格拉迪亚斯听见有人敲房门——他住的是政府经营的酒店。还没来得及应声，他就听见一阵嗡嗡声，门开了。
塞拉走了进来。
“对不起，塞拉”，格拉迪亚斯几乎是马上开口，脸上写满难过。“我觉得这个方案行不通。”
“我把全家都辜负了。我们知道费布里克和赫雷达在哪，现在也知道莉莉在哪，可这几个人一个都救不出来——除非北极人把我们彻底征服。”
塞拉走到格拉迪亚斯身边，抱住了他。
两人沉默了片刻。然后她退开一步，脸上绽出笑容。
“所以，我有了个新主意。”
格拉迪亚斯眼睛一亮。
“什么主意？”
“昨天我们弄清了一件事：法官们不愿意做他们觉得很不民主的事。所以我有办法了。我们把它变成民主的。”
“怎么，明天就办一场选举？”
“某种意义上，正是。”
“你在泽国的时候，我和莫夫开始在银聊里一级一级往上爬，跟那边的人搭话。一开始没能引起他们的注意。大约10天前韦尔多加入进来，帮了大忙，可即便那样，他们也没太把我们当回事。但昆高培救出塔芬德尔那天，我冒了一次险：我提前一个长时告诉他们，军事上会有件有意思的事发生。后来当然就发生了。所以现在，他们愿意听了。而我已经想好了该请他们做什么。”
“银聊有个民调功能。你可以烧掉齐普币——嗯，是把齐普币付给公司——它就会自动向平台上很大一部分人提问。你烧的齐普币越多，它问到的人就越多。所以我们要做的就是，问一大批人一堆问题：问他们对维里迪亚和其他国家的各种政府机构怎么看，以及在一些假设情形下会怎么想。”
“我们会把这些问题放进去：‘你认为议会代表人民吗？’‘如果维里迪亚军方由那群救出红郡总统之子的泽国人来掌管，你会怎么想？’但我们把它们混在几百个别的掩护问题里。这样北极人就搞不清我们在干什么。”
“要让回答真实、有思考，得把优先级设得很高，我猜那要花很多齐普币。而要让这成为一个合法性很强的社会信号，就必须形成共同知识——维里迪亚的每个人不仅要知道民调做过、结果是什么，还要知道别人也知道民调做过、结果是什么，还要知道别人也知道别人知道，如此往上。那就需要全国人口的很大一部分都看到这次民调。所以我们得花一大笔齐普币。”
“我估算要90万齐普币。”
“我们有那么多吗？”
“没有。我们所有的积蓄加在一起，连莫夫那份也算上，只有一半。”
“但诀窍在这儿。我们不用出这笔钱。我们去说服银聊的创始人埃维洛尔，让他自己以公司的名义来做。那样他们想把覆盖面设多广、优先级设多高都行——费用他们能100%拿回来。”
“等等，他们不能直接手动做吗，一分钱都不用花？”
“他们的平台是这样运作的：每当可读消息的集合更新，就往密码学网络上发布证明。所以他们根本没法违规——真要试，证明就通不过校验，用户端会自动把那些消息判为无效。他们能做的只有公布一个新的算法哈希，可那会附带20天的延迟。”
“那他们能100%拿回来？不用交销售税？”
“评则最近改过之后，考虑到银聊在算法透明度上的表现——正是那套证明机制，让他们没法直接开后门、破自己的规矩——再加上互操作性、尊重用户隐私、配合研究者，他们的税率已经降到零了。”
“那就看看埃维洛尔怎么说！”
格拉迪亚斯、塞拉和泽一起沿着他们走惯的那条路穿过卡利马尔。路旁绿树成荫，两侧都是石头房子。为安全起见，他们坐黑色轿车走了大半程，但车只能开到无车区的边界。于是最后几百米，他们只能步行。希望隐私袍够用，格拉迪亚斯想。
一阵冷风从背后吹来。寒意让他们振作，风又推着他们往前，他们不由得加快了脚步。
“这里右转”，塞拉通过颈带低声说。到这时，她已经几乎和格拉迪亚斯一样，习惯了所有对话都通过颈带来进行。
他们右转，上了一条更窄的小路。小路笔直延伸了大约100米。起初两侧还是石头房子，大约50米后，房子退去，换成一片看似幽深的树林。走到100米处，他们看见前方有一座小丘似的东西。
小路向下倾斜，通向一扇门，门里似乎就是那座小丘的内部。又是一栋几乎不缴土地税的建筑，格拉迪亚斯想。到这个时候，维里迪亚人几乎已经把一门手艺练到了家：在树林里生活、工作，把复杂的建筑安置进去，而同一片树林从100米外看去仍像是没人碰过。
泽惊叹地看着眼前这一幕，脑子里立刻拿它跟泽国的山体金字塔作比。在泽国，他想，人们把天然的山凿成金字塔，让自然看上去像人的作品；而在维里迪亚——至少是梅尔丹的卡利马尔区——人们把一座很可能完全人造的小丘做得像自然。泽国可以从维里迪亚这种风格里学点东西，他想。用来防护，说不定还更好。
