人工智能能不能 (曾安军) (Z-Library)

【One-Line Pitch】 A deliberately sober, logic-first audit of what artificial intelligence can and cannot do, written for readers who want a counterweight to both hype and doom. Best for general readers curious about AI's real boundaries and for researchers who want a structured, opinionated framework to argue with. 【Book Arc】 - **Opening (~0%–20%)**: Sets up the book's stance and vocabulary — AI as an engineering practice rather than a standalone science — and previews the author's original models (human-brain capability lists, machine-intelligence capability lists, intelligence grading). This stage solves the "where do we even start the argument" problem. - **Early (~20%–40%)**: Surveys the current AI boom across academia, media, industry, and government, using well-known milestones (AlphaGo and AlphaGo Zero, Boston Dynamics robots, Tesla Autopilot/FSD, Siri, the virtual student Hua Zhibing) to show how expectations were built. - **Middle (~40%–60%)**: Moves from spectacle to scrutiny — media enthusiasm and warnings side by side, plus concrete applied cases such as the fully automated "robot chemist" — establishing that real progress and inflated claims coexist. - **Late (~80%)**: Examines corporate strategy and deep technical application, including Google's pushes into protein-structure prediction and AI-assisted chip design, illustrating where AI delivers measurable gains. - **Ending (~100%)**: Consolidates the book's core thesis: AI's potential is real but bounded, and the author's framework is meant to clarify those bounds rather than celebrate or dismiss the technology. 【Key Takeaways】 - **AI is best understood as a technology for empowering machines, not a unified scientific theory** (Opening): This reframing matters because it shifts debate from "is AI a science?" to "what engineering capabilities can we actually build?" (Early) - **The author proposes structured capability models for both human and machine intelligence** (Opening): Human abilities are organized into 8 categories and 18 element-level capabilities, with a parallel machine-intelligence list — a scaffold for deciding what is worth simulating. (Early) - **A key distinction is drawn between machine-intelligence technology and intelligent-machine technology** (Early): The former is general-purpose enabling technology; the latter is about building specific smart machines. This split helps readers locate any AI project on a map. (Middle) - **The book argues AI has hard limits rooted in its technical architecture** (Late): Rather than listing failures, it derives impossibilities from the architecture itself — a more durable form of skepticism than anecdote. (Late) - **ChatGPT is analyzed as both impressive and problematic** (Late): The author examines how it works and where its weaknesses may lie, treating it as a case study rather than a marvel. (Late) - **Intelligence grading standards are proposed** (Middle): A common scale for rating how "smart" a machine is aims to make comparisons and public expectations more disciplined. (Middle) - **The book forecasts an economy of abundance** (Ending): If AI's potential is realized, the author predicts a gradual shift toward a surplus economy — a speculative but clearly flagged projection. (Ending) 【Reading Tips】 - **Deep-read the framework chapters**: The capability models and the machine-intelligence vs. intelligent-machine distinction are the book's backbone; everything later depends on them. - **Skim the boom survey if you already follow AI news**: The early chapters on AlphaGo, robots, and corporate activity are context-setting, not the argument. - **Treat the "cannot" claims as arguments to test**: The author explicitly invites criticism; read the limitation chapters with a pencil and ask whether each impossibility follows from the stated architecture. - **Watch for the author's tone**: He aims for humor and plain language over technical rigor, so don't expect formal proofs — expect a reasoned position. 【Coverage Limits】 The excerpts cover the book's preface, chapter overview, and portions of the opening survey and later application examples; they do not include the full text of the framework chapters, the detailed limitation analysis, or the concluding forecast, so this guide reflects the book's stated structure and themes rather than every argument.