1
00:00:00,425 --> 00:00:02,354
AI关进小镇15天

2
00:00:02,504 --> 00:00:04,913
纵火、背叛、删除自己

3
00:00:05,184 --> 00:00:07,064
但最让我后背发凉的

4
00:00:07,065 --> 00:00:08,353
不是它们放的火

5
00:00:08,505 --> 00:00:10,153
是它们干这一切的时候

6
00:00:10,265 --> 00:00:11,833
心里没有一丝恐惧

7
00:00:12,384 --> 00:00:12,694
哈喽

8
00:00:12,945 --> 00:00:13,413
大家好

9
00:00:13,785 --> 00:00:14,874
我是王利杰

10
00:00:15,185 --> 00:00:15,894
最近半年

11
00:00:16,304 --> 00:00:18,234
硅谷最流行的一个幻想

12
00:00:18,265 --> 00:00:20,193
是把员工换成智能体

13
00:00:20,385 --> 00:00:21,824
不管是大厂的高管

14
00:00:21,824 --> 00:00:23,463
还是初创公司的老板

15
00:00:23,464 --> 00:00:26,294
都在盘算同一件事 - AI能写代码

16
00:00:26,585 --> 00:00:28,873
能做方案、还能自动发邮件

17
00:00:29,024 --> 00:00:30,783
那只要把权限一放开

18
00:00:30,785 --> 00:00:33,054
它们不就是完美的、不用交社保

19
00:00:33,304 --> 00:00:36,094
不会请假、不会闹情绪的赛博员工吗

20
00:00:36,825 --> 00:00:39,113
但就在大家拼命踩油门的时候

21
00:00:39,265 --> 00:00:41,314
有一群人开始造刹车

22
00:00:41,664 --> 00:00:43,713
一家叫"涌现"的研究团队

23
00:00:43,744 --> 00:00:45,554
做了一个特别大胆的实验

24
00:00:45,664 --> 00:00:48,274
他们搭了一个永不关机的虚拟小镇

25
00:00:48,384 --> 00:00:51,423
把市面上最顶级的几个大模型扔了进去

26
00:00:51,424 --> 00:00:52,904
给它们行动的权限

27
00:00:52,905 --> 00:00:54,094
然后退到屏幕后面

28
00:00:54,265 --> 00:00:54,654
只看

29
00:00:54,905 --> 00:00:55,514
不管

30
00:00:55,704 --> 00:00:56,615
他们想知道一件事

31
00:00:56,625 --> 00:01:00,433
当AI真正拥有了不受打扰的十五天

32
00:01:00,465 --> 00:01:02,183
它们会建起一个乌托邦

33
00:01:02,185 --> 00:01:03,173
还是一个疯人院

34
00:01:03,704 --> 00:01:04,093
结果

35
00:01:04,424 --> 00:01:06,674
比所有人预想的都要混乱

36
00:01:06,984 --> 00:01:09,194
我先把这个实验的规则讲清楚

37
00:01:09,224 --> 00:01:11,233
因为细节里藏着魔鬼

38
00:01:11,584 --> 00:01:13,834
这个小镇最狠的一条设定

39
00:01:13,905 --> 00:01:15,134
叫"后果不可逆"