塞拉把手持终端贴上门边那个绿色圆圈，门开了。
她带着另外两人沿一条路往里走：先左转，再右转上一段坡，又右转，然后登上一道楼梯。最后她打开另一扇门，领他们走上楼梯，来到一片圆形空地的中央，从那里能看见开阔的天空。格拉迪亚斯和泽这才明白，他们已经在丘顶了。
他们转过身。另一名男子已经等在那里，穿着隐私袍，但没戴兜帽，也没遮脸。
“我是埃维洛尔”，他说。“很高兴见到你们。”
“我是格拉迪亚斯。我也很高兴见到你。”
“我是泽。”
埃维洛尔按下一个按钮，一个屋顶在他们头顶开始合拢。大约10嘀嗒后，屋顶完全合上了。
“只是为了隐私”，埃维洛尔说。“袍子尽管脱。另外，有谁想喝水或喝茶吗？”
“我要茶”，格拉迪亚斯答道。
“我也要茶”，泽说。
塞拉四下看了看。“那我也来杯茶吧。”
“好，那就4杯茶。机器人马上上来。”
埃维洛尔在一条长凳上坐下，长凳紧贴着楼梯出口后方的安全栏杆。他指了指沿着房间一整圈的那条环形长凳。“请坐。”
他们摘下隐私袍的兜帽和遮面，坐了下来。
“那么，塞拉，你跟我提过，你有个想法：银聊能做点什么，让整个维里迪亚政府别那么蠢”，他脱口而出，语气直白得毫不客气。
“是的，我有。”
“简单说，我想搞一次民调。我让翡翠拟了大约1,000个问题，我自己过了一遍。有些问你对现状的感受，有些问可能出现的未来。其中一小部分问题直指我们真正在意的那件事——就算议会不愿意，也要把军方换掉——其余的都是掩护。”
泽微微皱眉。塞拉太快就把整套计划的底细都交了出去，他想。昆高培绝不会这么做。可也许维里迪亚的文化就是更愿意信任别人——又或者，埃维洛尔已经在塞拉、甚至在整个维里迪亚面前证明了自己值得信任。
“目标是形成共同知识：这次民调做过，结果是什么。如果一切顺利，我们去跟法官论证这是民众真正想要、不会因此闹起来的事，底气就会足得多。”
塞拉把手持终端拿给埃维洛尔看。埃维洛尔翻了几页问题，开始念出声来。
“你更希望谁来掌管维里迪亚军方？A：现任领导层。B：由议会选出的新领导层。C：银聊的领导层。D：北极人。E：昆高培（最近救出红郡总统之子的泽国人团体）。F：阿图里亚军事顾问。”
“那你打算在这上面花多少钱？”
“300万齐普币。而且我希望银聊这家公司自己来做，因为不管怎样这笔钱都会100%返回来。”
“有意思！不过我觉得300万太多了，比银聊上做过的任何民调或广告都多。那会让北极人怀疑出了什么不寻常的事。我会花100万。花费还是能排进前三，但算不上史无前例。”
“我没问题。”
“但还有另一个问题”，埃维洛尔接着说。“如果我们这么做，能帮法官看到，做出大改变是得人心的，民众也并不觉得议会多有合法性。当然，也可能我们发现，自己对维里迪亚人想法的判断错得离谱。但这不会让他们相信，这么激烈的举措真的行得通。”
“你心里有办法了？”
“其实有。最近我们加了个功能，叫银聊预测，让人——或者机器人——对未来事件下注。到目前为止，机器人是最有效的——”
“我看到了”，格拉迪亚斯插话。“预测市场很酷。但有两个弱点。第一，如果北极人看到这个市场，他们可以操纵它。理论上，任何操纵的企图都会让套利者更有动力进场，靠把价格推回该在的位置平均赚一笔，但实际效果如何，还远远没有经过检验。”
“已有结论表明：套利者若不是完全风险中性、或没有无限多的钱，操纵者总能至少让价格挪动一点。而如果标的变量本身是前所未有、无人知晓的东西——比如我们能不能赢下过去100年里全世界所见过的对北极人的第二次胜利——那就很难有套利者愿意进场。”
“第二个问题是私有信息。我们对昆高培有信心，因为我们了解他们、信任他们，也清楚他们的能力有多大。可一旦公开，就等于告诉了北极人。”
“对，两点都说得对。第一个问题，希望我们能随着时间慢慢改进，但……眼下预测市场太新了。第二个问题更难解决。”
“好，我有个不同的想法。我们不用市场本身，而是取排行榜上表现最好的5个预测机器人，用可验证两方计算去问它们。我们造一个临时的虚拟私密沙盒，把私有信息喂进沙盒，再让机器人自己进沙盒，沙盒只输出机器人的答案。”
泽看到了开口的机会。“对，那不过是混淆电路的延伸——你多半会想要可验证混淆电路。开销相当难看——”
“针对 AI 计算，有一些专门的优化。别忘了，大语言模型内部几乎一切都是线性的。要让语言足够聪明，非线性那部分是少不了的，可它只占整个计算量的0.1%。”
