one-second-rule

Source Metadata

Original cangjie-skill frontmatter from the distillation run:

name: one-second-rule
description: |
  当用户想快速判断一项任务是否能被当前AI技术自动化时使用。
  用户在问"这个AI能做吗"、"技术上可行吗"、"能不能用AI自动完成X"时激活。
  不适用于: 需要深度可行性分析的项目评估、涉及情感共鸣或复杂推理的创意任务、LLM时代的长文本生成场景。
source_book: 《给所有人的AI入门课》 吴恩达 (Andrew Ng)
source_chapter: p06, p07
tags: [automation, feasibility, heuristic, task-selection]
related_skills:
  - slug: ml-feasibility
    relation: composes-with
  - slug: automate-task
    relation: depends-on
  - slug: ab-mapping
    relation: composes-with

The One-Second Rule — A Heuristic for Judging AI Feasibility

R — Reading

"As long as it takes less than a second of thought, we can now or soon automate it with supervised learning."

— Andrew Ng, p06

I — Methodological Skeleton (Interpretation)

The One-Second Rule is a quick-screening heuristic: if a human can make the required judgment in less than one second, then current supervised learning can most likely automate that task. The underlying logic is: judgments humans complete "instantly" correspond to "simple concepts"—and simple concepts are exactly what machine learning is best at learning. Conversely, if a human needs several minutes or even hours to complete it (such as writing a report or making a strategic decision), these "complex concepts" remain difficult for current AI. The value of this rule is that it turns the complex question of "technical feasibility" into a quick test anyone can perform—no need to understand algorithms, just simulate human thinking time. It is not a precise scientific law, but an efficient initial screening tool that helps teams quickly rule out clearly infeasible projects before investing substantial resources.

A1 — Applications in the Book (Past Application)

Case 1: Vehicle Detection

Problem: Can an autonomous driving system detect vehicles ahead in real time?
Use of the methodology: A human can tell "there is a car ahead" at a glance—far less than one second
Conclusion: The One-Second Rule judges it feasible
Result: Modern autonomous driving systems have indeed achieved real-time vehicle detection

Case 2: Phone Screen Scratch Detection

Problem: Can a factory use AI to automatically detect phone screen defects?
Use of the methodology: A quality inspector can tell "scratch/no scratch" at a glance—less than one second
Conclusion: The One-Second Rule judges it feasible
Result: Visual quality inspection has become a typical AI application in manufacturing

Case 3: X-ray Pneumonia Diagnosis

Problem: Can AI diagnose pneumonia from X-ray images?
Use of the methodology: An experienced radiologist can make a preliminary judgment at a glance—less than one second
Conclusion: The One-Second Rule judges it feasible
Result: Medical imaging AI has reached near-expert level on multiple diseases

Case 4: Empathetic Customer Reply (Counterexample)

Problem: Can AI automatically generate empathetic customer replies?
Use of the methodology: An excellent customer service agent needs to understand the customer's emotions and weigh wording—this takes several minutes
Conclusion: The One-Second Rule judges it infeasible
Result: Such tasks were indeed difficult to automate in 2019

Case 5: 50-Page Market Report (Counterexample)

Problem: Can AI automatically generate a complete market analysis report
Use of the methodology: Analysts need hours or even days to complete it
Conclusion: The one-second rule determines it is not feasible
Result: Long-horizon structured reasoning is indeed difficult under the supervised learning paradigm

A2 — Trigger Scenarios (Future Trigger) ★

In what situations will users need this skill?

A business team proposes an AI idea and needs to quickly determine whether it is worth deeper investigation
A product manager is prioritizing AI features and needs an initial screening criterion
An investor is evaluating the technical feasibility of an AI startup
A manager with a non-technical background wants to quickly understand "what AI can actually do right now"

Linguistic signals (these should activate it when they appear in the user's words)

"Can this AI do it"
"Is it technically feasible"
"Can AI automatically..."
"To what extent can AI do this now"
"Is this idea realistic"

Distinction from adjacent skills

Difference from ab-mapping : A→B mapping is formalizing the problem into an input-output structure, while the one-second rule judges, after formalization, "whether this task is simple enough for AI to do."
Difference from ml-feasibility : The two-factor framework is a more complete assessment (conceptual simplicity + data sufficiency), while the one-second rule is only the first step of rapid initial screening.
Difference from automate-task : Task decomposition is "breaking a job into tasks," while the one-second rule judges "whether a single task is feasible."

E — Executable Steps (Execution)

After the skill is activated, the agent should execute the following steps:

Clarify the specific task/judgment content
- Ask the user: "What specific judgment do you need AI to make? Please describe a specific decision scenario."
- Completion criterion: The user can describe a clear judgment task (such as "determine whether there is a cat in the image" or "determine whether this email is spam")
Simulate human thinking time
- Ask the user: "How long would an experienced human need to make this judgment? Can it be completed within one second?"
- Completion criterion: The user can give a rough time estimate, or agree/deny that it can be completed "within one second"
Give a feasibility determination and suggest the next step
- If "yes" → tell the user "it may be technically feasible; it is recommended to proceed to data availability assessment (refer to ml-feasibility)"
- If "no" → tell the user "current supervised learning may find it difficult to automate; it is recommended to consider task decomposition or defer it"
- Completion criterion: The user receives a clear determination of "possibly feasible/possibly not feasible" and a next-step recommendation
- Stopping condition: If the task involves emotion, creativity, or complex reasoning, directly determine "currently not feasible" and explain the reason

B — Boundary ★

Do not use this skill in the following situations

The task requires empathy, emotional understanding, or interpersonal interaction (such as psychological counseling, negotiation, or creative writing)
The task requires long chains of logical reasoning (such as legal argumentation or strategic planning)
The task involves complex language generation (such as writing reports or creating novels)
Project decisions that require precise cost-benefit analysis (in which case ml-feasibility or triple-due-diligence should be used)

Failure modes the author warns about in the book

Equating tasks that "can be judged in one second" with "certain to succeed"—ignoring data availability, a key factor
Ignoring dependencies between tasks—the feasibility of a single task does not mean the entire process can be automated

The author's blind spots / era limitations

This rule is based on the 2019 level of AI technology; the emergence of large language models (LLMs) has already made some "slow tasks" feasible (such as writing reports and long-text generation)
The rule assumes that "a human judging in one second = AI can learn it," but some tasks feel simple to humans because they rely on common sense and bodily perception (such as "is this glass of water hot"), which AI still struggles to replicate
The rule does not apply to the evaluation of generative AI—the definition of "simplicity" for generative tasks differs from that for classification tasks

Adjacent methodologies that are easily confused

Difference from the "Turing test": the Turing test determines whether a machine exhibits human intelligence, while the one-second rule determines whether a specific task can be automated
Difference from "technology maturity assessment": technology maturity assessment is more comprehensive, while the one-second rule is only the quickest preliminary screening

Related skills

ml-feasibility (composes-with): the one-second rule serves as the assessment tool for "conceptual simplicity" in the two-factor framework
automate-task (depends-on): after task decomposition, the one-second rule judgment needs to be made for each task
ab-mapping (composes-with): first define A→B, then use the one-second rule to judge whether B is simple enough

Audit information

Verification passed: V1 ✓ / V2 ✓ / V3 ✓
Test pass rate: pending testing
Distillation time: 2026-06-29