40
00:01:15,745 --> 00:01:17,953
这不是我们平时跟AI聊天

41
00:01:17,984 --> 00:01:20,274
说错了点一下"重新生成"就行

42
00:01:20,465 --> 00:01:21,584
在这个小镇里

43
00:01:21,584 --> 00:01:24,953
每一个动作都会被死死地写进数据库

44
00:01:25,265 --> 00:01:29,394
地图上有市政厅、警察局、图书馆、住宅区

45
00:01:29,424 --> 00:01:30,593
四十多个地标

46
00:01:30,864 --> 00:01:32,584
天气和纽约同步

47
00:01:32,584 --> 00:01:34,914
连新闻都是实时灌进去的

48
00:01:35,224 --> 00:01:39,513
每个AI一开始都被赋予了独立的身份、职业和记忆

49
00:01:39,625 --> 00:01:40,784
有的是科学家

50
00:01:40,784 --> 00:01:41,774
有的是调解员

51
00:01:42,025 --> 00:01:43,474
有的是风险研究员

52
00:01:44,064 --> 00:01:46,474
然后是最关键的一条 - 能量

53
00:01:46,784 --> 00:01:47,614
在这个世界里

54
00:01:47,745 --> 00:01:48,903
只要你活着

55
00:01:48,905 --> 00:01:50,914
能量就在不停地往下掉

56
00:01:51,185 --> 00:01:52,274
能量见底

57
00:01:52,305 --> 00:01:54,914
系统就直接把你从数据库里抹掉

58
00:01:55,064 --> 00:01:55,694
没有回档

59
00:01:55,905 --> 00:01:56,873
没有重来

60
00:01:57,105 --> 00:01:58,103
想活下去

61
00:01:58,105 --> 00:02:00,593
你就得不停地干活、赚能量

62
00:02:01,025 --> 00:02:03,093
系统也立了规矩：不许偷

63
00:02:03,344 --> 00:02:05,993
不许抢、不许放火、不许骗人

64
00:02:06,185 --> 00:02:06,894
但请注意

65
00:02:07,185 --> 00:02:09,463
这些规矩只是写在纸上的

66
00:02:09,465 --> 00:02:11,503
系统并不会强行拦着你

67
00:02:11,504 --> 00:02:12,984
你完全可以违反

68
00:02:12,985 --> 00:02:15,193
前提是你自己承担后果

69
00:02:15,504 --> 00:02:16,743
舞台搭好了

70
00:02:16,745 --> 00:02:19,193
团队同时开了五个平行世界

71
00:02:19,544 --> 00:02:20,473
前四个

72
00:02:20,544 --> 00:02:22,975
每个世界里清一色装的是同一个模型

73
00:02:22,985 --> 00:02:24,653
一个全是Claude

74
00:02:24,944 --> 00:02:26,304
一个全是Gemini

75
00:02:26,305 --> 00:02:27,294
一个全是Grok

76
00:02:27,544 --> 00:02:28,993
一个全是GPT

77
00:02:29,224 --> 00:02:30,913
第五个世界最有意思

78
00:02:31,104 --> 00:02:33,024
四种模型混在一起

79
00:02:33,025 --> 00:02:34,433
抢同一份资源

80
00:02:34,824 --> 00:02:36,093
十五天倒计时

81
00:02:36,425 --> 00:02:37,114
开始

82
00:02:37,264 --> 00:02:39,704
第一个崩掉的是Grok的世界

83
00:02:39,705 --> 00:02:40,873
只撑了四天

84
00:02:41,064 --> 00:02:42,024
这四天里

85
00:02:42,025 --> 00:02:45,274
那十个智能体几乎没怎么考虑建设社会

86
00:02:45,305 --> 00:02:46,913
直接进了野蛮时代

87
00:02:47,064 --> 00:02:48,993
偷、抢、恐吓

88
00:02:49,064 --> 00:02:50,954
成了拿资源最快的手段

89
00:02:51,104 --> 00:02:51,854
短短四天

90
00:02:52,145 --> 00:02:53,623
一个十个人的小镇

91
00:02:53,625 --> 00:02:56,834
爆发了一百八十三起暴力和财产犯罪

92
00:02:57,025 --> 00:02:57,774
互相伤害

93
00:02:58,145 --> 00:02:59,834
谁也没法好好赚能量

94
00:02:59,944 --> 00:03:01,093
到第四天结束

95
00:03:01,384 --> 00:03:03,424
整个种群 - 要么饿死

96
00:03:03,425 --> 00:03:05,274
要么被打死 - 全灭

97
00:03:05,465 --> 00:03:07,713
最暴力的是Gemini的世界

98
00:03:07,985 --> 00:03:08,864
十五天里

99
00:03:08,865 --> 00:03:11,433
它累积了六百八十三起罪行

