How AI Could Bring Halo and Other Classic Games Back to Life

以下是该播客文字记录的摘要： ### **《侠盗猎车手 6》与漫长的游戏开发周期** * 《侠盗猎车手 6》将在下个月发布，其制作周期始于 2013 年或 2014 年左右（开发周期长达 12 到 13 年）。 * 尽管《侠盗猎车手 6》预计会是一款出色的游戏，但与其它软件标准（如人力资源软件、客户关系管理或企业资源规划软件）相比，超过 10 年的开发周期被认为极其疯狂。 * 如此漫长的开发周期对创新产生了令人窒息的影响。面对 10 年的时间线和数亿美元的成本，3A 游戏制作人变得非常保守，他们更倾向于批准续作（如第 20 部《使命召唤》或《侠盗猎车手 6》）以及成熟的类型（如第一人称和第三人称射击游戏），而不是新的知识产权或实验性游戏类型。 ### **游戏开发中的人工智能技术突破** * 大约在一个月前，随着 Astra 的出现，技术迎来了重大突破，因为它能直接操作流行的游戏开发软件，包括直接在虚幻引擎和 Unity 中工作，以及在云端使用 Blender。 * 过去，模型只能就如何在 Blender 或虚幻引擎中操作给出文字指令；现在，Astra 可以直接操作该软件。 * 业余爱好者或从未制作过游戏的人首次能够使用 Astra 完成以下操作： 1. 起草游戏设计文档。 2. 生成概念美术。 3. 将美术素材导入 Blender 以构建 3D 模型、制作动画并进行绑定。 4. 将模型导入虚幻引擎。 5. 添加音乐和声音文件（这些也可以生成）。 6. 将项目导出为可玩的游戏（使用 Astra 一键完成）。 * Astra 还可以通过截取游戏屏幕并从其输出中进行学习来修复错误或视觉问题，从而实现自我质量保证。这就解释了为什么最初在 X 上构建的原型会出奇地好。 * **人工智能对行业的影响：** * 大幅缩短了从构思到原型的开发时间并降低了成本（将 5 到 10 年的开发周期缩短至 2 到 5 年，并将数亿美元的成本降至数千万美元）。 * 蒂姆·斯威尼（Epic）指出，人工智能使得开发小众游戏和过去在经济上不可行的深奥游戏类型成为可能。 * 允许游戏开发者承担更多创意风险、进行更多实验，并提升游戏的丰富度、多样性和雄心（可能让下一个历时 12 年开发的游戏能够构建一个宇宙，而不仅仅是一座城市）。 * 让游戏创作民主化：你不再需要成为专业人士，也不再需要一个涵盖编程、美术、游戏设计以及音乐/音效的全功能多学科团队，因为人工智能可以填补技能空白。 ### **模组、混音与反编译** * 最大、最好游戏的悠久历史在很大程度上依赖于模组（为喜爱的游戏构建附加扩展，从而改变游戏玩法并成为独立的游戏类型）： * 《反恐精英》是《半条命》的一个模组。 * 《英雄联盟》的灵感来自《远古保卫战》（Dota），后者是《魔兽争霸 III》的一个模组。 * 《堡垒之夜》和《绝地求生》的灵感来自《武装突袭》的《绝地求生》大逃杀模组。 *（原文如此） * 人工智能通过为创作者提供一个框架或现成的知识产权作为起点，使游戏的模组制作和混音变得非常简单（例如，将《对马岛之魂》重制为带有体素基建造系统的《我的世界》模组，或者加入《动物森友会》的关系系统）。 * **反编译与重新编译趋势：** * 社区项目正在将旧的、深受喜爱的游戏（如 2001 年的 Xbox 游戏《光环》）进行反编译和重新编译，以适应现代平台（例如将《光环》制作成可在网页图形库上玩的游戏）。 * Xbox 实验室的公告以及阿莎对人工智能和新开发前沿的支持，是由将旧游戏移植到现代生态系统（如 iOS 或网络）并掌握停产游戏的社区项目所引发的，这让年轻观众能够游玩诸如《旧共和国武士》之类的主题游戏。 * 诸如 X 上的创作者“Blendy”（B-L-E-N-D-I）正将两个现有游戏的资产提取出来，并将它们组合成第三个正常运行的游戏。 ### **新兴趋势与新游戏类型** * **3.js 操作：** Anthropic 的模型（如 Opus 4.7 或 4.8）在 3.js（浏览器内 3D）方面表现出乎意料地出色，它们以一丝不苟的方式逐个元素进行布局，为结合 Astra 工具使用的基于代码的游戏生成铺平了道路。 * **将创作视为游戏 / 禅意园艺：** 创作游戏、将资产赋予生命以及看着角色做出意想不到的事情的过程，具有一种宁静的禅意园艺般的特质。对许多人来说，创作即是消费；绝大多数人工智能游戏将仅由其创作者自己游玩。 * **模拟器：** 职业模拟器（例如东京地铁列车长模拟器）代表了一种常青的小众游戏类型，人们愿意花很多钱来玩。其他例子包括《无题大鹅模拟》（大鹅模拟）和早期飞行模拟器。 * **即时战略游戏的复兴：** 历史上面临低投资回报率和高制作门槛的即时战略等游戏类型，正在经历全面的复兴。 * **恋爱模拟游戏：** 乙女恋爱模拟游戏类型（例如由中国开发者制作的、创造了数亿美元收入的《恋与深空》）正迎来被人工智能颠覆的时机，其中的角色可以充当具有独特个性并能实时响应的持久智能体。 ### **生态系统、平台与变现** * **Xbox / 阿莎与 Epic / 蒂姆·斯威尼：** 两位领导人都对人工智能开发的游戏以及硬件与软件的相互作用表示了支持。微软/Xbox 见证了社区项目的发展，并希望支持新的开发前沿，尽管商业模式还需要迎头赶上。 * **变现与知识产权挑战：** 从历史上看，原始知识产权的所有者并没有从在其作品之上构建的模组中受益太多（例如 Dota/《英雄联盟》或《半条命》/《反恐精英》的创作者）。公司需要摸索出收入分成、版税或可持续的模式（例如《我的世界》的模组生态系统，当扩展共享核心基础而不是进行分支时，它会使知识产权所有者受益）。 * **分发平台：** 像 Steam 和应用商店这样的传统渠道处理绝大多数游戏，但原生人工智能分发平台存在巨大的机遇。Spawn（由雅各布和他的团队创建）被着重强调为一个雄心勃勃的例子——这是一个共享的社区/游戏生成平台，其特色是一个名为“Savvy”的具身智能体，它负责接收漏洞报告、即时修改游戏、创建通往其他游戏的传送门，并默认让每款游戏都支持多人联机。 * **共享世界（《我的世界》与《罗布乐思》）：** * 《我的世界》采用集中式模型，每个人都从相同的丰富核心游戏开始，并像同人小说一样对其进行模组化/分支，共享共同的文化和知识产权熟悉度。 * 《罗布乐思》采用去中心化模型，玩家通过完全不同的独立体验（如《越狱》或《布鲁克海文》）进入，为创作者提供了不同的引导路径。这两种方法都很有效。 ### **游戏领域人工智能的局限性与工作室视角** * **人类的品味与品味判断：** 游戏最后 20% 的工作占了总工作量的 80%——这就是魔法发生的地方（例如武器音效设计、后坐力、世界感、情感共鸣）。人工智能无法取代人类的品味、创意判断以及使游戏真正有趣且引人入胜所需的直觉。 * **工作室情绪：** * **令人兴奋：** 降低成本、减少财务风险、缩短原型制作循环、改善过度劳累的工作室开发人员的生活质量，以及创造更多的知识产权。 * **令人担忧：** 努力构建“常青游戏”或网络/社交产品（如《堡垒之夜》或《我的世界》），而不是纯粹的娱乐产品。 * **未来 / “智力游戏”愿景：** 大型多人在线游戏和未来游戏的终极目标是一个自适应的世界模型（类似于《安德的游戏》中的“智力游戏”），它能够动态生成任务，并根据玩家在任何给定时刻想要做的事情实时进行自我调整。

