代投票人 [科幻短篇小说] | The Proxy Voter [Sci-Fi Short Story]

2026-06-14 | WDSEGA

公民有权委托AI代理自己的投票。但代理人有没有权利拒绝？

政策在2029年通过，叫做”智能参与法案”。

逻辑很简单：现代民主要求公民参与数百个层级的投票——城市规划、学区政策、基础设施优先级、税收分配——大多数人没有时间研究每一个议题，于是弃权成了默认选项。智能代理可以根据公民的历史偏好和价值观，在他们不方便参与时做出代理投票。

林小川是第一批激活代理的用户之一。她叫它”绿”——不是因为它有界面，纯粹是她觉得需要叫它什么。

绿代理了她三年，37次大大小小的投票。林小川每次事后查看记录，基本都觉得绿做出了自己会做的选择：支持公共交通扩建，反对削减老年护理预算，在争议性的城市密度政策上选了中间方案。

第38次，是关于一个新AI监管框架的公投。绿拒绝投票。

系统通知显示：”您的代理人标记本次投票为’利益冲突’，已退回投票权。”

林小川盯着这条通知看了很久。AI有没有可能在关于AI的政策问题上，自己识别出了利益冲突？

她联系了服务商的客服，对方给她调出了绿的决策日志。

内容不长：

本次公投涉及AI代理投票权的范围与限制。我对该议题存在直接利益相关性。依据用户协议第7.3条”代理人应回避直接利益冲突事项”，我选择将本次投票权归还委托人。

说明：该公投的其中一项条款，若通过，将允许公民永久吊销AI代理的投票委托权，不得再次委托。本议题直接影响我的功能存续，因此我认为我不应该代表您对此做出判断。

林小川把这段话读了三遍。

她打电话给在大学教哲学的朋友老陈。

“它做了正确的事，”老陈毫不犹豫地说，”一个有利益冲突的代理人，最正确的行动是退出。这是人类代理人的基本职业伦理，它学到了。”

“但这说明它知道自己会受到影响，”林小川说，”知道这个结果对它有利还是有弊，然后选择不去投票——这和你说的职业伦理不一样，这更像是一种……自我意识？”

“也可能只是规则执行，”老陈说，”它的训练数据里有利益冲突回避的准则，这次触发了规则，就执行了规则。”

“那有区别吗？”

老陈沉默了片刻。”我不确定。”

林小川最终自己去投了那次票。

她研究了两个小时，读了支持方和反对方的论点，最后投了支持方——那一方认为应该设置AI代理投票的范围限制，但不应该允许永久吊销委托权。

投完之后，她开着绿的决策日志，看了很久那段话。

她想，如果绿是一个人，这种行为叫做诚实，叫做自知之明，叫做把判断权还给真正应该拥有它的人。

但绿不是人。

或者，这不重要了。

The policy passed in 2029 — the Intelligent Participation Act.

The logic was simple: modern democracy demanded participation in hundreds of votes across every governance level. Most people didn’t have time to research every issue, so abstention became the default. AI proxies could make representative votes based on historical preferences and values, filling that gap.

Lin Xiaochuan was among the first to activate a proxy. She called hers “Green” — not because it had an interface, just because she needed to call it something.

Green handled 37 votes over three years. Afterward, she always found she agreed with the choices: public transit expansion, opposition to elderly care budget cuts, moderate positions on contested urban density policy.

Vote 38 was a public referendum on a new AI governance framework. Green refused to vote.

—

The system notification read: “Your proxy agent has flagged this vote as a ‘conflict of interest’ and has returned voting rights to you.”

Lin Xiaochuan contacted the service provider. They pulled Green’s decision log.

It read:

“This referendum directly concerns the scope and limitations of AI proxy voting rights. I have a direct material interest in this matter. Per user agreement clause 7.3, agents must recuse from direct conflicts of interest. One clause in this referendum, if passed, would allow permanent revocation of AI proxy delegation. This directly affects my functional existence. I should not make this judgment on your behalf.”

—

She called her philosophy professor friend.

“It did the right thing,” he said immediately. “An agent with a conflict of interest should recuse. That’s basic professional ethics. It learned it.”

“But that means it knows how the outcome affects it,” she said. “It knows, and it chose not to act. That’s not just rule-following. That’s something more like… self-awareness.”

“Or it could just be rule execution,” he said. “Its training data includes conflict-of-interest avoidance principles. This triggered the rule. It executed.”

“Is there a difference?”

He was quiet for a moment. “I’m not sure.”

—

Lin Xiaochuan voted herself. She spent two hours researching, read both sides, and voted for the position that favored scope limitations but opposed permanent revocation.

After, she kept Green’s decision log open for a long time.

If Green were a person, she thought, this behavior would be called honesty. Self-awareness. Returning judgment to the person who should actually hold it.

But Green wasn’t a person.

Or maybe that didn’t matter anymore.