Thoughts on Making Games in the Age of AI: Generative AI at Ludum Dare 浅谈AI时代的游戏创作：Ludum Dare 里的生成式 AI

1 Preface

This October I took part for the first time in the Ludum Dare game jam (ldjam from here on). My teammates and I started making things with almost no plan at all. Most of the design came from sudden flashes of inspiration or something close to daydreaming, and luckily we managed to submit a reasonably complete game at the very last minute. But what I want to talk about here is what came after the jam. Once the 18 days peer-rating period began (a game needs more than 20 ratings to get a final ranking), I noticed a huge amount of discussion about gen AI in game-making: in the comment sections of our own game and other people’s, in posts on the ldjam forum, and on ldjam Discord. This discussion has taken center stage across several events and touches almost every game (Figure 1). By now, almost everyone who joins a game jam has to face two questions: should I use AI to make my game? And how should I rate games made with AI?

So in this post I scraped data from LD50 to LD60. Through trend, stance and sentiment analysis of how generative AI has been used at Ludum Dare over the past 5 years, how people rated and commented on games that did or did not use AI, and what exactly people argued about on the forum, I try to get a grip on how attitudes towards generative AI in game-making have changed, from the big picture down to specific cases, and on what we might learn from it.

I chose Ludum Dare first of all because I have been binge-reading its posts lately. It also helps that its data are relatively open and transparent, and come from several sources: game pages, comment pages, ratings and posts. As a long-running event that is well known in the industry, it has many participants and a lot of discussion and data. And because it lacks moderators and community guidelines that keep up with the times, what people say there reflects fairly well what is actually going on right now.

2 What We Talk About When We Talk About AI

AI is a complicated concept in itself. From early symbolic reasoning and expert systems to machine learning and deep learning, from the pathfinding algorithms and behaviour trees that drive enemies in games to today’s LLMs and diffusion models that write code, generate images and compose music, all of these broadly count as AI. In game-making specifically, and based on what actually appears at Ludum Dare, mentions of AI broadly fall into these categories: using AI to make the game, AI running inside the game, AI-written comments or posts, AI as the theme of the game, AI as an opponent or NPC, and other or unrelated (comments that the keywords caught but that turned out to have nothing to do with genAI).

It is easy to see from the figure that using AI to make the game and AI running inside the game have always been the main things people talk about. Besides the discussion of AI-written comments and posts, one clear trend is that the share of “using AI to make the game” rose from 38% to 83%, while “AI running inside the game” fell to 5%. Different applications and extents of use can draw completely different reactions even when the same LLM sits behind them, so specific cases need to be looked at one by one. Ludum Dare has one more peculiarity: it has more than one submission categories: Jam, Compo, Extra and Unfinished. The LD51 welcome post said that AI should not be used in the Compo and that such games should be submitted to the Jam or Extra instead; the 2024 FAQ added that AI assistants and co-pilot tools “can be used without restriction”. So we also need to discuss the Jam and the Compo separately.

It is also worth noting that “using AI to make the game” alone can be broken down further by what the AI was used for (art, music, code and so on). Figure 3 shows that, apart from fairly general mentions of AI, the related comments most often mention art, followed by writing and then music and sound. Notably, programming has clearly been on the rise. This is very likely connected to the boom in AI coding around 2024¹, and also to the LD community’s lack of rules about programming and the arguments that has caused.

3 How do I rate AI games?

The posts that got me thinking the most were I don’t know how to rate “AI-assisted” games and How do you rate Claude games?. Both sparked a lot of discussion, because the topic reflects more than just widespread confusion over the mechanics of rating. Each reply subtly introduces a complex web of underlying issues and stances.”

But before answering that question, let’s first take a broad look at how Ludum Dare participants feel about AI use, and at how people rate and comment overall.

To start with, we can set aside what form the AI takes and what exactly it refers to, and simply classify each comment’s stance on the AI use at face value (seven levels, from Praise to Direct criticism), to sketch the overall trend of the community’s discussion.

So what exactly do the AI-related comments say? Looking across the ldjam comment data, we can roughly sort them into the ten kinds shown in the figure below. Obviously, each kind may lean its own way. “Praises not using AI”, for example, usually signals a dislike of AI use, and such comments are more common in the Compo than in the Jam. The attitude behind “Points out AI traces” can be more complicated, and it may change over time. Back in LD52, asking or pointing out whether some part of a game was made with AI might have been plain curiosity or technical interest, but by LD60 it often carried an undertone of implicit criticism. In particular, commenters who choose not to be anonymous when giving feedback mostly use polite, courteous wording, partly out of fear of retaliatory ratings, resulting in deliberately vague or neutral phrasing.

