Web3, a new low-cost path can be carved out to achieve the myth of "Genshin Impact."

First, let's observe the payment logic of "Genshin Impact." Characters are the core payment point of "Genshin Impact." It is a game that is extremely centered around characters; all content in the game, including but not limited to the storyline, the vast world, and animations, ultimately serves the characters. Each character has exclusive storylines and narratives, and even the separately created expansive worlds that enhance the gaming experience (like the Chasm) are shared among each character, as you ultimately need to use them for exploration.

So why do players pay for characters? Because players develop feelings for the characters. Teyvat is indeed a beautiful stage, and the vast world created by miHoYo, accompanied by resonant music, diverse characters, and stories, deeply captivates people. Walking on this land feels like stepping into a true otherworld—beyond novels and animations, it allows one to genuinely experience a pastoral poem from a foreign land. Beautiful things are inherently full of temptation, and we flock to them.

On this picturesque continent, whether players are exploring, battling, conquering the Abyss, or nurturing characters, it makes you feel as if you are truly in the wilderness or raising offspring. The brain produces a certain pleasurable feeling, stemming from the deep emotional bond cultivated between players and the two-dimensional characters through each adventure, captivated by the characters' experiences and personalities.

For example, Hu Tao is a classic "contradictory" character—both lively and profound, understanding the doctrines of life and death while living freely and carefreely. On the surface, she seems carefree and indifferent, but deep down, she holds a profound respect and seriousness towards death. This complex and contradictory multi-dimensional personality further enhances the charm of Hu Tao as a character.

On the surface, Hu Tao is playful and carefree, but understanding her experiences reveals a solemn and serious side.

miHoYo has crafted highly three-dimensional and complex personalities for each character. After experiencing adventures together, players gradually understand and uncover each character's story, captivated by their personalities and charm, leading to a desire to collect them, thus generating a payment impulse.

This is also why "Genshin Impact" must continuously undergo high-intensity updates, as only a large amount of high-quality content centered around characters can attract players to develop empathy and closeness with the characters during exploration, thereby drawing players into a payment cycle.

Web3 SLG Will Create a New "Genshin Impact" Through AI Technology

So, can this payment logic be replicated in SLG types? It is evident that the development of AI technology makes this logic possible. GalaxyBlitz will use AI technology in the next version, allowing each character in GalaxyBlitz to become an AI Being that can absorb the protagonist's personality. Unsupervised reinforcement learning technology can enable AI Beings to possess cognitive awareness similar to humans. All actions of the characters, such as swinging swords, dodging, defending, walking, sprinting, bending, shooting, etc., are trained by players themselves, leaving an imprint of the protagonist and making decisions similar to those of the players. Each character essentially becomes an AI relative that grows and experiences alongside the player, tagged with the player's label, easily generating empathy with the player.

How can these AI technologies be realized? GalaxyBlitz has developed a neural network training system suitable for SLG mobile games called GalaxyBlitz Mobile Net in collaboration with the well-known AI company Phantom AI. GalaxyBlitz Mobile Net is a scalable, data-driven approach that, in conjunction with low-power player mobile devices and GalaxyBlitz's deep reinforcement data center, learns reusable movement skills for AI Beings.

Our framework overview is shown in the figure below:

GalaxyBlitz Mobile Net Framework

The GalaxyBlitz Mobile Net framework consists of two phases: pre-training and target training.

During the pre-training phase: the low-level policy π(a|s,z) is defined as a mapping that uses action space data set s as the basis for action a to achieve target z, modeling various actions using a reward function r. After pre-training, it can be transferred to the target task using specific tasks.

We have specially prepared a demo of pre-training to multiple specific target tasks.

The video includes AI beings completing three goals defined by the reward function r: rolling through obstacles in mid-air, continuously shooting at the target, and giving themselves a thumbs-up after successfully hitting the target.

The low-level policy can model various general actions, and based on the player habit state set defined by Environment0, it can train general actions like swinging swords, walking, sprinting, bending, and shooting.

Low-level policy action combination - walking through water

Low-level policy action combination - rolling jump

Low-level policy action combination - running with a gun

Low-level policy action combination - kneeling and shooting

During the target training phase: the high-level policy w(a|s,z) is defined as a mapping that uses action combination z from the action space data set to achieve goal g, where z=array(a).

As shown in the figure below, through the high-level policy, AI Beings that have learned the low-level policy can be instructed to perform various specific tasks, which are composed of a set of actions from the low-level policy.

High-level policy action task combination - kneeling and shooting at the target

High-level policy action task combination - walking around the flag

In the high-level policy, initially, the simulated characters generated by random samples from Environment1 perform skill combinations. As player operations become more frequent, the discriminator will separate actions that align with player habits, and the actions of AI Beings will increasingly match player habits.

As shown in the figure below, our actions do not include "lying down" and "standing up," but AI Beings will autonomously learn "lying down" and "standing up." If players prefer the AI Being's timid personality, then the AI Being will "crouch" when encountering danger and "bravely stand up" at critical moments, thereby uncovering the player's inner brilliance.

The high-level policy gives AI Beings the personality of being "usually timid but courageous at critical moments."

The Rise of New SLG in Web 3.0

If the current version of GalaxyBlitz 1.0 ultimately achieves an annual revenue of 200 million dollars, reaching a relatively high level in the current Web2 SLG field, GalaxyBlitz aims to achieve: investment demand + entertainment demand + emotional consumption demand in the next AI version, with an expected annual revenue level similar to "Genshin Impact" at 5 billion dollars.

GalaxyBlitz team's envisioned Gamefi evolution route

The new type of SLG represented by GalaxyBlitz's AI technology will feature characters as NFTs with unique traits. In the next generation of SLG, each character will possess players' unique preferences and data; they will no longer be uniform numerical tools as in traditional games but will embody players' emotions, greatly enhancing playability and extensibility.

Thus, the personalized AI Beings will bring about the next narrative climax in Web3 gaming.