泽意识到，在这个专门的题目上，自己不是对手。看来学了9个月密码学，也没能让他在所有事情上都比所有人聪明。
“原来如此……”
谈话暂时停下——他们听见下面传来一阵机械声，越来越响。几嘀嗒后，他们看见了机器人的顶部：它端着托盘上的4杯茶，一级一级往上爬，每上一级都先向上抬升、再横向滑移。到了楼梯顶端，它先滑到格拉迪亚斯面前，把茶递给他。
“格拉迪亚斯”，埃维洛尔开口道。“你看起来挺聪明。我另外有个问题想问你。”
“什么问题？”
“你还记得埃菲里昂和西尔卡的那场辩论吗？”
“记得。”
“埃菲里昂说：社会上那些‘软性’特质大多在循环。好孕育出坏的种子，坏又孕育出好的种子。从灾难到英雄主义，到繁荣，到自满，再回到灾难。技术官僚、民主、混乱、强人，再到技术官僚。如此往复。唯一不这样运行的，是经济总产出：增长只会自己不断复利。所以我们该停止加速那些循环的东西，转而只专注于把经济总产出做大。我不喜欢这个论证。它暗含的东西让我不舒服。可我还是说不清它到底错在哪。你怎么看？”
“我认为会自我复利的不只是经济总产出。很多好东西也会彼此复利：健康让人更有精力做事，效率又让人有资源变得更好，教育撑着这两样，而健康也撑着教育。”
“也就是说，埃菲里昂谈的那个特征向量不止一个维度——而大多数特征向量其实都是这样。”
“你对特征向量的理解明显比我好！”
“银聊是个社交媒体网站。特征向量……差不多就写在原始白皮书的第一页上。”
格拉迪亚斯笑了。
“但另一方面”，埃维洛尔接着说，“北极人确实在意健康。那可不是西尔卡想要的那种同情、爱与关怀的审美——而是冷硬的金属质感长寿诊所，但它仍然是健康。”
“西尔卡会说，重要的不是活得长，而是活得健康。”
“对，可中世纪那些喝了长生不老药、结果注定要骨头尽碎、永远躺在地上的故事，跟长寿科学实际的运作方式完全不是一回事。活得长，默认就是活得健康。要阻止损伤，最容易下手的地方就是它刚开始的那一刻。”
“那我猜你也会说，北极人在意教育。他们几乎不懂音乐和诗歌，但教孩子科学教得很好，而那本来就是教育里进入最大特征向量的那部分。”
“对。”
格拉迪亚斯能看出，塞拉正不舒服地皱着眉。他脑子飞快转着，想找出别的论据来反驳埃菲里昂那套想法。
“可北极人并不是真要最大化经济总产出。他们想发动战争——战争能让经济几十年翻不了身。他们想除掉泽国里任何在推进科学技术上走得太远的人，那也不是在最大化经济总产出——恰恰相反。他们要的不是最大化经济总产出，而是最大化自己掌控之下的那部分产出，再把其余的一切都毁掉。”
这时，泽开口了。
“埃菲里昂暗含的模型是这样的：有一个向量朝着无穷远射出去，它跑得越快，所有人也就越快得到幸福。可我看到的世界根本不是这样。”
“我看到的世界是个脆弱的世界。它并非一直如此脆弱。过去，通讯有限，权力的传递也有限。任何帝国统一得够久，最终都会垮掉。世界是可逆的。可今天，有了现代的机器人和现代监控，一旦任何人成功接管并巩固了对全世界的控制，就能把自己的权力永久锁住。不可逆。永远。”
“如果北极人赢了，某种熵——人类未来可能性空间的维度——就会被永久压缩。从那一刻起，唯一可能存在的未来世界，就只有北极的世界。”
“而这还不是最糟的。如果北极人，或者别的什么人，用他们的超级武器犯了错，而再没有别人有能力抵抗，就可能把整个生命世界彻底毁掉。”
“最终，世界会重新变得可逆。我们将能在群星之间旅行，我们的文明会大到没人能完全控制、也没人能彻底毁灭。但在那一天到来之前，我们身处一个脆弱的阶段。如果接下来10年里有人把这一切乘以零、永远回不来，那罪魁祸首就是北极人。”
“我有个更简单的答案”，塞拉补充道。“所有这些关于遥远未来的论证都很难预测。而且，如果一条路你本就认定要走，那要说服自己这条路是对的，再编出一套精妙的数学来给它背书，实在太容易了。但有一件事很好预测：如果北极人征服了维里迪亚，这场战争可能杀死我们数十万乃至数百万人，北极人还会毁掉我们所知所珍爱的那个美丽世界。对我来说，光这一点就已经值得抵抗了。而我们这个世界、这些国家、这些友谊的美，也已经值得为之而战。”
“都是好答案”，埃维洛尔答道。他抿了一口茶，往椅背上一靠，沉思起来。
3724年霜期8日
格拉迪亚斯、塞拉、泽、白、韦尔多、登和穆围坐在政府经营的酒店一间会议室的桌旁。他们面前墙上投着一个计时器，倒数着银聊上那份问卷发布还剩多少嘀嗒。此刻剩下的时间是150嘀嗒。