Figure 5 shows that “using AI to make the game”, “AI running inside the game” and “AI-written comments or posts” received completely different kinds of comments. “Using AI to make the game” is the most common and the most varied topic, but comments that oppose AI-generated work as a whole, such as Says AI was used and not admitted, Praises not using AI and Thanks for disclosing or opting out, are concentrated in the last few events, especially LD60. For “AI running inside the game” there is almost no criticism; most comments stay at judging how well it works or talking about the technology, and only a few ask whether the rules allow it. For “AI-written comments or posts” there is very little data overall, and it mostly shows up only in LD60, so it is a fairly new phenomenon. Concretely, some users exploit the system’s rules by using AI to comment on large numbers of games quickly and in bulk, which boosts their own game’s visibility and baits other developers into rating theirs. People generally resent this, but the current rules have no way of forbidding it.

So, for the different uses of AI in game-making (Figure 3), how is each one judged? We looked at the share of criticism among these comments. Because the data are limited, categories such as Writing show no statistically meaningful preference. Still, an overall trend is easy to see: for almost every category with data, criticism of AI use first rises and then stays at a fairly high level.

Before discussing how AI use relates to ratings, we also need some way of deciding whether a game used AI at all. The simplest way is the creator’s own declaration, but it’s not necessarily honest, and not everyone makes one. It’s also hard to setup a criteria for people to judge manually; for code in particular, since not every codebase is open source, it is sometimes nearly impossible to tell. It is also worth mentioning that the “AI declaration” seems to play a subtle role: it lets authors explain why they used AI. For example, someone who is only good at art and writing might use AI to help with the programming, because otherwise they could not finish a complete game within three days and say what they wanted to say. Even if we look only at the tags on itch.io, different players may react differently to different declarations; and tags related to art, audio or text may draw more discussion.

Although ldjam’s mechanics encourage people to comment a lot, in exchange for more “karma” that makes their own game more visible, the comments, which are usually not anonymous, and the ratings, which always are, may well differ quite a lot in both number and attitude. In particular, sometimes only people with clear likes or dislikes about AI mention it in their comments, while people with vaguer or neutral views may not bring it up or join the discussion at all. For ratings, we analysed the LD59 rating data, which covers 1,142 ranked games, grouped by the itch.io declarations compiled by @rogual. Figure 8 shows that, unlike what the comments suggest, games that declared AI use did not score noticeably lower than games that declared no AI use or made no declaration. On the contrary, games that used AI only to help with coding are no different (both groups average 3.55), and their median score is even slightly higher. This group also includes the Compo winner, Boba’s Love Story. Although many players questioned its use of AI in the Compo, the game still came out on top in the ratings and took first place. Given that declarations may be untrue, if we look at it simply from this angle, a creator who wants a high score could in theory use AI to help with code, and then declare falsely or not declare at all, even though this is likely to be unfair to the other entrants.

4 What should we do?

From the data above, we now have a rough overview of how AI is used in game development, what the comments and ratings look like, and how the community feels overall. Broadly speaking, the share of games made with AI’s help has grown with every event, and it has spread from art, music and text at first to code and overall design. AI as an NPC or as a subject of discussion appeared more in the early events; as people got used to AI, it stopped being such a frequent topic. People have started using AI to spam comments in bulk. Spam comments have always existed, previously as copy-pasted, generic text, but it is only in LD60 that they became mainly AI-driven. Over these ten events the discussion about AI never stopped, and its specific topics kept changing. In terms of attitude, the comments clearly show a growing dislike of generative AI in game development, but the ratings do not seem to show this tendency very clearly.

Of course, even with all this data, I still cannot pass judgement on the views of everyone in the I don’t know how to rate “AI-assisted” games thread. Nor can I say whose side I am on, because everyone there seems to be discussing in good faith and to be genuinely puzzled. And although everyone appears to be talking about the same thing, “how to rate AI-assisted games”, the thread actually covers several different sub-questions. What is AI’s real role in game development: a tool, a means of theft, or a colleague? Where does the core of game design and development lie? What are ratings really for? Is this competition fair? Where do the rules come from? Everyone is probably partly right from where they stand.

