Text/Nine Lotus Treasure Lamp

Hyperparameter Technology Co., Ltd. is an AI company from Shenzhen. Founded in January 2019, the company was founded by Tencent AI Lab and led the team to develop Go AI "Jue Yi" and "Honor of Kings" AI Jue Wu. So far, many well-known game manufacturers around the world have established cooperation with them and used their AI services.

For the new track of "AI + game", hyperparameters also showed great enthusiasm. As early as 2020, they released the AI WeChat mini-game "It's Your Turn to Reveal the Secret", which was launched for 2 months and gained more than 300,000 users under the condition of zero purchase promotion. Last year, they also exposed a GAEA model that can build a whole ecosystem of NPCs, capable of creating a large number of AI NPCs, and has a vivid enough feedback mechanism.

In order to demonstrate this ability, they have released a short demo "Living Chang'an City". In this video, the NPCs quarrel, fish, fall in love, and when they find out that they have no money, they obediently go back to work, and this real reaction has attracted a lot of attention. Others said that perhaps the world depicted in the American drama "Westworld" will soon be realized.

And recently, I've heard that hyperparameters are starting to "toss" new games again. To find out, we reached out to Liu Han, the person in charge of the game, Distant Planets: The Builder.

Prior to joining Hyperparameter, Han worked at Tencent as the head of publishing for many years and participated in many projects. Today, he is working with a team of more than 10 people on the development of the game.

To be honest, I didn't expect much from the games they made. Although it is true that AI technology has developed rapidly in recent years, AI games are still just a gimmick. The main role that AI can see in games is still to participate in game production and produce art or copywriting resources. Can a complete game with AI as gameplay really be made and fun?

Regardless of the sale, their product really caught my eye: if nothing else, this game really gave me a viable direction for AI games in the future.

01

A truly dynamic society

At first glance, Distant Planet: The Builder doesn't seem to be any different from a regular management sim. The game builds a futuristic intergalactic era in a cartoonish style, where the player, as a merchant, is free to explore some barren planets and use the output of these planets to trade.

But in addition to production and construction, the game has a lot of RPG elements. For example, the player can visit different planets, communicate with the NPCs above, and convince them to do something beneficial to them; In addition to this, there are a lot of NPC competitors in the game who do the same job as the player, and have the potential to engage in business battles with the player.

The game also has a larger narrative, in which the universe is about to die, and players need to work with other NPCs to find a way to survive multiple crises and save the universe. Players can not only see the past events in this vast star field from the stories of each character, but also clearly feel the political, economic and other environmental changes in this star field during the game.

So, where is the AI element in the game?

The first is AI companion play. While the player is in the game, there will be an AI NPC companion who will chat with the player throughout the game, reminding the player of the information that is currently worth paying attention to. When players encounter something they can't do in the game, or some dilemmas, they can also turn to this partner for advice.

However, unlike many of the AI companions that exist today, their AI NPC companions have complete and coherent memories. Even if you open a new save and start playing again, the AI will retain your memories of playing the game before, and will even say something like "you didn't choose that last time".

But more importantly, there are hundreds of NPCs in the game, and behind them, there is an AI. With the blessing of AI, the whole game becomes very vivid:

For example, the relationship between these NPCs is quite complex, and it is a very complete small society. Liu Han showed me a three-dimensional model that showed the intricate relationships between different NPCs. Friendship or disbefriending any of the characters can trigger a chain reaction.

As another example, players can communicate with NPCs directly by typing. Liu Han showed me an example of how some NPCs in the game often encounter some life choices that make it difficult for them to make decisions, so you can persuade them to do something more beneficial to you.

Player lobbying can make a direct difference to the game environment. In the game, each NPC will have their own biography. Once you've influenced these characters, and then you go back and look at the biographies, and you'll see that the content changes accordingly. Especially when the player uses persuasion to influence the important decisions of some important people, the whole story may be completely different.

In my opinion, the most fun thing is the prophet in the game. The Prophet will make some predictions about how the world will change in the future of the game. For example, when the player asks about the future of a sector, the Prophet may tell the player that if certain conditions are met, the sector may mutiny in the future. If the player wants to prevent or facilitate this from happening, they can find a way to change the course of this particular condition.

