Intelligent Agents 智能体

Agents and Environments

Sensors & Actuators

Sensors : Perceiving its environment through sensors.
Actuators : Acting upon that environment through actuators
智能体可以被视为通过传感器感知所处环境并通过执行器对该环境产生作用的东西
Examples
- Human agent
  - Sensors: eyes, Ears and other organs
  - Actuators: hands, legs, mouths, etc…
- Robotic Agent
  - Sensors: cameras, infrared range finder
  - Actuators: various motors
- Software agent
  - Sensors: receiving keystrokes, file contents, network packets
  - Actuators: display on screen, sending network packets, writing files
- Calculator as agent
  - Sensors: Inputs from keyboard, e.g. “2+2 =”
  - Actuators: Display, e.g. “4”

Percept & Percept Sequence

Percept : Agent’s perceptual inputs at any given instant. 只和一小段时间有关。我们使用感知信息来表示任何给定时刻智能体的感知输入。
Percept Sequence:
- The complete history of everything the agent has ever perceived
- An Agent’s choice of action at any given instant can depend on the entire percept sequence observed to date
- 智能体的感知序列是该智能体所受到的所有输入数据的完整历史。一般而言，智能体在任何时刻的行动选择取决于到那个时刻为止该智能体的整个感知序列。

Agent Function

Describe behaviors of an agent
我们可以用一个把任意给定感知序列映射到智能体的行动的智能体函数来描述智能体的行为
Agent function table（从外部反映智能体特性）
- Only record the external characterization of the agent
- Abstract mathematical description 是一个抽象的数学表示
Agent Program（从内部反映）
- Internally, the agent function of an intelligent agent is implemented by an agent program
- Concrete implementation and running on the agent architecture
  是Agent Function的具体实现

Good Behavior: The Concept of Rationality

Performance Measures

作为一般规则，最好根据实际在这个环境中希望得到的结果来设置性能度量，而不是根据智能体应该表现出的行为。

Rationality

Rationality depends on
- The performance measure that defines the criterion of successes 定义成功的标准
- The agent’s prior knowledge of the environment 代理对环境的先验知识
- The actions that the agent can perform 代理可以执行的动作
- The agent’s percept sequence to date 迄今为止代理的感知序列
Definition of a Rational Agent : For each possible percept sequence, a rational agent should select an action that is expected to maximize its performance measure, given the evidence provided by the percept sequence and whatever built-in knowledge. “性能度量最大化”
Characters
- It is not the same as perfection
- It is to maximize expected performance 理性是期望性能的最大化
- In contrast, perfection is to maximize actual performance 完美是理想性能的最大化
- Does not require omniscience
- Rational choice depends only on the percept sequence to date 理性的选择只取决于到当前为止的感知序列

Learning

Autonomy 自主性
- An agent lacks of autonomy if it ignore its own percepts and relies on the prior knowledge of its designer. 扩展一个智能体依赖于设计者的先验知识而不是它自身的感知信息。
- A rational agent should be autonomous
Successful agents should split the task of computing the agent function into 3 different periods
- When the agent is being designed : Some of the computation is done by its designers.
- When it is deliberating on its next action : The agent does more computation
- It learns from experience :It does even more computation to decide how to modify its behavior
Learning allows one to design a single rational agent that will succeed in a vast variety of environments. 当得到关于环境的充分经验后，理性智能体的行为才会有效的独立于它的先验知识。从而，与学习结合使得设计一种在很多不同环境下都能成为的理性智能体成为可能。

The Nature of Environments

The Task Environment

In designing an agent, the first step must always be specifying the task environment as fully as possible.