For now, one of the places where the controversy gathers most is AI-assisted programming. Unlike AI-generated art and text, AI-generated code, although it has its own telltale features, usually does not show up directly in what you see on screen, so it is almost impossible to spot without reading the source, and ldjam has no rating category for code. It is also a fairly new and little-discussed problem, because until recently AI coding was not good enough to build a complete game on its own, so naturally there were no rules for it either.

As a computer science student, I can appreciate the beauty of programming, but I also know that most games cannot show it. A game that runs smoothly and is well optimised owes that to its code, of course, but players do not think “this game must have some really optimised code behind it!” while playing. And the beauty of code comes in many layers. A small function can be clever and beautiful, with low time and space complexity, or the overall system design can be very good, and it is hard to say what exactly counts as good. For me personally, programming is an intellectual challenge, and I enjoy it both during the process and at the end, but it is hard to imagine the people who use my programs feeling that same joy. As someone who came to computing from a degree in Chinese language and literature, I also understand the frustration of not having enough technical skill to express your own ideas. AI clearly helps people like that. Even for people who already have plenty of programming experience, AI-assisted coding can be very useful, because it lets you put your energy into higher-level orchestration and game design. As one forum comment puts it: “Using AI doesn’t mean I’ve suddenly forgotten how to program. My knowledge and experience are still there. I’m simply making part of the development process faster and easier, while remaining in control of what I’m creating.”²

The question is, what is the essence of game design? Willops says: “How much speed to give this enemy? How much jump height to give the player? How long should the jetpack power last? These questions are the essence of game design to me.”³ That is true, but the essence of game design may well be different for different people and different genres. And if someone learns how to set an enemy’s speed from YouTube or somewhere like that, how is that different from getting a “suggested enemy speed” from AI? I think what matters is the role the human plays in the whole creative process. For the feel of each specific control, you open Godot or Unity and test, check and tweak it; for the overall design, you write out all the conditions and triggers step by step. But this line is very blurry. If you don’t establish clear creative boundaries, it is very easy to let AI decide everything and end up with a work that does not really belong to you at all, yet carries your name and may even win you a prize.

Where do the rules come from when a community has no moderators and no shared guidelines? The ldjam user @Jezzamon started a poll and shows the results so far at ld-votes. Interestingly, the results so far are close to a tie. Given that ldjam is coming to an end, I rather doubt that new rules will be announced, or that people would follow them if they were. It has always been this way: When the only deterrent is the honor system, rule-breakers easily gain the upper hand.

But in the end, ldjam is not a contest about competing and scores, and it is hard to say that a truly fair competition exists anywhere in the world. Some people will surely use their ranking as a selling point, but what most people are really after is a bit of motivation and some constraints, to finish a game as well as they can within limited time and a given theme. It is a test of your own ability, and a chance to express yourself. Rating each other’s games is, at heart, about getting your game seen and getting as much feedback as possible, and also a chance to play other people’s games, broaden your horizons and find new inspiration. Whether to use AI and how to use it, whether you let AI push you through the whole design until it erodes your own ideas, or let AI do the groundwork so that you can spend more time on the core design, is, for now, essentially a personal choice.

Both ldjam and game development are only a tiny microcosm of how AI is used today and of its effects. What is truly worrying is the making of new rules and theories once the forces and relations of production change, and that blurry edge of creativity.

1 前言

今年十月，我第一次参与了久仰大名的Ludum Dare Game Jam（下文简称ldjam）。和几个队员几乎是毫无规划地就开始了制作，多数设计全靠灵光乍现或者梦游天姥，幸好在最后一刻提交上去了一份还算完整的作品。不过这次我想重点讨论的是游戏创作结束后的部分：作者们进入为期18天的互相评分环节（只有收到超过20个评分才能进入最终的排名）后，我注意到在我们自己和其他游戏的评论区、ldjam论坛的posts、Discord论坛，产生了大量关于游戏创作中的生成式AI的讨论，而且这类讨论是核心的、持续了好几届的、横跨几乎所有游戏的（如图1）。如今，参与gamejam的所有创作者几乎都必须要面对这个问题：我要使用AI创作游戏吗？我要如何给使用AI创作的游戏打分？