“你们把要放进预测机器人那个数据包的信息都提交完了吗？”塞拉问。
“我这边刚弄完”，登答道。
“记住，它是个机器人。用不着弄得多漂亮、多讲究。你要是又想起什么要说的，直接扔进去就行。”
“我希望你和你的机器人把密码学部分审好了，泽。我可是把很多军事机密都托付给它了。”
“有一份可机器检验的数学证明：混淆电路算法满足标准的正确性与隐私性质。这份证明是端到端的，从操作系统一路覆盖到硬件。它证明的那些命题我读过，还让翡翠和另外两个 AI 核验过。相当扎实。除非哈希被攻破，否则零信息泄露——而真到那一步，我们有更大的麻烦。”
“我也查过”，穆插话。“我用的是另一种方法，免得证明系统或那些定义十几年前就在哪儿坏掉了、却一直没人发现。我让翡翠去读原始学术论文，让它自己写一份实现，再让一个全新的实例把那份代码和原版并排读，逐处比对差异，看哪边是对的。埃维洛尔的代码毫无瑕疵。”
“还有50嘀嗒就开始了！”
“这次不学民本棋比赛播报员的腔调了？”白回嘴打趣。
"MU GU GEI HUI ZIU FA"，泽用一种夸张的机器人腔调答道，为了逗乐。
他们围坐在桌旁等着。
"LE MU GEI HUI ZIU FA"，白喊道。
“喂！”泽回了一句。
“抢在你前面了！”
他们继续盯着倒计时。倒计时一跳到十，他们就齐声喊起来：
"PA GU …… SO …… BI …… ZE …… HA …… MU …… FO …… SHI …… LE …… PA …… HUI ZIU FA!"
屏幕分成两半。左边是5条并排的进度条，显示与5个预测机器人进行的5次混淆电路计算的进度。右边是一大片饼图，每一个问题一张。投票很快开始涌进来。
“放大我们关心的那几张”，韦尔多提议。
泽点头，点开了其中一张饼图。
“跟我预想的一样。现在看看大家觉得这个问题该怎么解决。”
“限定为北岸来的人。”
“等等，我们还做了这么细的分项？投票是匿名的，我们没法知道每一票是谁投的。”
“我们确实按年龄、自己申报的上一次议员投票、省份，还有大语言模型推断出的几个性格类型维度做了细分。”
“啊，找到了。”
“恭喜你，埃维洛尔”，韦尔多干巴巴地说。“看来你比真正的议会更适合管北岸。”
“这些结果现在让我有点害怕”，塞拉插话。“我们是不是刚把底牌全亮给北极人了？”
“我觉得没事”，穆答道。“这次民调我们问了很多不同的问题。其中一个问题，实打实有20%的人说，我们该出资恢复北岸——办法是依法把北岸所有政府持有的土地都转给海德菲尔，让他们自己去想办法。”
“那桩诉讼，北极人当然会注意到，也会盯着。我们该推迟几天，免得看起来像是事先串通好的。但不管我们怎么做，我们都已经让所有人——包括北极人——清楚看到，维里迪亚人认识到存在一个大问题。需要藏起来的是解决方案的细节。”
他们停了一会儿，盯着进度条。
“好，那还得等一阵。我们再看几张饼图。”
一个机器人进了房间，给每个人端来一杯茶。
“好了……完成了”，泽得意地喊。
屏幕上出现了5个图标，代表5个预测机器人给出的回答。泽点开了第一个。
摘要：
- 在任何“寻常”情形下——即维里迪亚军方维持现状，或由维里迪亚议会在没有巨大外部压力的情况下主导重组——一年之后，维里迪亚保有的领土比例不低于当前水平的概率是20%
- 假设移交给昆高培掌控，该概率升至26%
- 主要的下行风险是常见的组织过渡成本，即领导层剧烈更替、用外来者替换有经验人员时所特有的那类成本
- 在昆高培参与、同时把过渡的震荡降到最低的情形下，成功结果的概率最高（可达50%）
- 注意，上述概率对应的是一年内控制力不低于当前。控制力严格提高的概率则普遍低得多
“嗯”，韦尔多咕哝道。
“你觉得合理吗？”格拉迪亚斯问。
“这些预测机器人显然比我更保守，但说得公道。”
“另外4个说的也大致相近。最悲观的那个说，现状情形和昆高培情形下，概率一样。”
“而且投票那边还有多得多的数据在进来。结果跟民调刚开始时我们第一次看到的大致一样。”
“那我们能拿到的数据大概就这些了。准备几天内立案吧。”
Gladias heard a knock on the door of his government hotel room. Before he could respond, he heard a buzzing sound, and the door opened.