PEAS Description
- Performance
- Environment
- Actuators
- Sensors

Examples

List of dimensions of environments

Fully observable vs Partially observable

Fully observable
- If an agent’s sensors give it access to the complete state of the environment at each point in time, the sensors detect all aspects that are relevant to the choice of action. 一个智能体的传感器每个时间点上都能让它访问获取环境的完整状态
- 如果传感器能够检测所有与行动选择相关的方面，那么这个任务环境是有效完全可观察的，而相关性取决于任务的度量
- Convenient because agent need not to maintain any internal state to keep track of the world.
Partially observable
- Noisy and inaccurate sensors
- Parts of the state are missing from the sensor data

Deterministic vs Stochastic

Deterministic
- The next state of the environment is completely determined by the current state and the action executed by the agent. Otherwise, stochastic. 环境的下一个状态完全决定于当前的状态和智能体执行的动作
- If the environment is fully observable and deterministic, there is no uncertainty.
- If an environment is deterministic except for the actions of other agents, the environment is Strategic.环境确定+受其他智能体活动的yingxiang=Strategic.
Stochastic
- An partially observable environment is usually stochastic

Episodic vs Sequential

Episodic 片段式的
- Agent’s experience is divided into atomic episodes 智能体的经历被分成一个个片段
- Each episode consists of the agent perceiving and then performing a single action 每个片段组成包含了智能体所感知的信息以及进而执行的单个行动
- The next episode does not depend on the actions taken in previous episodes. So, action in each episode only depends on the episode itself. 下一个片段不依赖于以前的片段中采取的行动，行动的选择只取决于当前片段自身
- Classification tasks are usually episodic 大多数分类任务都是片段式的
Sequential 延续式的
- Current decision could affect all future decisions 当前的所有决策都会影响到未来的决策
- Chess playing and taxi driving are examples of sequential environments

Static vs Dynamic

Dynamic
- Environments which change while an agent is deliberating 环境在智能体思考的时候会变化，环境相对于智能体是动态的。（自动驾驶）
- Continuously asking the agent what it wants to do. If the agent is not yet decided what to do, it is counted as decided to do nothing at that time. 持续要求智能体决定做什么，如果智能体没有做决策，就认为它决定不做任何事情。
- If the environment does not change, but the agent’s performance score does, it is a semidynamic environment. 如果环境本身不随着时间的流逝而变化，但智能体的性能评价随时间变化，那我们称这个环境是半动态的。（国际象棋）
Static
- Easy, no need to keep looking at the world for deciding on an action（字谜）
- No need to worry about the passage of time

Discrete vs Continuous

Continuous
The discrete and continuous distinction can be applied
- To the state of the environment
- To the way time is handled
- To the Percepts and actions of the agent
- The speed and location of the taxi and of the other cars sweep through a range of continuous values over time. The actions of taxi-driving is continuous.
  e.g. steering angles
Discrete
- In chess game, there are finite number of distinct discrete states. Actions and percepts are also discret

Single agent vs multiagent

Single agent
- An agent solving crossword puzzle by itself is a single agent environment
Multiagent
- Two agents playing chess game together is a multiagent environment
- Competitive multiagent environment
  - In chess game, opponent entity B is trying to maximize its own performance measure
    - By the rule of chess game, it is to minimize agent A’s performance measure
    - So, they are competitive in the game
  - Partially Competitive multiagent environment
    - In taxi driving, parking is a partially competitive game. Only one car can
      occupy a parking space
    - In taxi driving, every driving is trying to avoid collisions to maximize its own performance measure and also all agents’ performance measures 大家都不想被撞
      - So, they are partially cooperative
        
        Not to gain a game point by reducing it from others
        
        But also no need to work with others to maximize its performance measure
- The key distinction of agent : Whether B’s behavior is best described as maximizing a performance measure whose value depends on agent A’s behavior.（如何分辨能否当智能体）

Example

Environment Class

在真实世界中存在一定的随机性（Stochastic），因此An agent should be examined in many environments drawn from an environment class。

The Structure of Agents

The job of AI is to design the agent program that implement agent function mapping percepts to actions（实现把感知映射到行为的智能体函数）
Agent architecture 体系结构：Sort of computing device with physical sensors and actuators that the program will run on.
Agent = Architecture + Program
Architecture makes the percepts from the sensors available to the program, runs the program and feeds the program’s action choices to the actuators as they are generated.接受感知信息，运行程序，并把行动选择送至执行器。