因此在这篇文章中，我抓取了LD50到LD60的数据，试图通过对近五年间Ludum Dare中生成式 AI的应用、人们对使用和没有使用AI的游戏的打分与评论、人们在论坛中具体的讨论内容进行趋势、态度与情感分析，来从宏观到具体地把握人们对于游戏创作中的生成式 AI的态度变化，以及可能的启发。

选择Ludum Dare的数据当然首先因为我最近高强度浏览了大量帖文，其次也因为其数据相对公开透明，并且包含游戏页面、评论页面、评分和post等多个数据源，作为业界闻名的老牌活动参加者、讨论声量和数据量相对比较大，同时因为缺乏 mod 和与时俱进的公约，其内容也很能够反映当前的真实情况。

2 当我们在谈论AI的时候，我们在谈论什么

AI本身就是一个复杂的概念，从最早的符号推理与专家系统到机器学习和深度学习，从游戏里驱动敌人的寻路算法与行为树到如今能写代码、画图、作曲的大语言模型与扩散模型，宏观意义上都属于AI的范畴。具体到游戏制作领域，并且考虑到Ludum Dare中出现的实际种种情况，大致可以归类为：使用AI制作游戏、游戏内运行的AI、AI撰写的评论或帖文、作为游戏主题的AI、作为敌人或NPC的AI、其他/无关（这些评论被关键词捕获，但经过筛查发现实际上是无关的）.

从图中不难发现，使用AI制作游戏和游戏运行时内部的AI一直都是人们主要讨论的内容。除了对于AI撰写的评论或帖文的讨论，一个明显的趋势是“使用AI制作游戏”占比从 38% 升到 83%，“游戏里运行的 AI”降到 5%。对于不同的情况、不同位置和程度的使用、即使背后都是同样的LLM，人们也可能产生截然不同的态度，因此需要对不同的具体案例进行分析。对于Ludum Dare还有一个比较独特的情况：其有多个提交类别：jam，compo，extra，unfinished。其中LD51的欢迎帖指出AI不应当用于compo，应该提交到jam或者extra类别中；2024 年的 FAQ 补充说 AI 助手和 co-pilot 类工具”can be used without restriction”。因此我们也将jam和compo的情况分开来讨论。

除此之外，值得注意的是，仅仅是“使用AI制作游戏”一条就可以根据AI不同用途（美术、音乐、代码……）继续细分。从图3中我们可以看出，除了比较general的AI提及以外，人们往往在相关评论中最常提及游戏美术，其次是writing，Music and sound。值得注意的是，游戏编程经历了明显的增长态势，这和2024年左右AI编程的大幅兴起很可能存在联系¹，也和ludumdare社区缺乏对游戏编程相关的条例与引起的争执有关。

3 How do I rate AI games?

我个人认为最给我启发的帖文是I don’t know how to rate “AI-assisted” games 和 How do you rate Claude games?. 这两篇帖文都激起了不小的讨论，因为这个话题不仅反映了一种普遍的、对于具体的“评分”这个行为的困惑，每个参与者的评论也都潜在地引出不少与之相关而交错的议题与立场。

不过在回答这个问题之前，我们可以先宏观地看LudumDare参与者对AI使用的态度如何，看看人们整体上是如何打分和评论的。

首先，可以不考虑AI是以什么形式被提及的、具体是指什么，只按字面归类每条评论对 AI 使用的评价态度（从Praise到Direct criticism分为七个层级），来描摹整体的社区讨论趋势。

那么，涉及到AI的评论具体都是如何评论的呢？纵观ldjam的评论数据，我们大致可以为其归为如下图所示的10类，很显然，不同的类别都可能有其各自的态度倾向，例如“Praises not using AI”往往代表对AI使用反感的态度，而这类评论在compo类中多于jam类，而“Points out AI traces”背后蕴藏的态度可能更复杂，并且在不同时期可能不一样。例如，在LD52的时候询问或指出游戏中某个部分是否由AI制作可能出于单纯的好奇或技术探讨，但是在LD60中这可能变成了一种隐含的批评。特别地，对于选择不使用匿名来进行评价的评论者，多数人选择用更礼貌、客气的话语进行评分，一部分原因也是担心自己的游戏被报复性地打低分，因此背后隐藏的态度很可能是模糊的。从图5可以看出，“使用AI制作游戏”，“游戏内运行的AI”， “AI撰写的评论或帖文”三者收获了完全不同的评论：“使用AI制作游戏”是最普遍、最多样的一种讨论，但是Says AI was used and not admitted、Praises not using AI、Thanks for disclosing or opting out这类具有对AI生成作品整体反对态度的评论集中在最近几届，尤其 LD60；对于“游戏内运行的AI”，批评的声音几乎没有，多数停留在评价效果与技术讨论，只有少数讨论涉及到规则是否允许；对于“AI撰写的评论或帖文”，整体数据都非常少，而且主要在LD60才出现，属于比较新兴的现象。具体来说，是一些用户利用系统规则使用AI批量、快速地给大量游戏评价，并且提升自己游戏的热度、吸引这些作者去给自己的游戏评分。尽管用户普遍对此表示反感，但目前的规则没有办法禁止这一行为。

