人工智能：一种现代的方法（ArtificialIntelligenceaModernApproach）第3版(StuartJ.Russell,PeterNorvig)

书: https://pan.baidu.com/s/1A6ZLSPMbCiZ-L4eRucUvXQ?pwd=kt7i
一些笔记分享给大家：

一、基础定义

1. “AI is the study of agents that receive percepts from the environment and perform actions.”
   （AI研究能感知环境并执行行动的智能体。）
2. “Rationality is doing the right thing, given what one knows; perfection is unattainable.”
   （理性是在已知条件下做正确的事，完美不可企及。）

二、搜索与规划

3. “A search is optimal if the heuristic is admissible (never overestimates the true cost).”*
   （若启发函数可容（永不高估真实代价），A*搜索最优。）
4. “Constraint satisfaction problems (CSPs) exploit domain-specific heuristics to reduce search space.”
   （约束满足问题利用领域启发式缩小搜索空间。）

三、知识表示

5. “First-order logic is expressive enough to represent ‘there exists an x such that…’, unlike propositional logic.”
   （一阶逻辑可表达“存在某个x使得…”，命题逻辑则不能。）
6. “The frame problem: how to succinctly specify what does not change when an action is performed.”
   （框架问题：如何简洁描述动作执行后未变化的部分。）

四、不确定性推理

7. “Bayesian networks compactly represent joint distributions via conditional independence.”
   （贝叶斯网络通过条件独立性压缩表示联合分布。）
8. “Markov decision processes (MDPs) formalize sequential decision-making under uncertainty.”
   （马尔可夫决策过程形式化不确定性下的序列决策。）

五、机器学习

9. “Overfitting occurs when a model fits noise in the training data rather than the underlying pattern.”
   （过拟合是模型拟合训练数据噪声而非真实规律。）
10. “The bias-variance tradeoff: simple models have high bias, complex models have high variance.”
    （偏差-方差权衡：简单模型偏差高，复杂模型方差高。）

六、神经网络

11. “Backpropagation is just gradient descent with the chain rule applied to compute gradients.”
    （反向传播是链式法则下的梯度下降。）

七、伦理与哲学

12. “The Turing Test is about deception, not intelligence—a philosophical limitation.”
    （图灵测试关乎“欺骗”而非智能，存在哲学局限。）
13. “Asimov’s laws of robotics fail in edge cases due to ambiguity in ‘human’ and ‘harm’.”
    （阿西莫夫机器人三定律在边缘场景中因“人类”“伤害”的模糊性失效。）

八、强化学习

14. “Exploration vs. exploitation: the agent must balance trying new actions and maximizing rewards.”
    （探索与利用：智能体需平衡尝试新动作与最大化奖励。）

九、自然语言处理

15. “The ambiguity of language requires world knowledge to resolve, not just statistical patterns.”
    （语言歧义需世界知识而不仅是统计模式来解决。）

十、AI未来

16. “General AI requires integration of perception, reasoning, learning, and action—no component suffices alone.”
    （通用AI需整合感知、推理、学习与行动，单一组件不足。）

---

附：经典问题表述

17. “The Chinese Room argument: syntax manipulation does not imply understanding.”
    （中文房间论证：语法操作不等于理解。）
18. “Moravec’s paradox: easy tasks for humans (vision, movement) are hard for AI, and vice versa.”
    （莫拉维克悖论：人类轻松的任务（视觉、运动）对AI极难，反之亦然。）
19. “The symbol grounding problem: how symbols acquire meaning without human interpretation.”
    （符号接地问题：符号如何不依赖人类解释获得意义。）
20. “AI’s impact will be limited not by its capabilities, but by our ability to specify goals correctly.”
    （AI的局限将源于人类定义目标的能力，而非技术本身。）