Agent program

Take the current percept as input from the sensors and return an action to the actuators.
Agent program takes the current percept as input. If an action depends on the entire precept sequence, the agent will have to remember the percepts. 如果智能体行动依赖整个感知序列，这个智能体不得不记住全部感知信息。Agent function takes the entire percept history.
Pseudocode of an agent

记录感知序列，然后使用感知序列作为索引，到行动表里查询以决策该做什么。行动表中明确地表示了智能体程序实现的智能体函数。这种方式要求我们必须构造包含针对每个可能的感知序列的适当行动的函数表。
- 这种方法存在的问题
  - No physical agent in this universe will have the space to store it 没这么大的表
- The designer would not have time to create the table 没时间建表
- No agent could ever learn all the right table entries from the experience 智能体学不了
- Even if the environment is simple enough to yield a
  feasible table size, the designer still has no guidance
  about how to fill in the table entries. 环境足够简单也没办法填表
- The key challenge for AI is to find out how to write programs that produce rational behavior from a small amount of code rather than from a large number of table entries. AI的关键是搞清楚如何编写程序，在可能的范围内，从少量的代码而不是大量的表条目中产生出理性的行为。
- Four basic types of agent program
  - Simple reflex agents
  - Model-based reflex agents
  - Goal-based agents
  - Utility-based agents

Simple reflex agents

Simple reflex agents（简单反射智能体） selects action on the basis of the current percept, ignore the rest of the percept history. 基于当前感知选择自己的行动，忽略其余的感知历史。
本图中，智能体的决策只建立在当前位置和是否包含灰尘的基础上。忽略了感知历史后cuts down the number of possibilities from 4T4^T4T to 444, 同时当前方格有灰尘时，行动不依赖于所在的地点。
Condition-Action rule 条件-行为规则
- 如果…那么…
  矩形表示智能体决策过程的内部状态，椭圆形表示该过程中用到的背景信息。
Characters
- Simple, but very limited intelligence 智能相当有限
- The SIMPLE-REFLEX-AGENT defined in previous will work only if the correct decision can be made on the basis of only the current percept 只有在可以仅根据当前的感知信息来完成当前决策的情况下才能工作。
  - That is, only if the environment is full observable 只有在环境完全可观察的情况，即使有少量不可观测的情况，也会引起严重问题。
- Infinite loops are often unavoidable in partially observable environment 有可能会出现无限循环的问题
  - An agent may escape from infinite loops if the agent can randomize its action. 避免无限循环-randomize

Model-based reflex agents

The most effective way to handle partial observability is for the agent to keep track of the part of the world it can not see now 让智能体追踪记录当时无法观察到的那部分世界
The agent maintain some sort of internal state that depends on the percept history 智能体维持某种取决于感知历史的内部状态
- Reflects at least some of the unobserved aspects of the current state 反映出当前状态的某些不可观测的方面。
Updating internal state information as time goes by requires 更新内部信息需要
- Information about how the world evolves independently of the agent 关于智能体之外的世界如何独立发展的信息
- Information about how the agent’s own actions affect the world 关于智能体自身的行动如何影响世界
Model of the world - knowledge about “How the world works” .世界的模型-关于世界如何运转 Agents use such a model is called a model-based agents 基于模型的反射型智能体

显示出如何把当前的感知信息与过去的内部状态结合起来产生当前状态的更新描述，使用一个内部模型记录世界的当前状态。然后按照与反射型智能体同样的方式选择行动。
The function UPDATE-STATE
- Responsible for creating the new internal state description 创建新的内部状态的描述
- Interpreting the new percept in the light of existing knowledge about the state
  - It uses information about how the world evolves to keep track of the unseen parts of the world
  - Must know about what the agent’s actions do to the state of the world

