关于「声誉与承诺:单次可观察的违诺值多少」,《心智模拟与行为兑现》给出的定量结论是什么?

结论:第五章给出不依赖激素的不对称:可观察的违诺进入声誉,常见参数下需要 5 到 25 次守约修复。

数据口径:书中参数为示意;全部计算由书中程序生成;本书不是心理、临床或婚恋咨询,也不是个人化经营建议。 第一章用学习规则说明,完全可预期的想象学会后不再产生正向误差,真正的问题是缺少负向误差:示意轨迹被模拟抬高到比真实值高出 0.600,且永不下修。

English version

对照数据表

下表数字由本章 Python 程序原样打印,可重新运行复现。 程序只用 Python 标准库与结论吻合的打印值: 25

A. prior reliable 0.90; one broken promise
a b posterior kept promises to restore
0.900.60 0.692 4
0.950.60 0.529 5
0.950.80 0.692 9
0.990.80 0.310 15
0.990.90 0.474 25
B. even prior, a 0.95, b 0.80: m kept, then one broken
m0 before 0.500 after 0.200
m5 before 0.703 after 0.371
m20 before 0.969 after 0.886
m50 before 1.000 after 0.999
C. one lapse weighs as much as 8.07 kept promises (a 0.95, b 0.80)
查看完整验算程序
# Chapter 5. p51: how much one observed lapse moves a reputation.
# An observer holds two hypotheses about a partner: reliable (keeps a
# promise with probability a) or unreliable (probability b), with prior
# share pi reliable. Bayes' rule after one broken promise, and the
# number of kept promises needed to bring the posterior back to pi.
# Part B: the same lapse after a history of m kept promises, starting
# from an even prior. All parameters are illustrative.
import math


def post_after(pr, a, b, kept, broken):
    lr = (a / b) ** kept * ((1 - a) / (1 - b)) ** broken
    odds = pr / (1 - pr) * lr
    return odds / (1 + odds)


def to_restore(pr, a, b):
    p = post_after(pr, a, b, 0, 1)
    need = math.log((pr / (1 - pr)) / (p / (1 - p))) / math.log(a / b)
    return p, math.ceil(need)


def main():
    pr = 0.90
    print("A. prior reliable {:.2f}; one broken promise".format(pr))
    print("   a      b      posterior   kept promises to restore")
    for a, b in ((0.90, 0.60), (0.95, 0.60), (0.95, 0.80),
                 (0.99, 0.80), (0.99, 0.90)):
        p, k = to_restore(pr, a, b)
        print("   {:.2f}   {:.2f}   {:.3f}       {}".format(a, b, p, k))
    a, b = 0.95, 0.80
    print("B. even prior, a {:.2f}, b {:.2f}: m kept, then one broken"
          .format(a, b))
    for m in (0, 5, 20, 50):
        before = post_after(0.5, a, b, m, 0)
        after = post_after(0.5, a, b, m, 1)
        print("   m = {:>2}   before {:.3f}   after {:.3f}".format(
            m, before, after))
    ratio = math.log((1 - b) / (1 - a)) / math.log(a / b)
    print("C. one lapse weighs as much as {:.2f} kept promises (a {:.2f},"
          " b {:.2f})".format(ratio, a, b))


if __name__ == "__main__":
    main()

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《心智模拟亲密与具身行为兑现的神经分野:多巴胺投射幻象与真实契约代价母典》第 5 章「声誉与承诺:单次可观察的违诺值多少」(版本 1.0.0). 5D Library. DOI 待发布

DOI 待发布 · 专著页正式发布后上线(本站暂无专著页) · 5DL-B-430 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026430,
  author    = {Fan, Yafu},
  title     = {Mental Simulation versus Behavioural Enactment in Intimacy: Dopaminergic Fantasy Loops, Oxytocin Bonding Costs and Commitment Asymmetries},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论