What does Human-AI Symbiosis and Parallel Agents conclude, in numbers, about “An Instruction as a Noisy Channel”?

Answer: Chapter 6 treats an instruction as a noisy channel: at a misreading rate of 0.15, checks covering 0.9 of decisions give 0.6355 at 2.06 uses per decision, better than restating five times (0.4452).

Scope of the numbers: Its agents, rates, hours and values are illustrative models; all calculations are generated by its programs; the book is not legal, financial or employment advice and makes no claim about any particular AI product. Marked boxes separate published evidence, runnable Python calculations (16 programs), classical sources and the author's hypotheses.

中文版

Data table

The numbers below are printed verbatim by the chapter's Python program and can be reproduced by running it. standard-library Python onlyprinted values that match the answer: 0.4452, 0.6355, 0.9, 2.06

optionuses per decisionerrorP(all 30 right)
restate r = 11.000.1500.0076
restate r = 33.000.0610.1526
restate r = 55.000.0270.4452
restate r = 77.000.0120.6940
test c = 0.51.590.0750.0964
test c = 0.92.060.0150.6355
test c = 1.02.180.0001.0000
Show the full program
# Chapter 6. p12: Restating versus testing. Same channel as p11: each of
# 30 decisions is misread with chance 0.15. Option R: state each decision
# r times and let the agent follow the majority (cost r uses). Option T:
# state it once and add an executable check that covers a share c of
# decisions; a misread covered decision fails its check and is redone
# until it passes (cost 1 + c uses, plus expected redo uses). Residual
# error per decision for T is p (1 - c).
import math
from math import comb

P, D = 0.15, 30

def maj(p, r):
    return math.fsum(comb(r, k) * p ** k * (1 - p) ** (r - k)
               for k in range(r // 2 + 1, r + 1))

if __name__ == "__main__":
    print("   option            uses per decision   error   P(all 30 right)")
    for r in (1, 3, 5, 7):
        e = maj(P, r)
        print("   restate r = {:<4}  {:17.2f}   {:5.3f}   {:15.4f}".format(
            r, r, e, (1 - e) ** D))
    for c in (0.5, 0.9, 1.0):
        e = P * (1 - c)
        redo = c * P / (1 - P)
        print("   test c = {:<6}   {:17.2f}   {:5.3f}   {:15.4f}".format(
            c, 1 + c + redo, e, (1 - e) ** D))

Single source and citation

Fan, Y. (2026). Human-AI Symbiotic Cognition, Parallel Agent Pipelines and Prompt Information Bounds (Version 1.0.0), Chapter 6: An Instruction as a Noisy Channel. 5D Library. DOI pending

DOI pending · monograph page goes live after publication · 5DL-B-103 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026103,
  author    = {Fan, Yafu},
  title     = {Human-AI Symbiotic Cognition, Parallel Agent Pipelines and Prompt Information Bounds},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

Other chapter answers in this book