Summary of Key Points
"Remote Planet: The Builder" is an "AI-native game" developed by an independent team that breaks away from the traditional approach of using AI merely as a dialogue mechanism in games. Instead, it creates a dynamic space world driven entirely by AI: 600 NPCs with unique identities and decision-making logic, whose choices real-time affect the economic system and power structures. Every decision made by the player (such as deceiving an NPC) is remembered by the game world. While exploring the deep integration of AI into the core gameplay, the team faces challenges related to technology (managing large groups of NPCs), cost (controlling the costs of advanced models), and talent (finding individuals who understand both gaming and AI). They believe that AI-native games will become the norm in the industry and could even compete for top game awards.
I. AI-Native Games: Not Just a "Topping" on the Food, but the Main Ingredient
Many people think of AI games as simply adding the ability for NPCs to talk, but this game is different—its entire gameplay relies on AI.
- Traditional AI Games: These are like adding a bit of AI seasoning to a traditional dish; for example, using AI to write stories or optimize graphics. The game can still be played without AI (for instance, a martial arts game that uses AI to generate dialogue but has traditional combat mechanics).
- AI-Native Games: AI is the main ingredient; without it, the gameplay wouldn't function properly. In "Remote Planet," the decisions of 600 NPCs (such as whether to cooperate or raise prices) and the changes in the economic system (like a sudden shortage on a planet leading to price spikes) are all driven by AI. Without AI, this world that grows and remembers past actions would cease to exist.
The team's goal is to create a game similar to "Westworld," where NPCs have their own goals and take initiative (for example, thieves steal, and the police hunt them down), rather than waiting for the player to trigger events.
II. What Does AI Do in the Game? It Needs to Be Both "Lively" and "Stable"
The AI in this game isn't a single, large model; it's a combination of various elements:
1. Dynamic Economic System: Supply and demand, as well as prices, are not fixed values but are determined by the choices of 600 NPCs. For example, if you frequently buy minerals at a low price on Planet A and sell them on Planet B, the NPCs there might realize you're profiting from the difference and decide to raise prices next time.
2. NPC Memory and Decision-Making: Each NPC has a distinct identity (such as a merchant or pirate) and long-term goals (like becoming the leader of a guild). Their behavior adjusts based on the player's actions. If you deceive them once, they might refuse to do business with you; if you help them, they might offer you discounts.
3. Traditional Safeguards: AI isn't omnipotent; the developers have set rules (for example, NPCs can detect lies during negotiations) and limits (such as a maximum of five rounds of dialogue to prevent meaningless spamming). After all, the game needs a predictable and enjoyable experience; AI can't get out of control.
III. Challenges in Implementing AI-Native Games: Technology, Cost, and Balancing
To make AI-native games work, three major issues need to be addressed:
1. Technology: Managing 600 NPCs making decisions simultaneously can cause server crashes. The team uses "engineering design" techniques, such as pausing AI operations when the player is offline to avoid wasting computing power, and breaking down NPC decision-making into smaller, manageable tasks.
2. Cost: Using large models incurs costs (known as Token fees). If not controlled properly, selling the game for $30 would be unprofitable. The team's strategies include limiting AI output to only show content visible to the player and choosing open-source, cost-effective models (like Tongyi Qianwen) rather than the latest but more expensive ones (like DeepSeek), as stability is crucial.
3. Balance: Players want freedom (e.g., deceiving NPCs), but the game must remain controllable. The team allows some deception (e.g., saying "I'm your boss's friend" to negotiate lower prices) while setting limits (e.g., preventing prices from falling below cost levels). This ensures both fun and a stable gameplay.
IV. The Future: Talent Shortage, Players Becoming Creators, and Awards Within Reach?
1. What Kind of Talent is Needed?
- "AI planners" who understand both gaming and AI, capable of translating game design requirements into achievable AI functions (e.g., making NPC decisions more consistent with their character personalities).
- Designers who can clearly articulate the desired gameplay experiences (e.g., creating scenarios where NPCs retaliate after being deceived).
2. Will Players Become Creators?
In the future, players will be both consumers and creators. AI can help generate stories and modify NPC behavior (e.g., using mods to turn a NPC into a pirate). However, developers need to establish a clear framework (e.g., the game's worldview and basic rules) to prevent chaotic outcomes.
3. When Will We See Awards?
The team believes it won't be long before AI-driven NPCs with self-decision-making become the norm. Maybe within the next year or two, an AI-native game will win a top award like the TGA for "Best Game." By then, players will find it natural for NPCs to act on their own, just as they currently take it for granted that NPCs can talk.
In Conclusion
AI-native games are not about showing off technical prowess but about using AI to redefine the "sense of world" in games. The future of gaming will see worlds that come alive, where every player's choice can change the direction of the story. This could be the next major trend in the industry.