Goal-based agents

Knowing about the current state of the environment is not always enough to decide what to do. 只知道当前的状态不足以帮助决定做什么 As well as a current state description, the agent needs some sort of goal information that describes situation that are desirable 智能体需要某种目标信息来描述想要达到的状况
The agent program combines the goal with information about the results of possible actions for selecting actions that achieve the goal. Search and Planning are the subfields of AI devoted to finding action sequences that achieve the agent’s goal. 搜索和规划帮助完成基于目标的行动选择。
Decision making in goal-based agents
- Is fundamentally different from the condition-action rules
- It involves consideration of the future
  - What will happen if I do such-and-such
  - Will that make me happy?
- This information is not explicitly represented in reflex agent esigns 反射型的不考虑这些
Characters
- Appear to be less efficient
- It is more flexible because the knowledge that supports its decision is represented explicitly and can be modified 支持它决策的知识都被显式的表示且可更改。

Utility-based agents

Goals alone are not really enough to generate high-quality behavior in most environments. A more general performance measure should allow a comparison of different world states according to exactly how happy they would make the agent if they could be achieved. We use the term “utility” to say that if one world state is preferred to another. 只有目标没有效用是不行的，所以引用"utility"进行评价。

最终会选择期望值最高的action
Utility function
- Maps a state or a sequence of states onto a real number which describes the associated degree of happiness
- A complete specification of the utility function allows rational decisions in two kinds of cases where goals are inadequate
  - When there are conflicting goals or only some of goals could be achieved, tradeoff according to the utility function. 当目标在冲突时，确定了适当的折中。
    e.g. Speed and safety in driving
  - When there are several goals that the agent can aim for, none of which can be achieved with certainty, utility provides a way in which the likelihood of success can be weighed against the
    importance of the goals多个目标没有一个是有把握的，根据目标的重要性对成功的似然率加权。

Learning agent

Learning agent’s conceptual components
- Learning element 学习元件
- Performance element 执行元件
- Critic 评论原件
- Problem generator 问题产生器

AI - Intelligent Agents相关推荐

7 Papers Radios | 6年前的ResNet重回巅峰；AI辩手登上Nature封面
机器之心 & ArXiv Weekly Radiostation 参与:维度.楚航.罗若天本周的重要论文包括IBM 海法研究院开始研发能够与人类辩论的 AI 系统--Project Deba ...
ai人工智能在手机的应用_强化学习在人工智能中的应用
ai人工智能在手机的应用 The reinforcement learning is being used in many Intelligent Systems and the developers ...
人工智能AI在智能交通领域中的应用
摘要人工智能(AI)是指机器执行人类能够轻松完成的感知.推理.学习和解决问题等认知功能的能力.过去20年以来,由于互联网产生的海量数据的可用性,人工智能在全球范围内获得了关注.最近,使用高级算法处理 ...
从冯诺伊曼结构看AI
AI, Artificial Intelligence, 人工智能. 就像每个人眼中都有一个自己的哈姆雷特一样,每一个看AI 都是不一样的.作为一个老程序员,也只是一个工作时间长一些的程序员而已,本没 ...
从冯诺伊曼结构看AI 1
版权声明:本文为半吊子子全栈工匠(wireless_com,同公众号)原创文章,未经允许不得转载. https://blog.csdn.net/wireless_com/article/details ...
ai人工智能将替代人类_教AI学习人类如何有效计划
ai人工智能将替代人类 Human planning is hierarchical. Whether planning something simple like cooking dinner or ...
网易100天---26、What Is Artificial Intelligence (AI)?
What Is Artificial Intelligence (AI)? 温馨提示:正文推送半小时,您将收到精编讲义 Passage 1 人工智能的定义 The Definition of Arti ...
OpenAI开发AI版《文明》，一块CPU就能重现AI生存战争史
原文链接:https://mp.weixin.qq.com/s?__biz=MzIzNjc1NzUzMw==&mid=2247515805&idx=2&sn=3a3952337 ...
[wiki翻译]分布式人工智能-Distributed AI
分布式人工智能(DAI)也被叫做去中心化人工智能,他是人工智能研究领域的一个子领域,致力于开发分布式解决方案. DAI与多智能体系统(Multi-agent sytem)领域域切相关,并且是多智能体系 ...
人工智能AI工具汇总（AIGC ChatGPT时代个体崛起）
Name Category Website Description 描述 <AIGC时代:超级个体的崛起> 小报童 https://xiaobot.net/p/SuperIndividua ...

AI - Intelligent Agents