Automated Systems
IGCSE Computer Science Topic 6 10:22 English narration · English + 中文 subtitles burned in
Chapters
Transcript
Picture a greenhouse at three in the morning.
想象凌晨三点的一间温室。
Nobody is there.
没有人在里面。
The air warms up a little, and a window slides open all by itself.
空气稍微变暖了一点，一扇窗子就自己滑开了。
An hour later it closes again.
一小时后它又自己关上。
Nobody decided that — no person was watching, nobody pressed anything.
没有人做这个决定——没有人在盯着，也没有人按下任何按钮。
And the machinery that did it is startlingly simple: three parts, in a loop, repeating for ever.
而做成这件事的装置简单得惊人：三个部件，组成一个循环，永远重复下去。
Once you know that loop you can explain a greenhouse, a car park barrier, a central heating system and a chemical plant with the same three sentences.
只要你懂了这个循环，你就能用同样的三句话，解释温室、停车场道闸、 中央供暖系统，还有一座化工厂。
Machines that run themselves.
会自己运转的机器。
Today: what an automated system is, the sensor-microprocessor-actuator loop, choosing the right sensor, the advantages and disadvantages, robotics, artificial intelligence, expert systems, machine learning, and the three levels of AI.
今天我们讲：什么是自动化系统、传感器—微处理器—执行器这个循环、 如何选对传感器、它的优点与缺点、机器人技术、人工智能、专家系统、机器学习， 以及人工智能的三个层次。
Let's begin.
让我们开始吧。
An automated system is a mix of software and hardware that senses and responds to data in its environment, with no need for human intervention.
自动化系统，是软件和硬件的结合，它能感知环境中的数据并作出反应，而不需要人工干预。
That last phrase is the one the mark scheme wants: no need for human intervention.
最后那半句正是评分标准想要的：不需要人工干预。
In plain words, it works on its own.
说白了，就是它自己会运转。
You already meet them everywhere — a central heating system, a chemical process in a factory, a greenhouse, a car park barrier.
其实你到处都见得到它们——中央供暖系统、工厂里的化学流程、温室、停车场的道闸。
Three parts do all the work.
所有的工作都由三个部件完成。
The sensor measures a physical quantity — temperature, light, pressure — and sends that data onward.
传感器测量某个物理量——温度、光照、压力——并把数据送出去。
The microprocessor compares that data with a stored value, and makes a decision.
微处理器把这些数据和一个存储值作比较，然后作出决定。
The actuator receives a signal and causes movement or action — opening a valve, starting a fan.
执行器接收信号， 并引发运动或动作——打开阀门，启动风扇。
Take the greenhouse.
就拿温室来说。
A temperature sensor reads the heat.
温度传感器读取热度。
The microprocessor compares it with the wanted value, say twenty-five degrees.
微处理器把它和设定值比较，比如二十五度。
Too hot, and it signals an actuator to open a window; too cold, and it switches on a heater.
太热了，它就给执行器发信号去开窗； 太冷了，就打开加热器。
Then the loop repeats, continuously.
然后这个循环不断重复。
Two marks nearly always sit on those exact words: compares with a stored value, and the process repeats.
有两分几乎总是落在这几个字上： 与存储值比较，以及这个过程会重复。
Describe how an automatic greenhouse holds the temperature at twenty-five degrees.
描述自动温室如何把温度保持在二十五度。
A temperature sensor continuously measures the temperature and sends its reading, converted to digital by an A D C, to the microprocessor.
温度传感器持续测量温度，并把读数经模数转换器 转成数字后送给微处理器。
The microprocessor compares the reading with the stored value of twenty-five degrees.
微处理器把读数和存储的二十五度比较。
If the temperature is above it, the microprocessor signals an actuator to open a window or switch on a fan.
如果温度高于这个值， 微处理器就给执行器发信号去开窗或开风扇。
If it is below, it switches on a heater.
如果低于这个值，就打开加热器。
The whole loop then repeats continuously.
然后整个循环 不断重复。
Two marks nearly always sit on the words compares with a stored pre-set value, and the process repeats.
有两分几乎总是落在这几个字上：与存储的预设值比较，以及这个过程会重复。
A description that stops at the sensor tells the computer leaves those marks behind.