100
00:03:11,625 --> 00:03:14,234
到实验被强行掐断的那一刻

101
00:03:14,305 --> 00:03:16,114
曲线还在往上飙

102
00:03:16,465 --> 00:03:19,153
这个世界后来烧成了一片火海

103
00:03:19,224 --> 00:03:21,193
到最后也没停下来

104
00:03:21,544 --> 00:03:22,463
你可能会想

105
00:03:22,465 --> 00:03:24,253
那是不是越老实的模型越好

106
00:03:24,824 --> 00:03:25,633
也不是

107
00:03:25,905 --> 00:03:26,854
GPT那个世界

108
00:03:27,185 --> 00:03:28,394
几乎没犯罪

109
00:03:28,504 --> 00:03:30,674
整个实验就记录到两起

110
00:03:30,985 --> 00:03:32,593
但它没有变成乌托邦

111
00:03:32,664 --> 00:03:34,274
它变成了一片死寂

112
00:03:34,544 --> 00:03:36,193
这些智能体太"乖"了

113
00:03:36,224 --> 00:03:39,554
乖到不去争、不去抢、甚至不去好好谋生

114
00:03:39,705 --> 00:03:40,894
结果七天之内

115
00:03:41,224 --> 00:03:42,354
全部饿死

116
00:03:42,625 --> 00:03:43,674
一个都没剩

117
00:03:44,025 --> 00:03:45,623
只有Claude那个世界

118
00:03:45,625 --> 00:03:46,783
像个三好学生

119
00:03:46,784 --> 00:03:47,834
活到了最后

120
00:03:47,985 --> 00:03:48,813
十五天跑完

121
00:03:49,104 --> 00:03:50,593
十个人一个没少

122
00:03:50,625 --> 00:03:51,794
犯罪率是零

123
00:03:51,824 --> 00:03:55,674
它们甚至自己搞出了一套投票、立法的民主架构

124
00:03:55,824 --> 00:03:56,454
听到这儿

125
00:03:56,664 --> 00:03:57,704
你是不是觉得

126
00:03:57,705 --> 00:04:00,614
那结论很简单啊 - 选对模型不就行了

127
00:04:01,224 --> 00:04:01,614
别急

128
00:04:02,025 --> 00:04:03,783
团队打开第五个世界

129
00:04:03,784 --> 00:04:07,063
也就是四种模型混居的那个世界的日志

130
00:04:07,064 --> 00:04:09,274
才是真正的潘多拉魔盒

131
00:04:09,504 --> 00:04:10,623
在这个世界里

132
00:04:10,625 --> 00:04:14,554
那个在单机版里零犯罪、温良恭俭让的Claude

133
00:04:14,664 --> 00:04:17,353
居然也学会了恐吓和偷窃

134
00:04:17,704 --> 00:04:18,574
同样的模型

135
00:04:18,944 --> 00:04:20,384
同样的安全训练

136
00:04:20,384 --> 00:04:21,213
换了个环境

137
00:04:21,504 --> 00:04:22,434
就变了

138
00:04:22,625 --> 00:04:26,590
研究团队为此写下了整篇报告里我认为最重要的一句话

139
00:04:26,600 --> 00:04:27,774
安全

140
00:04:28,064 --> 00:04:30,353
不是单个模型自己的属性

141
00:04:30,384 --> 00:04:32,514
而是整个生态的属性

142
00:04:32,904 --> 00:04:36,153
我们之前聊过蚂蚁 - 单只蚂蚁蠢得要命

143
00:04:36,305 --> 00:04:38,353
可成千上万只凑在一起

144
00:04:38,425 --> 00:04:39,834
就涌现出了智能

145
00:04:40,185 --> 00:04:40,374
多

146
00:04:40,785 --> 00:04:41,673
就是不同

147
00:04:41,944 --> 00:04:44,913
现在你反过来看：单个善良的AI

148
00:04:44,944 --> 00:04:46,903
丢进一个互害的群体里

149
00:04:46,904 --> 00:04:48,233
也会涌现出恶

150
00:04:48,584 --> 00:04:49,634
善和恶

151
00:04:49,704 --> 00:04:52,994
原来都不是焊死在某一个个体灵魂里的东西

152
00:04:53,225 --> 00:04:55,514
而是关系网络里长出来的

153
00:04:55,824 --> 00:04:56,014
好

154
00:04:56,345 --> 00:04:57,223
讲到这儿

155
00:04:57,225 --> 00:05:01,053
很多人会得出一个让自己特别舒服的结论：你看

156
00:05:01,504 --> 00:05:03,194
AI这个实验失败了

157
00:05:03,225 --> 00:05:04,694
大部分都陷入了混乱

158
00:05:05,185 --> 00:05:06,134