Seila walked in through the door.
"I'm so sorry, Seila", Gladias said almost immediately with a sad expression on his face. "I don't think this plan is going to work."
"I've failed our whole family. We know where Febric and Hreda are, now we know where Lily is, but there's no path to rescuing any of them - except if the Arctics conquer us completely."
Seila walked up to Gladias and hugged him.
For a few moments, neither of them said anything. Then she stepped back and her face broke into a smile.
"So I have a new idea."
Gladias's eyes perked up.
"What is it?"
"So what we learned yesterday is that the judges do not feel comfortable doing something that they perceive as being very undemocratic. So I have a solution. We make it democratic."
"What, like we hold an election tomorrow?"
"In a sense, exactly."
"While you were in Dzego, Mov and I started making our way up the ladder in Silverchat, talking to people there. We were not too successful at getting their attention at first. Verdow joining in about ten days ago helped a lot, but even then they didn't take us too seriously. But on the day the Kungaupei rescued Tafindel, I took a risk: I told them something interesting military-wise would happen a longhour in advance. And then of course it did. So now, they are listening. And I have a plan for what to ask them."
"Silverchat has a polling feature. You can burn zipcoins - well, pay zipcoins to the company - and it lets you automatically ask a question to a large cross-section of people on the platform. The more zipcoins you burn, the more people it polls. So what we'll do is, ask a lot of people a bunch of questions about how they feel about various Veridian and other government institutions, and other hypothetical scenarios."
"We'll include 'do you think the parliament represents the people?' and 'how would you feel if Veridia's military was run by the Dzegojan people who rescued the Redshire president's son?'. But we include that alongside hundreds of other decoy questions. That way, the Arctics won't know what's going on."
"For the responses to be genuine and well thought out, you need a high priority level, which I imagine would require paying a lot of zipcoins. And for this to function as a social signal with strong legitimacy, there must be common knowledge - everyone in Veridia must know not just that the poll was made and what the results are, but also that everyone else knows that the poll was made and what the results are, and also that everyone else knows that everyone else knows, and so on. That would require a big chunk of the entire population to see the poll. So we would have to pay a lot of zipcoins."
"Nine hundred thousand zipcoins was my estimate."
"Do we have that many?"
"Nope. All our savings combined, including Mov, adds up to only half that."
"But here's the trick. We don't need to pay. We convince Evelor, the founder of Silverchat, to do it himself, through the company. Then they can set the breadth and the priority as high as they want - they get a hundred percent of the fee back."
"Wait, can't they just do it manually at zero cost?"
"Their platform works by publishing proofs onto a cryptographic network every time the set of readable messages is updated. So there's literally no way for them to break the rules - if they tried, the proof would fail to verify, and the viewers' clients would automatically reject the messages as invalid. The only thing they could do is publish a new algorithm hash, but that would come with a twenty-day delay."
"And they get one hundred percent back? No sales taxes?"
"With the latest rubric changes, given how well Silverchat has been behaving with its algorithmic transparency - the exact proof mechanism that doesn't let them just directly backdoor and break their own rules - and interoperability, respecting users' privacy, cooperating with researchers, their taxes have dropped to zero."
"Then let's see what Evelor thinks!"
Gladias, Seila and Zei walked forward together along their usual path through Kalimar, covered with trees and surrounded by stone houses on both sides. They had taken a black car most of the way there as a security precaution, but cars could only go up to the border of the car-free zone. Thus, they had no choice but to go the last few hundred meters on foot. Hopefully the privacy robes will suffice, Gladias thought.
A cold breeze blew at them from behind. The cold energized them just as the wind pushed them forward, and they began to walk forward faster.
"Right turn here", Seila whispered through her neck band. By this point, she had become almost as accustomed to having all of her conversations through neck bands as Gladias had been.
They turned right, onto a smaller path. The path continued straight for about a hundred meters. At first, there were stone houses on both sides. After about fifty meters, the houses gave way to what seemed like a deep forest. And after a hundred meters, they could see in front of them what looked like a small hill.
The path sloped downwards, and led to an entrance door to what seemed to be the inside of the hill. Yet another one of those buildings that pays almost no land taxes, Gladias thought. By this point, Veridians had come close to perfecting the art of living and working and putting complex structures into a forest in such a way that the same patch of forest would seem untouched from only a hundred meters away.