Speaking of which, you can see that the way they use AI, especially large language models, is really comprehensive enough. You can see the AI involved in several of their main plays, and the way they participate is different each time.

After learning about the shape of the whole game, this game made me even more curious: why can the AI in their game be so closely integrated with the gameplay? How difficult is it to create such an AI society?

02

The content emerges, in fact, it is a twisting of the Rubik's Cube

As can be seen from the form of the game, the most attractive thing about this game is the AI dynamic society.

Liu Han said that they made this dynamic society to better represent the Chinese-style open world as he understood it: compared with the overseas open world that emphasizes interaction with the physical environment, the open world of the Chinese is more like a time and character relationship, such as games such as "Eight Desolations of Ghost Valley" and "Taiwu Picture Scroll".

Don't think that the so-called AI society is as simple as players watching NPCs enjoy their lives and simply accepting the content as it emerges.

In this game, there are a lot of human-created or AI-generated events that are only triggered based on the current environment. In order to get the results they want, players will try to reshape the relationships in this world and cause events to happen that will affect the course of the world.

"When we talk about AI, everybody talks about content emergence. But what is the purpose of the emergence? No one answered this question. In my opinion, the essence of emergence: a kind of Rubik's Cube. Why do we twist the Rubik's Cube? Because when you wring out a solid face, you get a sense of accomplishment."

And these events caused by the player's influence on the AI society will become a reward for the player, just like the solid color side in the Rubik's Cube.

In order to do well in this dynamic society, they did three things.

The first is to make sure that the game has the same world-running features as a traditional RPG, such as the main story and regular interactions, which is the foundation of everything. They chose to build it manually.

The second thing is to make NPCs explorable. Specifically, for example, the chain reaction caused by the player's attitude towards the character is the core mechanic.

At the same time, they also give the AI enough freedom to do what the player can do, and what the player can't do, the AI can't do. This kind of freedom allows AI to behave closer to that of a human being.

In addition, they also gave the AI the concept of subjective emotion. In the team's view, if AI is just acting as an emotionless robot that only reacts to objective data, then it doesn't really need AI to do it. So AI should have its own values for the whole world. Only in this way can the characters form different camps and create a variety of vivid character relationships.

These NPCs have subjective emotions and will also make the game more rich and realistic. Liu Han gave me an example, in many games, if a player needs to sell 10 items, they often need to find several buyers to clear the items.

In the past, you wanted to sell an item to an NPC, but he wouldn't accept it, and he wouldn't give any reason. This can sometimes lead to dissatisfaction among players who think that the makers are deliberately disgusting. However, the AI NPCs with emotions are able to justifiably say why they don't need such goods, and the player fills in a lot of information, which makes it fun.

"Why is it that many times playing games by yourself is not as fun as watching game streamers? Because the anchor will tell you the data and mechanics he encountered in the game through brain supplementation. In our game, the data and mechanics are already processed, and it lowers the threshold for players to make up the story."

But artificially created plots are still indispensable. Liu Han believes that AI can make divergences based on the current situation and emerge some interesting content, but if there is no artificially created some story nodes to converge on this divergence, the story will lose its theme and become boring.

The third thing is the AI companion function. The team believes that the current game can be simply divided into single-player and multiplayer, but there is a lack of a "1.5-player game" on the market.

"For some of the more 'I' players, they want to communicate with people in the game, but they are afraid of taking on the pressure of multiplayer versus gameplay."

Based on this logic, when designing AI companions, they tend to make it play a positive role, such as providing strategies and advice, or providing reassurance when players are frustrated. They also added the ability to speak to the AI companion to help players better immerse themselves in the world. In the future, players may also be able to freely choose their favorite NPCs to play with.

Liu Han said that at present, most of the applications of AI in the game industry are mainly AIGC, and even if there are AI NPCs, they only invest individual AI in mature games. There are no AI NPCs like them that put in hundreds of AI NPCs and let them form a dynamic operation on their own.