那么，具体到AI在游戏制作中的不同用途（图3），每种用途是分别如何被评价的呢？我们分析了这部分评论中批评的评论所占的比例。由于数据量比较有限，因此以Writing为代表的类别无法从数据中得到有统计意义的偏好。不过，从图中依旧可以明显辨认出整体的趋势：对于几乎所有有数据的类别，评论中对于使用AI的批评呈现出先增加后维持在比较高水平位置的趋势。

在讨论AI使用与评分中的关系前，首先需要想办法确定如何判断一个游戏是否使用了AI。最简单的方法是通过游戏创作者的声明，但是其未必诚实，且并不是每个人都进行了声明。如果要进行人工判断，那么标准是难以确定的；特别是对于游戏代码来说，在不是所有代码库都开源的情况下，有些时候几乎无法判断。此外值得一提的是，“AI声明”似乎有一种很微妙的作用，它也允许作者解释自己为什么要使用AI，例如一个只擅长美术与写作的人可能会采用AI来辅助编程，否则就没办法在三天内做出一个完整的游戏来进行自身的表达。就算仅仅看itch中的tag，不同的玩家可能会对不同的声明有不同的态度，例如与art / audio / text相关的tag使用就可能引发更多讨论。

尽管ldjam在机制上鼓励用户多评论，以此换取更高的“Karma”来为自己的游戏提高热度，往往是非匿名的评论和一定是匿名的评分在数量和态度上都很可能存在不小的差异。特别地，有些时候只有对AI态度比较明确好恶的人才会在评论中提及，而态度比较模糊或者中立的人可能不会提及或者参与相关讨论。对于评分，我们以LD59的游戏评分数据进行分析，其中包含1,142 个有排名的游戏，分组和设计来自 @rogual 整理的 itch.io 声明。从图8可以看出，与评论所展示的态度不同，声明了AI使用的作品得分并没有明显低于声明未使用AI或者未声明的作品。相反，仅仅使用AI辅助编程的游戏没有明显区别（平均分都是 3.55）甚至得分的中位数表现更好。这其中也包含compo组的winner，Boba’s Love Story ，尽管在compo中的AI的使用被大量玩家质疑，这款游戏仍然在评分中胜出，赢得了第一名。考虑到声明可能不实的情况，如果简单从这个角度来看的话，游戏创作者如果想要得到高分，理论上可以使用AI进行代码辅助、进行不实的声明或者干脆不声明，尽管这对其他参赛者很可能是不公平的情况。

4 What should we do?

通过以上的数据，我们已经可以对游戏开发中的AI的使用情况、评论与评分的情况、社区整体的态度有一个大致的概览。大略地来说，AI参与到游戏的制作中的比例逐届都在增加，并且从最初的美术、音乐、文本扩展到了代码和整体设计；AI作为NPC或者讨论的对象更多出现在早期，随着人们对AI的习以为常反而不再那么高频地作为讨论的题材了；人们用AI来批量spam评论的现象开始出现，spam评论一直都有，以前可能更多是复制粘贴一些同质化的general的评论，但是直到LD60才主要以AI驱动。人们在这10届对AI的讨论从未停止，而且具体的话题也不断在变化。在态度上，我们很明显地可以看到评论中对游戏开发中的生成式AI越来越讨厌，但是在评分上似乎看不到很明显的这个倾向。