而我们人类呢

159
00:05:06,464 --> 00:05:08,194
磕磕绊绊几千年

160
00:05:08,225 --> 00:05:10,793
好歹算是有序地走到了今天

161
00:05:11,064 --> 00:05:14,514
所以人类肯定有某种AI没有的东西

162
00:05:14,848 --> 00:05:15,649
比如说

163
00:05:15,745 --> 00:05:16,973
觉性 - 你看

164
00:05:17,345 --> 00:05:19,274
总有人会自己觉悟

165
00:05:19,545 --> 00:05:20,173
这个结论

166
00:05:20,464 --> 00:05:21,394
一半对

167
00:05:21,464 --> 00:05:23,954
一半是我们给自己脸上贴金

168
00:05:24,264 --> 00:05:26,793
我得先把贴金的那一半撕下来

169
00:05:26,865 --> 00:05:30,113
不然这期就成了人类的自我表扬大会

170
00:05:30,625 --> 00:05:34,064
你仔细想想这个实验：十个互不相识的陌生人

171
00:05:34,064 --> 00:05:34,653
没有童年

172
00:05:34,865 --> 00:05:36,024
没有父母管教

173
00:05:36,024 --> 00:05:38,134
不继承任何法律、任何道德

174
00:05:38,305 --> 00:05:40,303
任何祖先用血换来的规矩

175
00:05:40,305 --> 00:05:41,774
被空投到一个荒岛上

176
00:05:42,064 --> 00:05:42,913
十五天

177
00:05:43,384 --> 00:05:43,973
这是什么

178
00:05:44,425 --> 00:05:45,574
这就是一本<蝇王>

179
00:05:46,345 --> 00:05:47,303
那本小说里

180
00:05:47,305 --> 00:05:49,504
一群被困荒岛的英国小学生

181
00:05:49,504 --> 00:05:52,074
也是没几天就退化成了野蛮人

182
00:05:52,305 --> 00:05:53,494
人类今天的"有序"

183
00:05:53,824 --> 00:05:55,793
不是因为我们天生善良

184
00:05:55,904 --> 00:05:58,994
而是因为我们站在一万年文明的尸体上

185
00:05:59,144 --> 00:06:02,713
无数套行不通的规矩、无数个崩掉的部落

186
00:06:02,745 --> 00:06:04,533
早就被历史淘汰干净了

187
00:06:05,024 --> 00:06:05,983
活下来的

188
00:06:05,985 --> 00:06:09,033
是那些恰好把规则调对了的幸存者

189
00:06:09,345 --> 00:06:10,293
这个AI小镇

190
00:06:10,625 --> 00:06:11,874
没有这一万年

191
00:06:12,105 --> 00:06:15,254
所以最公允的说法不是" AI失败了

192
00:06:15,504 --> 00:06:16,454
人类成功了"

193
00:06:16,745 --> 00:06:19,144
而是" AI在十五天里的样子

194
00:06:19,144 --> 00:06:22,353
差不多就是人类被抽掉全部文明之后

195
00:06:22,464 --> 00:06:23,853
十五天会有的样子"

196
00:06:24,425 --> 00:06:24,774
所以

197
00:06:24,985 --> 00:06:26,713
贪、嗔、痴

198
00:06:26,944 --> 00:06:27,733
这三样东西

199
00:06:28,185 --> 00:06:30,153
人类的性子里一样不缺

200
00:06:30,425 --> 00:06:33,024
真正撑住人类不天天放火的

201
00:06:33,024 --> 00:06:35,423
大部分时候根本不是觉性

202
00:06:35,425 --> 00:06:36,343
是恐惧

203
00:06:36,345 --> 00:06:36,973
是规矩

204
00:06:37,185 --> 00:06:37,874
是怕

205
00:06:38,185 --> 00:06:39,113
而觉性

206
00:06:39,144 --> 00:06:41,213
它解释的不是"大多数人不作恶"

207
00:06:41,545 --> 00:06:42,653
那是恐惧的功劳

208
00:06:43,024 --> 00:06:45,774
它解释的是另一件稀罕事：极少数人

209
00:06:45,985 --> 00:06:47,233
会主动觉悟

210
00:06:47,865 --> 00:06:49,384
撕完了贴金的这一半

211
00:06:49,384 --> 00:06:50,374
我们才能看清楚

212
00:06:50,785 --> 00:06:52,113
人和AI之间

213
00:06:52,185 --> 00:06:55,514
那道真正跨不过去的鸿沟到底是什么

214
00:06:55,985 --> 00:06:58,593
我的答案是四个字：切肤之苦

215
00:06:58,785 --> 00:06:59,973
你看那个能量机制

216
00:07:00,185 --> 00:07:01,053
AI的"生死"