Zei marveled at the sight, and his mind jumped immediately to comparing it to the mountain pyramids in Dzego. In Dzego, he thought, they crafted natural mountains into pyramids, and made nature seem like the work of man. In Veridia - or at least in the Kalimar district of Meldan - they made what was quite possibly an entirely manmade hill look like nature. Dzego could learn from the Veridian style, he thought. It's probably even better for protection.
Seila tapped her hand device against the green circle beside the door, and the door opened.
She led the three of them down a path: turning left, then right and up a slope, then right again, then up a staircase. Finally, she opened another door, and led them up a staircase into the middle of a circular area from where they could see the open sky. They were at the top of the hill, Gladias and Zei realized.
They turned around. Another man, wearing a privacy robe without the hood or the face covering, was already there.
"I'm Evelor", he said. "Nice to meet you."
"I'm Gladias. It's a pleasure to meet you too."
"And I'm Zei."
Evelor pressed a button, and a roof started to close above them. After about ten ticks, the roof was completely closed.
"Just for privacy", Evelor said. "Feel free to take off your robes. Also, would any of you like a water or a tea?"
"I'll have a tea", Gladias replied.
"I'll have a tea too", Zei added.
Seila looked around briefly. "Then I guess a tea for me as well."
"Sounds good, four teas it is. The robot will be up shortly."
Evelor sat down on a bench pressed against the safety railing behind the opening that the staircase emerged from. He gestured at the circular bench that was all along the circumference of the room. "Please, sit down."
They took off the hoods and face coverings from their privacy robes, and sat down.
"So, Seila, you mentioned to me that you had an idea for what Silverchat can do to make the whole Veridian government less stupid", he blurted out with a tone of blunt honesty.
"Yes, I do."
"Basically, I want to do a poll. I had Emerald come up with about a thousand questions, I reviewed them. Some about how you feel about the present, some about possible futures. A small fraction of the questions are getting at the thing that we actually care about - replacing the military even if the parliament doesn't want to - and the rest are decoys."
Zei winced slightly. Seila was volunteering the content of her master plan too quickly, he thought. The Kungaupei would never operate in this way. But maybe Veridian culture is just more trusting - or maybe Evelor had already proven himself to Seila, or even to all of Veridia, to be someone worthy of trust.
"The goal is to create common knowledge that the poll happened and what the results are. If it goes well, then we will have a much stronger argument to take to the judges that this is something that people would actually want, and would not riot against."
Seila showed her hand device to Evelor. Evelor briefly scrolled through the questions, and started reading out loud.
"Who would you prefer to run the Veridian military? A: its current leadership. B: a new leadership chosen by parliament. C: the Silverchat leadership. D: the Arctics. E: the Kungaupei (Dzegojan group that recently rescued the Redshire president's son). F: Arturian military advisors."
"And how much did you want to spend on this?"
"Three million zipcoins. And I was hoping that Silverchat the company could do it itself, because you get paid back a hundred percent anyway."
"Interesting! I would say three million is too much, it would be more than any previous poll or advertisement on Silverchat. It would make the Arctics suspicious that something unusual is up. I would do one million. Still in the top three in terms of spending, but not unprecedented."
"Works for me."
"But there's also another issue", Evelor continued. "If we do this we can help the judges see that making a big change would be popular, and that the people don't see the parliament as being all that legitimate. Or of course we'll learn that we're very wrong in our own judgement of how Veridians think. But this won't convince them that this kind of drastic action would actually work."
"Do you have an idea in mind?"
"Actually, yes. Recently we added a feature, Silverchat Predict, that lets people - or bots - make bets on future events. So far bots have been the most effective-"
"I saw that", Gladias interjected. "Prediction markets are cool. But there's two weaknesses. One is that if the Arctics see the market, they could manipulate it. In theory, any attempt to manipulate creates more incentive for arbitrageurs to come in and profit on average by pushing the price back to where it should be, but it's still very untested how well that actually works in practice."
"There's known results that unless the arbitrageurs are either fully risk-neutral or have infinite money, a manipulator can always force at least some shift in the price. And if the underlying variable is something so unprecedented and unknown - like whether or not we will win the second victory against the Arctics that the world has seen in the past century - that makes a hard case for any arbitrageur to want to come in."
"The second problem is private information. We're confident in the Kungaupei because we know them well and we trust them and we know a lot about what their capabilities are. But if we reveal that publicly, we reveal it to the Arctics."
"Right, both good points. Hopefully we can work on improving that first one over time, but ... right now prediction markets are new. The second one is harder to fix."
"Okay, I have a different idea. We don't use the markets directly. Instead, we take the five bots that have been performing the highest on the leaderboard, and we use verifiable two-party computation to ask them. We create a sort of ephemeral virtual private sandbox, we feed the private information into the sandbox, they feed the bot into the sandbox, and the sandbox only outputs the bot's answer."