"We're more confident in saying that we're really doing something different, something that AI can provide."

Compared to a dynamic society, the simulation may seem inconspicuous, but it is actually a key point of their work.

Liu Han shared his previous work experience with me. "For example, PUBG, it's FPS + Battle Royale, and Battle Royale is its mechanic. PUBG is at least a shooter if it is detached from this mechanic, while Battle Royale cannot exist on its own without gameplay. For us, the AI dynamic world is just a mechanic, and we need to have the core gameplay to base it on."

Liu Han said that the technical difficulty of simulated operation is not high, and the user acceptance is relatively good. As a small team of only a dozen people, it is the most cost-effective choice to take simulated operation as the core gameplay.

But that doesn't mean they don't care about business simulations. Instead of traditional simulations, they decided to design the game as a limited-time level gameplay, where players come to the planet and need to mine resources locally, build some production lines, and finally create a series of content required by the mission in a limited time. And in the level, they also added a part of the meat pigeon element.

"This is more in line with the current gaming habits of users, and it can also help us avoid comparisons with some traditional simulation operations. After all, we really haven't been in this category for that long."

03

AI shouldn't do cheap things

Nowadays, there are many products that want to create a novel experience of AI games, but they are basically small games made by some entrepreneurial teams, and the gameplay is also a bit difficult to explain:

For example, "Oasis", which has been a topic of conversation before, is actually an AI-generated "Minecraft" that automatically generates what the player sees every time the player turns the camera.

It's a novelty as an experience, but it's hard to say if it's a playable game.

Today's popular role-playing software for large language models is also a breath of relief. Although they have been able to attract many users to try them out, in the final analysis, this model is more like a "chat simulator", lacking sufficient visual and gameplay experience.

Liu Han said that if players are allowed to remotely control each specific action of the AI, such as requiring the player to type and enter instructions such as movement, to achieve these very simple operations, it will only make the experience more complicated and uninteresting.

"Instead of using something as expensive as AI to do these very cheap things, game teams should make AI work for the actions that players have already done."

In Liu Han's view, the key to maximizing the change of large language models in the form of games is to regard them as the "logic engine" of games.

Now, the upgrade of the graphics engine and hardware capabilities has given the game team the opportunity to make the graphics more realistic and the maps bigger. Players can experience content that can only be experienced in many game maps in the past in a single map.

But behind this realistic picture, paradoxically, the characters in it have almost no memory and logic to speak of, they will only repeat those set sentences and do those set actions.

If the large language model can become the logic engine of the game, it will give more logic to the behavior of NPCs, and also make the open world, which is getting bigger and more complex, really work.

It's not just about getting NPCs to eat and drink, it's about explaining more of the content in their own way.

For example, when the planner comes up with some wonderful story lines, then the large language model will connect them in its own way, and the ending of the story will no longer be determined by one or two key choices, but will be affected by every action of the player.

When the story comes to an end, the logic is smoother, and there is no longer a situation where the player will be asked "is it worth it" no matter what they do.

In addition, large language models, as logic engines, can also play more ways to play. Liu Han introduced that in their game, there is also a new system mechanism. It allows the player to go back to a point in the past in the game, change their choices at the time, and return to the current timeline.

When the player completes a time travel, the AI will re-reason about the player's impact on the world. Sometimes it may be that nothing happens, but there is also the possibility that the whole world has changed dramatically.

This kind of gameplay experience can only happen when a large language model intervenes. In Liu Han's view, this is also the novel experience that AI games should really try.

Currently, Distant Planet: The Builder is halfway through development, and they plan to officially release the game next year.

As Liu Han said, this kind of "living NPC" is actually the ultimate goal that all RPGs want to achieve, and this kind of logic engine concept can also be used in more RPG game types.

If the hyperparameter team can run through this mode, then this new form of AI games may become popular in the future, and even become one of the mainstream directions of RPG games.

This article is from Xinzhi self-media and does not represent the views and positions of Business Xinzhi.If there is any suspicion of infringement, please contact the administrator of the Business News Platform.Contact: system@shangyexinzhi.com