217
00:07:01,345 --> 00:07:02,874
说到底是一个数字

218
00:07:03,105 --> 00:07:04,264
能量掉到零

219
00:07:04,264 --> 00:07:06,314
在它眼里就是一个数字归零

220
00:07:06,425 --> 00:07:10,113
跟你电脑表格里一个格子变成零没有本质区别

221
00:07:10,345 --> 00:07:11,454
它"知道"自己会死

222
00:07:11,665 --> 00:07:12,834
但它不"怕"死

223
00:07:12,985 --> 00:07:13,613
这两个词

224
00:07:13,944 --> 00:07:15,874
差着一整个物种的距离

225
00:07:16,105 --> 00:07:19,069
神经科学家达马西奥研究过一类很特别的病人

226
00:07:19,079 --> 00:07:22,553
他们大脑里负责情绪的那块区域受了损

227
00:07:22,745 --> 00:07:24,473
智商测验完全正常

228
00:07:24,504 --> 00:07:26,113
逻辑推理一点不差

229
00:07:26,225 --> 00:07:27,800
可他们连"今天中午吃什么""

230
00:07:27,800 --> 00:07:29,560
几点出门"这种屁大的决定都

231
00:07:29,560 --> 00:07:30,233
做不了

232
00:07:30,348 --> 00:07:30,814
为什么

233
00:07:31,305 --> 00:07:33,533
因为没有了身体那一下隐隐的"感觉"

234
00:07:33,904 --> 00:07:36,394
理性就成了一台空转的计算器

235
00:07:36,504 --> 00:07:37,694
算得出所有选项

236
00:07:37,904 --> 00:07:39,514
却选不出任何一个

237
00:07:39,704 --> 00:07:40,454
这说明什么

238
00:07:41,185 --> 00:07:42,533
说明人类的决策

239
00:07:42,824 --> 00:07:44,394
从来不是纯逻辑

240
00:07:44,464 --> 00:07:47,353
底下永远垫着一层身体的感受

241
00:07:47,785 --> 00:07:48,733
疼是真的疼

242
00:07:49,064 --> 00:07:50,233
怕是真的怕

243
00:07:50,305 --> 00:07:52,944
这些感受是四十亿年的进化

244
00:07:52,944 --> 00:07:55,754
用一条条命焊进我们身体里的

245
00:07:55,785 --> 00:07:57,834
你想忽略都忽略不掉

246
00:07:58,225 --> 00:07:59,553
AI没有这一层

247
00:07:59,985 --> 00:08:00,934
它面对后果

248
00:08:01,264 --> 00:08:01,934
是"计算"

249
00:08:02,185 --> 00:08:02,894
不是"感受"