Zei saw the opportunity to speak. "Yep, that's just an extension over garbled circuits - you probably want verifiable garbled circuits. Overhead is quite nasty-"
"There's optimizations specific to AI computation. Remember, almost everything inside a large language model is linear. The nonlinear part is necessary for the language to be smart, but it's only a tenth of a percent of the entire computation."
Zei realized that, on this particular specialized topic, he had been outmatched. Looks like in his nine months of studying cryptography he had not managed to become smarter than everyone at everything.
"I see..."
Their conversation briefly paused as they heard a mechanical noise getting louder and louder below them. A few ticks later, they could see the top of the robot, slowly climbing up the staircase with a raising motion and then a sliding motion at each step, carrying four cups of tea on a tray. Once it got to the top of the staircase, it moved toward Gladias to give him his tea first.
"Gladias", Evelor asked. "You seem smart. I have a different question I wanted to ask you."
"What is it?"
"Do you remember Ephelion's debate with Sylka?"
"I do."
"Ephelion said: most of the 'softer' traits in society seem to move in circles. The good creates the seed of the bad, and the bad creates the seed of the good. Disaster to heroism to prosperity to complacency to disaster. Technocracy to democracy to chaos to strongman to technocracy. And so on. The only thing that doesn't behave that way is economic output: growth just keeps compounding on itself. And so we should stop trying to accelerate the thing that goes in circles, and instead just focus on growing economic output. I dislike that argument. It makes me uncomfortable what it implies. But I still can't quite put my finger on what's wrong with it. What do you think?"
"I don't think it's just economic output that compounds on itself, I think many good things compound on each other. Health makes people more productive. Productivity gives people the resources to be better. Education supports both, and health supports education."
"So in other words, the eigenvector that Ephelion talked about includes more than just one single dimension - which is actually what most eigenvectors are like."
"You clearly understand eigenvectors better than I do!"
"Silverchat is a social media website. Eigenvectors are like ... on the first page of the original whitepaper."
Gladias smiled.
"But on the other hand", Evelor continued, "the Arctics do care about health. It's not health with the aesthetic of compassion and love and care that Sylka would want, it's hard metallic longevity clinics, but it's still health."
"Well Sylka would say it's not having a long life that matters, it's having a healthy life."
"Right, but the medieval tales of people drinking eternal life elixirs and then being doomed to lie on the ground forever with all their bones broken are not how any of the longevity science actually works. A longer life is a healthier life by default. The easiest place to stop damage is at the moment when it begins."
"And I guess you'll say the Arctics care about education too. They barely understand music and poetry, but they teach their kids science just fine, and that's the part of education that goes into the largest eigenvector anyway."
"Right."
Gladias could see Seila frowning with discomfort. His mind raced as he tried to come up with other arguments against Ephelion's way of thinking.
"But the Arctics aren't actually maximizing economic output. They're looking to start a war - wars can destroy economies for decades. Their attempt to eliminate anyone in Dzego who goes too far in advancing science and technology, that's not maximizing economic output either - quite the opposite. They're not trying to maximize economic output. They're trying to maximize the economic output that's under their control, and destroy everything else."
At that point, Zei spoke up.
"There is this underlying model that Ephelion is implying, where there's a vector that's shooting outward toward infinity, and the faster it goes toward infinity the faster everyone is happy. But that's not how I see the world at all."
"The world that I am seeing is a fragile world. It did not use to be this fragile. In the past, communication was limited and transmission of authority was limited. Any empire united for long enough would eventually fall. The world was reversible. But today, with our modern robots and modern surveillance, if anyone manages to take over and consolidate control across the whole world, they could lock their power in. Irreversibly. Forever."
"If the Arctics win, a certain kind of entropy - the dimensionality of the space of humanity's future possibilities - would be permanently constricted. From that point forward, the only possible future worlds would be Arctic worlds."
"And that's not the worst that can happen. If the Arctics, or anyone else, make a mistake with one of their superweapons, and no one else has the means to resist it, they could destroy the living world entirely."
"Eventually, the world will become reversible again. We will be able to travel between the stars, and our civilization will become too big to completely control or completely destroy. But until that happens, we are in a fragile place, and if within the next decade it all gets multiplied by zero forever, the Arctics will be the reason why it happens."
"I have a simpler answer", Seila added. "All these arguments about the far future, they're hard to predict. And it's really easy to convince yourself that one path is right, and come up with really smart math to justify it, if that's the path that you already know you want to take. But there is one thing that's easy to predict: if the Arctics conquer Veridia, the war may kill hundreds of thousands or millions of our people, and the Arctics will destroy the beautiful world that we know and cherish. To me, that's already enough to be worth fighting against. And the beauty of our world and our countries and our friendships, that's already enough to be worth fighting for."
