关于「机制之争:内化、背谱与构成偏差」,《围棋 AI 后的人类决策质量》给出的定量结论是什么?

结论:第五章区分内化、背谱与构成偏差,说明棋手固定效应能把虚假交互项从 0.831 压到 -0.053,却对背谱无效。

数据口径:书中除论文数字外的参数均为示意,全部计算由书中程序生成;本书是非个人化经营建议,不针对任何个人决定。 第四章用香农熵度量定式集中度,示意角部局面的策略熵从 1.608 升到 2.360 比特,从众学习模型显示外部评估者让群体提早约 12 年离开局部最优。

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: -0.053, 0.831

scenariotrue b3pooled b3FE b3FE b1
S10.5+0.415+0.435+0.619
S20.0+0.831-0.053+0.634
S30.0+0.816+0.768+0.578
S3 memorised share x bonus = 0.4 x 1.5 = 0.60
查看完整验算程序
# Chapter 5. p51: what the after x novel interaction can and cannot
# tell apart
# Model (simulated panel, illustrative): 200 players, 150 moves each
# before and after the AI era; DQI = 97 + ability + b1 after + b2 novel
# + b3 after x novel + noise (sd 4); ability ~ N(0, 1). Scenario S1:
# b1 0.6, b2 -0.6, b3 0.5, novelty 0.10 before and 0.20 after for all.
# S2: b3 = 0, but after the AI era only strong players (ability > 0.5)
# play novel moves often (0.35 versus 0.05). S3: b3 = 0 for thinking,
# but 40 % of post-AI novel moves are memorised AI lines worth +1.5.
# Each scenario is fitted by pooled OLS and by player fixed effects
# (within transformation). Seeded; sums by math.fsum.
import math
import random

SEED = 20261112
PLAYERS = 200
MOVES = 150


def solve(a, b):
    n = len(b)
    m = [row[:] + [b[i]] for i, row in enumerate(a)]
    for c in range(n):
        p = max(range(c, n), key=lambda r: abs(m[r][c]))
        m[c], m[p] = m[p], m[c]
        piv = m[c][c]
        m[c] = [v / piv for v in m[c]]
        for r in range(n):
            if r != c:
                f = m[r][c]
                m[r] = [v - f * w for v, w in zip(m[r], m[c])]
    return [m[i][n] for i in range(n)]


def ols(xs, ys):
    k = len(xs[0])
    xtx = [[math.fsum(x[i] * x[j] for x in xs) for j in range(k)]
           for i in range(k)]
    xty = [math.fsum(x[i] * y for x, y in zip(xs, ys)) for i in range(k)]
    return solve(xtx, xty)


def panel(rng, scen):
    rows = []
    for i in range(PLAYERS):
        a = rng.gauss(0.0, 1.0)
        for after in (0, 1):
            if after and scen == "S2":
                q = 0.35 if a > 0.5 else 0.05
            else:
                q = 0.20 if after else 0.10
            for _ in range(MOVES):
                nv = 1 if rng.random() < q else 0
                b3 = 0.5 if scen == "S1" else 0.0
                parts = [97.0, a, 0.6 * after, -0.6 * nv,
                         b3 * after * nv, rng.gauss(0.0, 4.0)]
                if scen == "S3" and after and nv and rng.random() < 0.4:
                    parts.append(1.5)
                rows.append((i, after, nv, math.fsum(parts)))
    return rows


def fit(rows):
    pooled = ols([[1.0, a, n, a * n] for _, a, n, _ in rows],
                 [y for *_, y in rows])
    groups = {}
    for r in rows:
        groups.setdefault(r[0], []).append(r)
    xs, ys = [], []
    for i in sorted(groups):
        g = groups[i]
        cols = [[float(a), float(n), float(a * n), y] for _, a, n, y in g]
        mean = [math.fsum(c[j] for c in cols) / len(cols)
                for j in range(4)]
        for c in cols:
            xs.append([c[j] - mean[j] for j in range(3)])
            ys.append(c[3] - mean[3])
    fe = ols(xs, ys)
    return pooled[3], fe[2], fe[0]


if __name__ == "__main__":
    rng = random.Random(20261112)
    print("scenario  true b3  pooled b3   FE b3    FE b1")
    truth = {"S1": 0.5, "S2": 0.0, "S3": 0.0}
    for scen in ("S1", "S2", "S3"):
        p3, f3, f1 = fit(panel(rng, scen))
        print("  {}       {:.1f}     {:+.3f}     {:+.3f}   {:+.3f}".format(
            scen, truth[scen], p3, f3, f1))
    print("S3 memorised share x bonus = 0.4 x 1.5 = {:.2f}".format(0.6))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《围棋人工智能后人类决策质量跃迁、新颖着法率与棋经十三篇策略熵母典》第 5 章「机制之争:内化、背谱与构成偏差」(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-230 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026230,
  author    = {Fan, Yafu},
  title     = {Post-AI Leaps in Human Go Decision Quality, Novel-Move Rates and Strategic Entropy in the Thirteen-Chapter Classic of Go},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论