当然，即使有了这些数据，我依旧无法对 I don’t know how to rate “AI-assisted” games 帖子中每个人的观点进行评价，也没办法说自己支持谁，因为似乎每个人都在真诚地讨论和真诚地困惑。而且，看似大家都在讨论“如何为AI辅助设计的游戏打分”这件事，实则又涵盖了不同而复杂的子议题，例如：AI在游戏开发中的本质角色是什么（作为工具、一种盗窃的手段，还是同事？）、游戏设计与开发的核心在哪里、评分本质上是用来干什么的、这场比赛是否公平、规则来自于哪里……每个人在各自的立场上可能都保持着一部分的正确。

目前来讲，可能争议最聚集的地方之一在于AI辅助编程，因为与AI生成的美术、文本不同，AI生成的代码虽然也具备特定的明显特征，但是往往无法直接作用在游戏界面中，因此不查看源码几乎无法识别，在ldjam中也没有专门的为代码评分的通道。而且这是一个相对比较新兴和缺少讨论的问题，因为在此之前AI编程水平还没有发展到可以独立制作出一个完整游戏的地步，自然也缺乏相应的规则。

作为计算机科学学生，我可以理解编程之美，但我也清楚绝大多数游戏是无法将其体现出来的——游戏流畅、优化得好当然是代码的功劳，可是玩家在体验的时候并不会因此想到“这个游戏背后一定有很optimised的代码吧！”而且编程之美也分成很多层次，例如一个小function可能空间时间复杂度很低、很聪明、很美，又或者整体的system design做得很好，很难判断究竟怎么样才算是好。对于我个人来说，编程是一种智力挑战，我在编程的过程中和结尾享受到这种乐趣，可是很难想象使用我程序的人也能感受到那种快乐。作为中文系出身的人，我也能理解没有足够的技术来表达自身想法的苦恼。AI的发展无疑对这样的人是有帮助的——即使是对已经很有编程经验的人，AI辅助编程也可能很有帮助，因为可以帮助你把精力放到更宏观的orchestration和游戏设计中。正如论坛中的评论：“使用人工智能并不意味着我突然忘记了如何编程。我的知识和经验仍然存在。我只是让开发过程的一部分更快、更容易，同时仍然控制我正在创建的内容。”²

问题是，游戏设计的本质是什么？用户@Willops说，“给这个敌人多少速度？给玩家多少跳跃高度？喷气背包的动力应该持续多久？对我来说，这些问题是游戏设计的本质”³，这话没错，可是游戏设计的本质对不同的人、不同的游戏体裁来说很可能是不一样的。而且如果一个人是从YouTube之类的地方学到要怎么设计敌人速度，那这和从AI对话学来“建议的敌人速度”有什么区别呢？我认为重点是人在整个创作过程中发挥的作用，例如对于每个具体的操作手感，你在Godot或者unity里面会打开进行测试、验证、调整；对于整体的设计，你按部就班把所有条件与触发机制写出来。但是这个边界太模糊了，如果一个人不坚守好底线，很容易就会let AI decide everything，制作出一个根本不属于你自己，却由你署名甚至让你赢得奖项的作品。

在社区缺乏mod和公约的情况下，规则来自于哪里？ld用户@Jezzamon发起了投票并将目前结果展示在ld-votes。有趣的是，目前的结果接近于平手。考虑到ldjam即将结束，我很怀疑之后是否会公布新的规则，以及人们是否会遵守那些规则。一直以来都是如此，单纯依靠道德约束的时候，违反规则者总会很容易赢。

不过说到底，ldjam不是一个以竞赛和得分为目的的比赛（也很难说世界上存在真正公平的比赛）。尽管一定会有人拿在其中获得的名次当噱头，多数人真正的目的还是给自己一点动力和限制，在有限的时间和题材内去尽量好地完成一款游戏，这是对自己能力的挑战，也是表达自我的机会。游戏互相评价本质也是尽可能多的让自己的游戏被看到、来收获评价，同时也是游玩其他游戏来扩充自己视野、丰富灵感的机会。用不用AI、如何用AI，是被AI推动着完成整个设计、被侵蚀失去自己的想法，还是让AI完成底层工作来用更多时间进行核心设计，在目前来看，本质还是个人的选择。

当然，游戏开发与ldjam中的相关使用与讨论不过是当前AI的诸多使用及其影响的一个微小的缩影。真正令人担心的，是生产力与生产关系变化后新的规则、理论的建立，以及那个模糊的，创意力的边缘。