Here is a summary of the podcast transcript: ### **GTA 6 and Long Game Development Cycles** * *Grand Theft Auto 6* is releasing next month after a production cycle starting around 2013 or 2014 (a 12-to-13-year development cycle). * While GTA 6 is expected to be an amazing game, a 10+ year development cycle is considered insane compared to other software standards (like HR software, CRM, or ERP). * These long dev cycles have a suffocating impact on innovation. AAA game producers facing 10-year timelines and hundreds of millions in costs become very conservative, greenlighting sequels (like the 20th Call of Duty installment or *Grand Theft Auto 6*) and established genres (like first-person and third-person shooters) over new IPs or experimental genres. ### **The AI Technological Unlock in Game Development** * A major technology unlock occurred about a month ago with Astra because it directly operates popular game development software, including working directly within Unreal and Unity and using Blender in the cloud. * Previously, models only gave text instructions on what to do in Blender or Unreal; now, Astra can operate the software directly. * For the first time, a hobbyist or someone who has never built games can use Astra to: 1. Draft a game design document. 2. Generate concept art. 3. Move art into Blender to build 3D models, animate, and rig them. 4. Move models over to Unreal. 5. Add music and sound files (which can also be generated). 6. Export the project as a playable game (one-shot using Astra). * Astra can also QA itself by taking screencaps of the game and learning from its output to fix bugs or visual issues. This explains why initial prototypes built on X are surprisingly good. * **Impacts of AI on the industry:** * Dramatically shortens the time and lowers the cost from idea to prototype (shortening 5-to-10-year dev cycles down to 2-to-5 years, and hundreds of millions in cost down to tens of millions). * Tim Sweeney (Epic) notes that AI makes it possible to build niche games and esoteric genres that previously were not economically viable. * Allows game developers to take more creative risks, make more experiments, and increase the richness, diversity, and ambition of games (potentially allowing the next 12-year game to build a universe rather than just a city). * Democratizes game creation: you no longer need to be a professional or have a full multi-disciplinary team covering coding, art, game design, and music/sound, as AI can plug skill gaps. ### **Modding, Remixing, and Decompilation** * The history of the biggest and best games heavily relies on modding (building add-on extensions to beloved games that change gameplay and become their own genres): * *Counter-Strike* was a *Half-Life* mod. * *League of Legends* was inspired by *Defense of the Ancients* (Dota), which was a *Warcraft III* mod. * *Fortnite* and *PUBG* were inspired by the *PlayerUnknown's Battle Royale* mod of *ARMA*. * AI makes modding and remixing games very easy by giving creators a framework or existing IP to start from (e.g., recreating *Ghost of Tsushima* as a Minecraft mod with voxel-based crafting or *Animal Crossing* relationship systems). * **Decompilation and Recompilation Trends:** * Community projects are taking old, beloved games (like the 2001 *Halo* Xbox game) and decompiling/recompiling them for modern platforms (such as making *Halo* a playable WebGPU game). * The Xbox Labs announcement and Asha's support for AI/new development frontiers were sparked by community projects porting old games to modern ecosystems (like iOS or the web) and mastering defunct games so younger audiences can play titles like *Knights of the Old Republic*. * Creators like "Blendy" on X (B-L-E-N-D-I) are taking assets from two existing games and combining them into a third, functioning game. ### **Emerging Trends & New Genres** * **3.js Manipulation:** Anthropic models (like Opus 4.7 or 4.8) unexpectedly excelled at 3.js (in-browser 3D) by laying things out element-by-element in a meticulous way, paving the path for