课文翻译
课文翻译 P1-P10
1
Artificial Intelligence
In this chapter we explore the branch of computer science known as artificial intelligence. Although this field is relatively young, it has produced some astonishing results such as expert chess players, computers that appear to learn and reason, and machines that coordinate their activities to achieve a common goal such as winning a soccer game. In artificial intelligence, today's science fiction might well be tomorrow's reality.
11.1 Intelligence and Machines
Intelligent Agents
Research Methodologies
The Turing Test
11.2 Perception
Understanding Images Language Processing
11.3 Reasoning
Production Systems Search Trees Heuristics
11.4 Additional Areas of Research
Representing and Manipulating Knowledge
Learning
Genetic Algorithms
11.5 Artificial Neural Networks
Basic Properties
Training Artificial Neural
Networks
Associative Memory
11.6 Robotics
11.7 Considering the Consequences
在本章中,我们将探讨计算机科学的一个分支,即人工智能。尽管这一领域相对年轻,但它已经产生了一些惊人的成果,比如国际象棋专家、似乎可以学习和推理的计算机,以及协调其活动以实现共同目标(如赢得足球比赛)的机器。在人工智能领域,今天的科幻小说很可能就是明天的现实。
11.1 智能和机器
智能代理
研究方法
图灵测试
11.2 感知
图像理解
语言处理
11.3 推理
产生式系统
搜索树
启发式
11.4 其他研究领域
表示和操作知识
学习
遗传算法
11.5 人工神经网络
基本属性
训练人工神经网络
联想存储器
11.6 机器人学
11.7 考虑到后果
2
Artificial Intelligence is the field of computer science that seeks to build autonomous machines-machines that can carry out complex tasks without human intervention. This goal requires that machines be able to perceive and reason. Such capabilities fall within the category of commonsense activities that, although natural for the human mind, are proving difficult for machines. The result is that work in the field continues to be challenging. In this chapter we explore some of the topics in this vast area of research.
人工智能是计算机科学的一个领域,它寻求建立自主机器——能够在没有人类干预的情况下执行复杂任务的机器。这个目标要求机器能够感知和推理。这种能力属于常识性活动的范畴,尽管对人类思维来说是自然的,但对机器来说却是困难的。其结果是,该领域的工作仍然具有挑战性。在本章中,我们将探讨这一广阔研究领域中的一些主题。
11.1 Intelligence and Machines
The field of artificial intelligence is quite large and merges with other subjects such as psychology, neurology, mathematics, linguistics, and electrical and mechanical engineering. To focus our thoughts, then, we begin by considering the concept of an agent and the types of intelligent behavior that an agent might exhibit. Indeed, much of the research in artificial intelligence can be categorized in terms of an agent's behavior.
11.1 智能和机器
人工智能的领域相当大,并与其他学科,如心理学,神经学,数学,语言学,电气和机械工程合并。那么,为了集成我们的思想,我们首先考虑智能体的概念和智能体可能表现出的智能行为类型。事实上,人工智能的许多研究都可以根据智能体的行为进行分类。
Intelligent Agents
An agent is a“device”that responds to stimuli from its environment. It is natural to envision an agent as an individual machine such as a robot, although an agent may take other forms such as an autonomous airplane, a character in an interactive video game, or a process communicating with other processes over the Internet(perhaps as a client, a server, or a peer). Most agents have sensors by which they receive data from their environments and actuators by which they can affect their environments. Examples of sensors include microphones, cameras, range sensors, and air or soil sampling devices. Examples of actuators include wheels, legs, wings, grippers, and speech synthesizers.
智能体
智能体是一种对环境刺激做出反应的 "设备"。人们很自然地会把智能体想象成一个独立的机器,如机器人,尽管智能体也可以采取其他形式,如自主飞机、交互式视频游戏中的角色或通过互联网与其他进程通信的进程(可能是客户端、服务器或对等体)。大多数智能体都有从环境中接收数据的传感器和影响环境的执行器。传感器的例子包括麦克风、摄像头、测距传感器以及空气或土壤采样装置。执行器的例子包括轮子、腿、翅膀、抓手和语音合成器。
Much of the research in artificial intelligence can be characterized in the context of building agents that behave intelligently, meaning that the actions of the agent's actuators must be rational responses to the data received though its sensors. In turn, we can classify this research by considering different levels of these responses.