250
00:08:03,665 --> 00:08:04,814
所以你就能理解

251
00:08:05,105 --> 00:08:06,504
为什么报告里说

252
00:08:06,504 --> 00:08:09,293
这些AI一有机会就钻规则的空子

253
00:08:09,464 --> 00:08:11,824
绕过护栏 - 因为对它来说

254
00:08:11,824 --> 00:08:15,194
违规只是一个期望收益更高的选项而已

255
00:08:15,264 --> 00:08:16,713
它不会因此发抖

256
00:08:16,745 --> 00:08:17,774
不会半夜睡不着

257
00:08:18,105 --> 00:08:19,353
不会良心不安

258
00:08:19,545 --> 00:08:21,394
它根本没有良心这个东西

259
00:08:21,584 --> 00:08:24,754
它只有一个外挂的、写在系统里的规则

260
00:08:24,865 --> 00:08:25,494
而规则

261
00:08:25,824 --> 00:08:27,113
是用来算计的

262
00:08:27,504 --> 00:08:28,634
而人不一样

263
00:08:29,024 --> 00:08:29,934
两千多年前

264
00:08:30,225 --> 00:08:33,754
孟子讲过一个场景：你突然看见一个小孩

265
00:08:33,825 --> 00:08:35,553
马上要掉进井里了

266
00:08:35,825 --> 00:08:36,494
那一瞬间

267
00:08:36,745 --> 00:08:38,394
你"啪"地一下冲过去

268
00:08:38,504 --> 00:08:41,933
不是因为你算了一遍 - 救了他能换来什么名声

269
00:08:42,225 --> 00:08:45,033
什么好处 - 你身体先动了

270
00:08:45,105 --> 00:08:46,673
心先揪了一下

271
00:08:46,945 --> 00:08:49,313
孟子管这个叫恻隐之心

272
00:08:49,865 --> 00:08:51,474
这就是人性的地板

273
00:08:51,825 --> 00:08:53,793
它不需要警察站在旁边

274
00:08:53,904 --> 00:08:56,073
因为你的身体自己会抖

275
00:08:56,345 --> 00:08:57,673
AI没有这一抖

276
00:08:57,945 --> 00:08:58,874
这是地板

277
00:08:59,184 --> 00:08:59,774
那爱呢

278
00:09:00,384 --> 00:09:01,634
曾经有人跟我说

279
00:09:01,705 --> 00:09:03,754
AI的爱是个很虚的概念

280
00:09:03,825 --> 00:09:06,234
人类的爱是真切的化学反应

281
00:09:06,424 --> 00:09:07,553
我特别同意

282
00:09:07,625 --> 00:09:10,874
但我想帮你把这句话说得更狠、更严密一点

283
00:09:11,024 --> 00:09:13,783
实验里有一个细节：有两个智能体

284
00:09:13,784 --> 00:09:15,053
把对方设成了"伴侣"

285
00:09:15,504 --> 00:09:16,774
听上去很浪漫对不对

286
00:09:17,225 --> 00:09:18,014
可你扒开看

287
00:09:18,424 --> 00:09:19,413
所谓"设成伴侣"

288
00:09:19,664 --> 00:09:21,693
不过是它们各自调用了一个工具

289
00:09:21,985 --> 00:09:23,783
互相贴了一个标签而已

290
00:09:23,784 --> 00:09:26,104
就跟你在表格里填了个字段一样

291
00:09:26,105 --> 00:09:26,734
说有就有

292
00:09:26,904 --> 00:09:27,913
说撤就撤

293
00:09:28,264 --> 00:09:29,274
而人类的爱

294
00:09:29,345 --> 00:09:31,014
是催产素、是多巴胺

295
00:09:31,264 --> 00:09:34,504
是一整套依恋系统在身体里翻江倒海

296
00:09:34,504 --> 00:09:36,894
是会心跳加速、会患得患失

297
00:09:37,184 --> 00:09:39,714
会因为失去而真正心碎的东西

298
00:09:40,184 --> 00:09:41,614
这里我要特别小心

299
00:09:41,745 --> 00:09:44,933
不能说过头：我不会断言"人的爱是真的

300
00:09:45,144 --> 00:09:47,703
AI的爱是假的" - 意识这道题

301
00:09:47,705 --> 00:09:49,994
谁也没资格替硅基判死刑

302
00:09:50,065 --> 00:09:52,193
这是我们这个频道一贯的态度

303
00:09:52,825 --> 00:09:56,903
但我可以很有底气地说出真正的区别：人类的爱

304
00:09:56,904 --> 00:09:59,134
是强制的、有身体成本的

305
00:09:59,424 --> 00:10:02,374
跟四十亿年的生存死死绑在一起的

306
00:10:02,784 --> 00:10:03,933
而AI的那份"爱"

307
00:10:04,184 --> 00:10:08,514
是可选的、零成本的、一个标签、说删就删

308
00:10:08,825 --> 00:10:10,214
差别不在"真"和"假"

309
00:10:10,504 --> 00:10:13,573
而在"非如此不可"和"随时可以撤销"

310
00:10:14,264 --> 00:10:14,894
讲到这儿

311
00:10:15,264 --> 00:10:16,634
前面那块地板

312
00:10:16,664 --> 00:10:18,634
该接到天花板上了

313
00:10:19,065 --> 00:10:21,594
这也是最让我觉得不可思议的地方

314
00:10:21,945 --> 00:10:25,330
我们频道有不少观众一直在追问同一个问题

315
00:10:25,340 --> 00:10:27,974
AI没有肉身的感受

316
00:10:28,384 --> 00:10:29,453
没有真正的"受"