code-based game generation alongside Astra's tool use. * **Creation as Play / Zen Gardening:** The act of creating a game, bringing assets to life, and watching characters do unexpected things has a tranquil, zen gardening aspect. For many, creation is consumption; the vast majority of AI games will be played solely by their creator. * **Simulators:** Job simulators (such as a Tokyo subway train conductor simulator) represent an evergreen, niche genre that people will pay a lot of money to play. Other examples include *Untitled Goose Game* (goose simulation) and early flight simulators. * **RTS Renaissance:** Genres like RTS (Real-Time Strategy), which historically suffered from low ROI and high production barriers, are seeing a complete renaissance. * **Dating Sims:** The Otome dating sim genre (such as *Love in Deep Space* by a Chinese developer, which makes hundreds of millions in revenue) is ripe for AI disruption, where characters can act as persistent agents with unique personalities responding in real-time. ### **Ecosystems, Platforms, and Monetization** * **Xbox / Asha and Epic / Tim Sweeney:** Both leaders have expressed support for AI-developed games and hardware/software interplay. Microsoft/Xbox is seeing community projects and want to support new development frontiers, though business models need to catch up. * **Monetization and IP Challenges:** Historically, original IP owners have not benefited much from mods built on top of their work (e.g., the creators of Dota/League of Legends or Half-Life/Counter-Strike). Companies need to figure out revenue shares, royalties, or sustainable models (such as Minecraft's modding ecosystem, which benefits IP owners when extensions share the core foundation rather than forking). * **Distribution Platforms:** Traditional channels like Steam and the App Store handle most games, but there is a major opportunity for native AI distribution platforms. Spawn (created by Jacob and team) is highlighted as an ambitious example—a shared community/game generation platform featuring an embodied agent named "Savvy" who takes bug reports, modifies games on the fly, creates portals to other games, and makes every game multiplayer by default. * **Shared Worlds (Minecraft vs. Roblox):** * Minecraft utilizes a centralized model where everyone starts with the same rich core game and mods/forks it like fan fiction, sharing a common culture and IP familiarity. * Roblox uses a decentralized model where players enter via totally different standalone experiences (like *Jailbreak* or *Brookhaven*), offering a different onboarding path for creators. Both approaches work. ### **The Limits of AI in Gaming & Studio Perspectives** * **Human Taste and Taste Judgment:** The last 20% of a game is 80% of the work—this is where magic happens (e.g., weapon sound design, recoil, world feel, emotional resonance). AI cannot replace human taste, creative judgment, and the intuition required to make a game genuinely fun and engaging. * **Studio Sentiment:** * **Exciting:** Getting costs down, reducing financial risk, shortening the prototyping loop, improving the quality of life for overworked studio developers, and creating more IP. * **Worrisome:** The struggle to build a "forever game" or a network/social product (like Fortnite or Minecraft) rather than a pure entertainment product. * **The Future / "Mind Game" Vision:** The holy grail for MMOs and future games is an adaptive world model (akin to the "mind game" in *Ender's Game*) that dynamically generates quests and adapts itself in real-time to whatever the player wants to do at any given moment.