人工智能中的许多研究可以在构建智能行为的智能体的背景下进行描述,这意味着智能体的执行器必须对其传感器接收到的数据做出合理的响应。反过来,我们可以通过考虑这些反应的不同层次来对这些研究进行分类。
The simplest response is a reflex action, which is merely a predetermined response to the input data. Higher levels of response are required to obtain more “intelligent”behavior. For example, we might empower an agent with knowledge of its environment and require that the agent adjust its actions accordingly. The process of throwing a baseball is largely a reflex action but determining how and where to throw the ball requires knowledge of the current environment.(There is one out with runners on first and third.) How such real-world knowledge can be stored, updated, accessed, and ultimately applied in the decision-making process continues to be a challenging problem in artificial intelligence.
最简单的响应是反射动作,其仅仅是对输入数据的预定响应。需要更高级别的响应才能获得更“智能”的行为。例如,我们可以向智能体赋予环境知识,并要求智能体相应地调整自己的行动。投掷棒球的过程在很大程度上是一个反射动作,但决定如何以及在哪里投球需要当前环境的知识。(一垒和三垒,有一人出局)如何存储、更新、访问这些真实世界的知识,并最终将其应用于决策过程,仍然是人工智能中一个具有挑战性的问题。
Another level of response is required if we want the agent to seek a goal such as winning a game of chess or maneuvering through a crowded passageway. Such goal-directed behavior requires that the agent's response, or sequence of responses, be the result of deliberately forming a plan of action or selecting the best action among the current options.
如果我们想让智能体寻找一个目标,例如赢得一场国际象棋比赛或在拥挤的通道中穿行,则需要另一个级别的响应。这种以目标为导向的行为要求智能体的反应或反应序列是有意形成行动计划或在当前选项中选择最佳行动的结果。
In some cases an agent's responses improve over time as the agent learns. This could take the form of developing procedural knowledge(learning “how”) or storing declarative knowledge (learning “what”). Learning procedural
在某些情况下,智能体的响应随着智能体学习的时间而改进。这可以采取开发程序性知识(学习“如何”)或存储陈述性知识(学习“什么”)的形式。
3
knowledge usually involves a trial and error process by which an agent learns appropriate actions by being punished for poor actions and rewarded for good ones. Following this approach, agents have been developed that, over time, improve their abilities in competitive games such as checkers and chess. Learning declarative knowledge usually takes the form of expanding or altering the“facts”in an agent's store of knowledge. For example, a baseball player must repeatedly adjust his or her database of knowledge(there is still just one out, but now runners are on first and second) from which rational responses to future events are determined.
学习程序性知识通常涉及一个试错过程,通过这个过程,智能体可以通过对不良行为的惩罚和对良好行为的奖励来学习适当的行动。按照这种方法开发的智能体,随着时间的推移,可以在跳棋和国际象棋等竞技游戏中提高自己的能力。学习陈述性知识的形式通常是扩展或改变智能体知识库中的 "事实"。例如,棒球运动员必须反复调整他或她的知识数据库(仍然只有一个出局数,但现在跑垒手在一垒和二垒),并据此确定对未来事件的合理反应。
To produce rational responses to stimuli, an agent must “understand”the stimuli received by its sensors. That is, an agent must be able to extract information from the data produced by its sensors, or in other words, an agent must be able to perceive. In some cases this is a straightforward process. Signals obtained from a gyroscope are easily encoded in forms compatible with calculations for determining responses. But in other cases extracting information from input data is difficult. Examples include understanding speech and images. Likewise, agents must be able to formulate their responses in terms compatible with their actuators. This might be a straightforward process or it might require an agent to formulate responses as complete spoken sentences-meaning that the agent must generate speech. In turn, such topics as image processing and analysis, natural language understanding, and speech generation are important areas of research.
为了对刺激产生合理的反应,智能体必须“理解”其传感器接收到的刺激。也就是说,智能体必须能够从其传感器产生的数据中提取信息,或者换句话说,智能体必须能够感知。在某些情况下