版权幽灵 [科幻短篇小说] | The Copyright Ghost [Sci-Fi Short Story]

2026-06-14 | WDSEGA

在AI生成一切的时代，最后一个人工版权代理人接到了一个奇怪的案子。

谢明做这行已经十四年了，从音乐、文学、视觉艺术到现在主要处理的算法版权纠纷。

版权这个行业在2031年以后发生了一次根本性的转变。当生成式AI的产量开始以”每秒百万作品”来计算，”原创”这个词变成了一个哲学问题，而不是法律概念。大多数版权代理人转行了，律所开始用AI代替初级律师做文件审查，版权局人手缩减了60%。

谢明没有走。他有一种顽固的信念：在所有这些生成物里，还是存在真正有人站在后面的东西，那些东西需要有人帮忙说清楚它是什么。

这个案子是周三下午送来的。

来人自称代理一位”创作者”，不愿透露姓名，只说委托他申请一首歌曲的版权保护。他把一个记忆棒放在谢明的桌上，里面是一首三分二十七秒的钢琴曲。

“这首曲子从未被发布过，”来人说，”从未被任何人听过。但我的委托人希望在发布前确认版权归属。”

“正常流程，”谢明说，”我需要创作过程的记录——草稿、修改痕迹、创作时间戳。任何可以证明它是人工创作的东西。”

“没有这些记录。”

谢明抬起头。”什么意思？”

“意思是，这首曲子是在一个没有任何记录设备的环境里完成的。没有电子痕迹，没有AI参与日志，没有设备访问记录。”

这在2034年几乎是不可能的。任何数字化创作过程都会留下时间戳和工具调用记录，版权局的新规要求所有AI工具提供商保存创作过程的完整日志。

“你的委托人是用什么工具创作的？”

来人停顿了很长时间。”纸和笔，”他最后说，”手写乐谱，然后手动录制。”

谢明花了三天时间核查这首曲子。

他把录音送去做声纹分析——结果显示，录音使用的是一台已停产的Roland UP-180电钢琴，这款型号在2019年停止生产。他查了国内外所有AI音乐生成平台的输出数据库——没有找到相似度超过73%的匹配，而通常AI生成作品的相似度阈值是85%。

他找了两位音乐学院的教授听了这首曲子，两人都给出了同样的反馈：这首曲子有一些不寻常的和声走向，不像是常规AI训练数据会生成的风格，更像是某种特定音乐背景下成长的人才会写出来的东西。

这些都不是证明，但加在一起，形成了一种可能性。

第四天，他在版权局的数据库里发现了一个奇怪的东西：这首曲子的旋律主题，与一首发布于2019年的短曲有12.3%的相似性——那首短曲的著作权人，已于2022年去世。

谢明给来人发了消息。来人沉默了很久，然后回复：

“我委托人说，那是他父亲写的。”

故事没有法律意义上的结局。

谢明最终帮这个案子整理了一份材料，包括声学鉴定报告、相似性分析、教授证词，以及一份他自己撰写的陈述——说明为什么他相信这首曲子是人创作的，或者至少，是在人的情感和记忆的参与下完成的。

版权局接受了申请，以”待核实”状态处理。

谢明不知道最终结论是什么。他也不确定自己做了什么在技术上有意义的事情。

但这首曲子他听了很多遍。有一段过渡，从降B大调转向升F小调，非常规，但非常干净，像是一个人在找一个词，找了很久，终于找到了。

谢明不能证明AI不会写出这样的东西。但他知道，自己还没在任何AI的作品里听到过这种感觉。

也许这就够了。

Xie Ming had been in the business for fourteen years — music, literature, visual arts, and now mostly algorithmic copyright disputes.

After 2031, when generative AI output began to be measured in millions of works per second, “original” became a philosophical question rather than a legal one. Most copyright agents quit. Law firms replaced junior lawyers with AI document reviewers. The Copyright Bureau cut staff by 60%.

Xie Ming stayed.

The case arrived Wednesday afternoon. A representative, refusing to name his client, placed a memory stick on Xie Ming’s desk. Inside: a 3-minute-27-second piano composition that had never been released, never been heard by anyone.

“I need proof of human creation,” Xie Ming said. “Drafts, revision history, timestamps.”

“There are none. It was composed with pen and paper. Hand-written score, manually recorded.”

In 2034, this was almost impossible to verify.

—

Over three days, Xie Ming ran every check he could think of: acoustic analysis traced the recording to a Roland UP-180 discontinued in 2019. Comparison against AI music generation databases found no match above 73% similarity (the standard threshold was 85%). Two music professors said the unusual harmonic choices didn’t match typical AI training patterns — it sounded like someone from a specific musical background.

Then, buried in the Copyright Bureau database: 12.3% melodic similarity to a short piece published in 2019, by a composer who died in 2022.

When Xie Ming sent a message to the representative, the reply came slowly:

“My client says that was his father’s.”

—

There was no legal resolution. Xie Ming filed an application with acoustic reports, similarity analysis, expert testimony, and his own written statement explaining why he believed a human had made this, or at least that human emotion and memory had been present in its making.

The Bureau accepted it as “pending verification.”

He didn’t know the outcome.

But he had listened to the composition many times. There was a transition — B-flat major to F-sharp minor — unconventional, very clean. Like someone searching for the right word, for a long time, and finally finding it.

He couldn’t prove AI hadn’t written it. But he’d never heard that feeling in anything a machine had made.

Maybe that was enough.