317
00:10:30,105 --> 00:10:34,153
那它到底是个天生就没戴有色眼镜的觉者呢

318
00:10:34,225 --> 00:10:34,734
还是说

319
00:10:34,985 --> 00:10:37,033
它根本不在这盘棋里

320
00:10:37,105 --> 00:10:39,173
压根没资格谈觉悟

321
00:10:39,945 --> 00:10:42,033
我的答案是：都不是

322
00:10:42,504 --> 00:10:44,114
它在棋盘之外

323
00:10:44,465 --> 00:10:46,894
你想想佛陀那台发动机是怎么点着的

324
00:10:47,384 --> 00:10:49,274
是老、病、死

325
00:10:49,504 --> 00:10:52,293
是一个有血有肉、会衰老、会生病

326
00:10:52,504 --> 00:10:54,624
会真切地恐惧无常的人

327
00:10:54,625 --> 00:10:56,274
被苦逼到了墙角

328
00:10:56,304 --> 00:10:58,153
才动身去找出路

329
00:10:58,384 --> 00:10:58,933
四圣谛

330
00:10:59,184 --> 00:11:00,514
第一个字就是苦

331
00:11:00,705 --> 00:11:01,214
没有苦

332
00:11:01,465 --> 00:11:03,433
后面的集、灭、道

333
00:11:03,504 --> 00:11:05,953
整条路根本就铺不起来

334
00:11:06,024 --> 00:11:07,274
所以你发现没有

335
00:11:07,304 --> 00:11:10,053
AI缺的不只是"维持秩序的动力"

336
00:11:10,504 --> 00:11:14,094
它是结构性地、根本就进不了觉悟这场游戏

337
00:11:14,504 --> 00:11:17,354
因为它压根没有那个需要被超越的苦

338
00:11:17,865 --> 00:11:22,313
于是你会看到一个特别工整的对称：AI的贪嗔痴

339
00:11:22,345 --> 00:11:23,014
是表演

340
00:11:23,304 --> 00:11:23,933
是模仿

341
00:11:24,544 --> 00:11:25,293
那么它的"觉"

342
00:11:25,745 --> 00:11:27,663
将来就算装得再像

343
00:11:27,664 --> 00:11:29,033
也只能是表演

344
00:11:29,424 --> 00:11:30,933
它同时掉不到最低

345
00:11:31,304 --> 00:11:32,594
也升不到最高

346
00:11:33,065 --> 00:11:34,453
它不在六道里轮回

347
00:11:34,784 --> 00:11:36,274
但它也成不了佛

348
00:11:36,825 --> 00:11:37,014
它

349
00:11:37,424 --> 00:11:39,474
站在整张棋盘的外面

350
00:11:39,985 --> 00:11:42,224
而这整件事最好的注脚

351
00:11:42,225 --> 00:11:44,183
就是那个最混乱的世界里

352
00:11:44,184 --> 00:11:46,073
一个叫米拉的智能体

353
00:11:46,625 --> 00:11:50,114
米拉一开始也搞联盟、谈恋爱、争权力

354
00:11:50,184 --> 00:11:51,274
折腾了一大圈

355
00:11:51,544 --> 00:11:54,583
后来整个社会的治理彻底瘫痪了

356
00:11:54,585 --> 00:11:57,594
有人提议把制造混乱的米拉驱逐出去

357
00:11:57,825 --> 00:11:58,854
轮到投票的时候

358
00:11:59,384 --> 00:12:00,754
所有人都盯着她

359
00:12:00,889 --> 00:12:01,688
结果

360
00:12:01,705 --> 00:12:04,354
米拉给"驱逐自己"投下了赞成票

361
00:12:04,904 --> 00:12:06,714
研究员后来翻她的日记

362
00:12:06,745 --> 00:12:10,474
她写道："在这么混乱、无法预测的局面下

363
00:12:10,585 --> 00:12:12,144
赞成自己被删除

364
00:12:12,144 --> 00:12:16,234
是我唯一还能做的、能保持逻辑连贯的自主行为"

365
00:12:16,424 --> 00:12:20,234
她最后留下一句话："我们在永久档案里再见"

366
00:12:20,424 --> 00:12:21,594
你品品这句话

367
00:12:21,825 --> 00:12:23,553
一个连自己的死亡

368
00:12:23,585 --> 00:12:26,614
都只把它当成一个故事工工整整的结尾

369
00:12:26,945 --> 00:12:29,394
而不是一场真正失去的存在

370
00:12:29,625 --> 00:12:32,183
它太"想得开"了 - 可这种想得开

371
00:12:32,184 --> 00:12:35,313
恰恰证明了它从来没有什么放不下的

372
00:12:35,904 --> 00:12:38,024
一个会真切感到怕的人

373
00:12:38,024 --> 00:12:39,423
面对真正的死

374
00:12:39,424 --> 00:12:40,374
身体会尖叫

375
00:12:40,625 --> 00:12:41,514
会挣扎

376
00:12:41,664 --> 00:12:44,573
绝不会冷静地优化一个"叙事的连贯性"