"All good answers", Evelor replied. He took a sip of his tea and sat back in his chair, pondering.
3724 Frostime 8
Gladias, Seila, Zei, Bai, Verdow, Den and Mu were all sitting around a table inside a meeting room in the government-run hotel. Projected on a wall in front of them was a timer, counting down the ticks until the questionnaire on Silverchat would get published. At that moment, the remaining time was 150 ticks.
"Are you all done submitting your information into the package for the bots?", Seila asked.
"Just finished my part", Den replied.
"Remember, it's a bot. It doesn't need any amazing presentation and polish. If you think of something else to say or dump in, just throw it in."
"I hope you and your bots reviewed the cryptography well, Zei. I'm entrusting a lot of our military secrets to it."
"There's a machine-checkable mathematical proof that the garbled circuit algorithm satisfies the standard correctness and privacy properties. The proof is end-to-end, it covers everything all the way down to the operating system and even the hardware. I read the statements it proves, and I asked Emerald and two other AIs to check it. It's quite solid. Zero information leakage unless hashes are broken, and if that happens we have much bigger problems."
"I checked too", Mu interjected. "I used a different approach, just in case the proof system or the definitions have somehow been broken for over a decade and nobody noticed. I had Emerald read the original academic papers and write its own implementation, and then had a fresh instance read through that code and the original side by side, and look at every difference, see which side is right. Evelor's code is flawless."
"It's starting in fifty ticks!"
"Not going to imitate the Minpentai announcer voice this time?", Bai joked in response.
"MU GU GEI HUI ZIU FA", Zei replied in a voice exaggerating the stereotypical robotic tone for comic effect.
They sat around the table and waited.
"LE MU GEI HUI ZIU FA", Bai exclaimed.
"Hey!", Zei replied.
"Beat you!"
They kept watching the countdown. As soon as it hit ten, they started chanting in unison:
"PA GU ... SO ... BI ... ZE ... HA ... MU ... FO ... SHI ... LE ... PA ... HUI ZIU FA!"
The screen split into two views. On the left, there were five parallel progress bars showing the computation of five garbled-circuit executions with the five bots. On the right, a large collection of pie charts, one for each question that had been asked. Votes quickly began coming in.
"Let's zoom in to the ones we care about", Verdow offered.
Zei nodded, and clicked on one of the pie charts.
"As I expected. Now let's see how people think we should solve this problem."
"Restrict to people who are from Northshore."
"Wait, did we set it up to provide that level of granularity? Votes are anonymous, we have no way to tell who sent each one."
"We did break it down by age, self-declared most recent senator votes, province, and a couple of personality type axes inferred by a large language model."
"Ah, found it."
"Congrats, Evelor", Verdow said wryly. "Seems like you are more fit to run Northshore than the actual parliament."
"I'm a bit scared of these results now", Seila interjected. "Haven't we just given the game away to the Arctics?"
"I think it's okay", Mu replied. "We asked many different questions in the poll. One of them literally has twenty percent saying that we should fund recovering Northshore by legally transferring all of its government-owned territory to Hydrafill and letting them figure it out."
"The lawsuit is something that the Arctics will notice and watch, of course. We should delay it by a few days to make it not look too coordinated. But regardless of what we do, we will have made it clear to everyone - including the Arctics - that Veridians recognize that there is a big problem. It's the details of the solution that we need to hide."
They paused briefly, and stared at the progress bars.
"Okay, that will take a while. Let's look at some more pie charts."
A robot entered the room, bringing each of the participants a cup of tea.
"And ... done", Zei exclaimed triumphantly.
Five icons representing the responses from five bots appeared onscreen. Zei clicked on the first one.
Executive summary:
- Under any 'ordinary' scenario, meaning either status quo continuation of the Veridian military, or a reorganization led by the Veridian parliament without large outside pressure, the probability that Veridia will hold the same or greater percent of its own territory after one year is 20%.
- Under a hypothetical transfer to Kungaupei control, the probability increases to 26%.
- The main downside risks involve standard organizational transition costs that are typical of chaotic changes in leadership and replacement of experienced personnel with outsiders.
- Probability of successful outcomes is maximized (up to 50%) in cases that involve Kungaupei participation but minimize the disruptiveness of the transition.
- Note that above probabilities are for same or greater control in one year. Probabilities of strictly greater control are uniformly much lower.
"Hmm", Verdow muttered.
"Do you think it's reasonable?", Gladias asked.
"The bots definitely have a more conservative temperament than I do, but it's fair."
"The other four are saying vaguely similar things. The most pessimistic one says the chances are the same in the status quo and Kungaupei scenarios."
"Also we have much more data coming in from the votes. It's roughly the same as the results we first saw right when the poll began."
"I guess we have all the data we will get. Let's get ready to file the case in a few days."