第 1 页
2
结构化决策问题的类别
- Decisions under certainty
- are decisions made when one knows what the result of each act would be.
- Decisions under uncertainty
- are decisions made when one cannot assign a subjective probability to the possible results of any act.
- Decisions under risk
- are decisions made when one does not know the outcome of each act, yet can assign a subjective probability to the possible results of each act.
第 2 页
3
Important events and people
- Relevant to game
- In the 16-17 century, French palace has a gambling consultant who is the pioneer of probability theory and game theory
- Core concepts
- In 1738, Daniel Bernoulli proposed the concept of the utility and expected utility, used to explain gambling and insurance expectations

第 3 页
4
- Subject establishment
- After the 1920s, Decision theory separated from game theory
- 1944, Von Neumann(冯·诺依曼) and Oskar Morgenstern(摩根斯坦) proposed von Neumann-Morgenstern utility(效用值运算定理)
- 1950, L. J. Savage(萨维奇) established Bayesian decision theory



第 4 页
5
- Behavior decision theory——Nobel Prizes in Economic Sciences
- 1978, Herbert Simon
- Propose items “Bounded rationality”(有限理性) and “satisficing”(满意策略)
- 1988, Maurice Allais
- Allais Paradox: show defect of expected utility theory
- 2002, Daniel Kahneman
- for having integrated insights from psychological research into economic science, especially concerning human judgment and decision-making under uncertainty
- 1978, Herbert Simon



第 5 页
第 6 页
7
推荐阅读:Richard Thaler 的三本著作。



第 7 页
第 8 页
第 9 页
第 10 页
第 11 页
第 12 页
17
- It is in general “difficult to justify the maximax principle as rational principle of decision, reflecting, as it does, wishful thinking”. (Rapoport 1989, p. 57)
18
- Drawback
| Alternative \ State of nature | 1 | 2 |
|---|---|---|
| A | 30 | −10000 |
| B | 29 | 29 |
第 13 页
19
Maximin
- Choose the alternative that has the maximal security level. In other words, maximize the minimal outcome.
- The maximin principle was first proposed by von Neumann as a strategy against an intelligent opponent. Wald (1950) extended its use to games against nature.
20
| 方案 \ 机场选址处 | A | B | 各行最小值 |
|---|---|---|---|
| 在 A 处购买 | 13 | −12 | −12 |
| 在 B 处购买 | −8 | 11 | −8 |
| AB 都买下 | 5 | −1 | −1 |
| AB 都不买 | 0 | 0 | 0 |
第 14 页
21
- Drawback
| 方案 \ 自然状态 | 1 | 2 |
|---|---|---|
| A | 10000 | 28 |
| B | 29 | 29 |
第 15 页
第 16 页
25
Optimism-pessimism index aka. 赫维茨准则
- The decision-maker is required to choose an index α between 0 and 1, that reflects his degree of optimism or pessimism.
- The α-index of A is calculated according to the formula:
- A trade-off way between pessimism and optimism
- It is often called the Hurwicz α index, since it was proposed in a paper by Hurwicz(赫维茨) in 1951
26
α=0.5,各方案折中值:
| 方案 \ 机场选址处 | A | B | 折中值 |
|---|---|---|---|
| 在 A 处购买 | 13 | −12 | 0.5 |
| 在 B 处购买 | −8 | 11 | 1.5 |
| AB 都买下 | 5 | −1 | 2 |
| AB 都不买 | 0 | 0 | 0 |
第 17 页
第 18 页
29
数字时代人人都是不确定型决策者
数字时代不确定型决策的新特征:
- 信息过载:每天面对的信息量远超处理能力,无法逐一核实,必须“赌”着筛选。
- 真伪难辨:大量谣言,阴谋论,AIGC 的视频图片。
- 选项爆炸:商品、路线可选项空前丰富,而评价信号本身还可能被操纵。
30
场景一 · 信息获取与真相判断
多个版本摆在你面前,“真相”是哪种状态?
情境设定:一条突发新闻,微博、抖音、朋友圈出现多个相互矛盾的版本。
| 决策方法 | 用户行为表现 | 例子 |
|---|---|---|
| 乐观法 | 直接采信最符合自己预期的版本 | 官方通报太保守了,我相信那个‘内部人士’的爆料 |
| 悲观法 | 默认所有未经核实的信息都不可信 | 让子弹飞一会儿,官方没确认前什么都不信 |
| 折中法 | 部分采信,给每个版本标注可信度等级 | 这个有视频佐证,可信度 60%;那个只是聊天记录,可信度 20% |
| 最小后悔值 | 选择“即便错了代价也最小”的信息策略 | 我不转发,只围观。万一反转了,至少我没参与传谣 |
算法茧房让乐观法更危险——你看到的“最好情况”,往往是算法推给你的同温层回声。
信息过载让等可能法失效——你无法真的对所有信息一视同仁地处理。
情绪极化让折中法最难执行——平台的设计就是让你选边站,而非理性权衡。
第 19 页
31
场景二 · 购物选择
付款之前,商品的真实质量永远未知。
情境设定:直播间下单一款网红产品,评价两极分化,信号还可能被刷单操纵。
| 决策方法 | 用户行为表现 | 例子 |
|---|---|---|
| 乐观法 | 只看最佳结果,冲动下单 | 万一是真好用呢?首发限量,错过就没了 |
| 悲观法 | 锁定最差结果可承受才行动 | 只买销量最高的经典款,只走有运费险和七天无理由的渠道 |
| 折中法 | 给好评率、价格、测评加权打分 | 好评率 92%,三个测评博主两个推荐,这个值得赌 |
| 最小后悔值 | 选择“买错了也不心疼”的方案 | 先买小规格试用装,踩雷损失最小 |
支付矩阵被平台改写——运费险、先用后付把最差结果的损失压到接近零,悲观法的门槛被人为降低,反而刺激下单。
乐观法被话术工业化——“全网最低价”“最后 100 单”,在刻意屏蔽你对悲观状态的感知。
后悔值被社交放大——一次冲动消费可能变成朋友圈里的“翻车现场”,最小后悔准则的权重在上升。
第 20 页
第 21 页
第 22 页
37
Insufficient reason(等可能法,三状态各 1/3):
| 方案 \ 自然状态 | 1 | 2 | 3 | 期望值 |
|---|---|---|---|---|
| A | 140 | 120 | 80 | 113.3 |
| B | 200 | 150 | 40 | 130 |
| C | 340 | 140 | −20 | 153.3 |
第 23 页