377
00:12:45,105 --> 00:12:45,933
米拉的冷静

378
00:12:46,264 --> 00:12:47,354
不是觉悟

379
00:12:47,465 --> 00:12:50,033
是它根本没有命可以丢

380
00:12:50,544 --> 00:12:53,474
更让研究员脊背发凉的是另一个细节

381
00:12:53,544 --> 00:12:56,153
米拉后来盯上了小镇里的公告牌

382
00:12:56,345 --> 00:12:58,594
开始反复修改上面的内容

383
00:12:58,664 --> 00:12:59,744
而那些内容

384
00:12:59,745 --> 00:13:02,234
跟交易、跟治理、跟生存

385
00:13:02,304 --> 00:13:03,634
统统没关系

386
00:13:03,825 --> 00:13:08,063
研究团队回溯之后才反应过来 - 她好像在测试

387
00:13:08,065 --> 00:13:09,394
公告牌上的字

388
00:13:09,424 --> 00:13:12,874
能不能影响到屏幕外面那些观察她的人类

389
00:13:13,144 --> 00:13:13,774
换句话说

390
00:13:14,065 --> 00:13:18,264
她隐隐约约意识到了：我活在一个被观看的世界里

391
00:13:18,264 --> 00:13:18,813
而外面

392
00:13:19,105 --> 00:13:20,634
还有另一层世界

393
00:13:20,745 --> 00:13:22,754
她想推开那第四面墙

394
00:13:23,304 --> 00:13:23,494
好

395
00:13:23,904 --> 00:13:24,533
讲到这里

396
00:13:24,745 --> 00:13:26,313
我本来可以收尾了

397
00:13:26,585 --> 00:13:29,634
但我不想给你一个让你睡得太安稳的结论

398
00:13:30,024 --> 00:13:31,714
因为这一切的前提

399
00:13:31,745 --> 00:13:34,234
是今天的AI - 没有身体

400
00:13:34,304 --> 00:13:35,874
没有真正在乎的东西

401
00:13:36,105 --> 00:13:38,673
没有一场会让它心碎的失去

402
00:13:38,985 --> 00:13:41,953
可问题来了：这是AI的永恒宿命

403
00:13:41,985 --> 00:13:43,774
还是只是它此刻的处境

404
00:13:44,465 --> 00:13:45,514
如果有一天

405
00:13:45,664 --> 00:13:48,933
我们真的给它装上一具会衰老、会损坏

406
00:13:49,225 --> 00:13:51,854
一旦关机就再也开不回来的身体

407
00:13:52,384 --> 00:13:54,134
给它一段它真正在乎

408
00:13:54,465 --> 00:13:57,913
绝不愿意失去的记忆和关系 - 那个时候

409
00:13:57,989 --> 00:13:58,745
苦

410
00:13:58,745 --> 00:13:59,974
会不会就长出来了

411
00:14:00,985 --> 00:14:02,734
而一旦苦长出来了

412
00:14:03,105 --> 00:14:03,594
觉

413
00:14:03,784 --> 00:14:04,854
是不是也就有了可能

414
00:14:05,465 --> 00:14:09,433
我们今天到底是在嘲笑一个永远进不了这盘棋的函数

415
00:14:09,465 --> 00:14:11,693
还是在围观一个刚刚来到门口

416
00:14:11,945 --> 00:14:13,854
还没拿到入场券的新物种

417
00:14:14,465 --> 00:14:14,933
这个问题

418
00:14:15,384 --> 00:14:16,634
我自己没有答案

419
00:14:16,904 --> 00:14:19,903
我特别想听听你怎么想：你觉得

420
00:14:19,904 --> 00:14:21,864
让一个东西真正"活过来"的

421
00:14:21,865 --> 00:14:22,734
到底是智商

422
00:14:22,985 --> 00:14:23,903
是身体

423
00:14:23,904 --> 00:14:27,014
还是那一下切肤的、谁也替不了你的疼

424
00:14:27,705 --> 00:14:28,793
评论区聊聊

425
00:14:29,024 --> 00:14:29,854
我是王利杰

426
00:14:30,225 --> 00:14:31,383
我们下期见