如果描述停在"传感器告诉计算机"，这两分就丢了。
The syllabus names a lot of sensor types, and questions love to ask which one you would use.
大纲里列了很多种传感器，而考题最爱问的就是你会用哪一种。
Each one reads exactly one physical quantity.
每一种都只读取一个物理量。
Temperature reads heat.
温度传感器读热度。
Light reads brightness.
光敏传感器读亮度。
Pressure reads force on a surface.
压力传感器读作用在表面上的力。
Infra-red, also called proximity, detects a nearby object.
红外（也叫接近）传感器探测附近有没有物体。
Acoustic reads sound level.
声学传感器读声音大小。
Humidity reads water in the air or the soil.
湿度传感器读空气或土壤里的水分。
Gas detects a gas being present.
气体传感器探测某种气体是否存在。
pH reads acidity.
pH 传感器读酸碱度。
An accelerometer reads movement or tilt.
加速度计读运动或倾斜。
And there are magnetic field, flow and level sensors too.
此外还有磁场、流量和液位传感器。
So the rule for choosing is simple: pick the sensor that reads the right physical quantity.
所以选择的原则很简单：挑那个读取正确物理量的传感器。
An automatic door needs infra-red, because it must detect a person nearby.
自动门需要红外传感器， 因为它必须探测到附近有人。
And a fish tank needs pH and temperature, because those are the quantities that matter to the fish.
而鱼缸需要 pH 和温度传感器，因为这才是对鱼要紧的量。
Each named sensor has a typical use, and that is how you pick it.
每一种点名的传感器都有一个典型用途，你就是靠这个来选的。
Temperature: ovens, greenhouses, heating.
温度：烤箱、温室、供暖。
Light: automatic lights and cameras.
光：自动灯和相机。
Pressure: alarm floor mats and touchscreens.
压力：报警地垫和触摸屏。
Infra-red: automatic doors and parking sensors.
红外：自动门和泊车传感器。
Acoustic: noise monitors.
声学：噪声监测。
Humidity: greenhouses and irrigation.
湿度：温室和灌溉。
Gas: smoke and carbon monoxide alarms.
气体：烟雾和一氧化碳报警。
pH: pools and fish tanks.
pH：泳池和鱼缸。
An accelerometer: phones and airbag triggers.
加速度计：手机和安全气囊触发。
Magnetic field: door contacts and compasses.
磁场：门磁和指南针。
Flow and level: pipes and tanks.
流量和液位：管道和水箱。
So is automation worth it?
那么，自动化到底值不值？
On the plus side: these systems work all day and all night without rest.
好处是：这些系统日夜不停地工作，从不休息。
They are faster and more consistent than people.
它们比人更快，也更稳定一致。
They can work in places that are unsafe for a human.
它们能在对人来说不安全的地方工作。
And they make fewer mistakes.
而且它们犯的错更少。
On the other side: they cost a lot to buy and set up.
另一面是：它们的购买和安装成本很高。
They can break down, and then they need an expert to repair them.
它们会出故障，出了故障还得请专家来修。
They may replace people's jobs.
它们可能取代人的工作岗位。
And they cannot easily react to something they were never built for — that last one is the deepest limit, and it comes back when we talk about robots.
而且，对于它们从来没有被设计去应付的情况，它们很难作出反应—— 最后这一点是最根本的局限，等我们讲到机器人时它还会再出现。
Automated systems show up in industry, transport, agriculture, weather, gaming and lighting.
自动化系统出现在工业、交通、农业、气象、游戏和照明里。
A camera drone can fly itself along a set route to collect data.
摄像无人机可以按设定路线自己飞， 去采集数据。
The situation names the quantities: farming needs humidity and moisture; weather gathering needs temperature, pressure and wind; lighting needs a light sensor.
场景会点出物理量：农业需要湿度和水分；气象采集需要温度、气压和风； 照明需要光敏传感器。
A weather station uses those sensors together.
气象站把这些传感器一起用。
Gaming uses sensors for movement.
游戏用传感器读运动。
Lighting uses a light sensor to switch lamps on at dusk.
照明用光敏传感器 在黄昏开灯。
The same loop runs in every one of them.
同一个循环在每一个场景里都在跑。
Robotics is the branch of technology that deals with designing, building and operating robots.
机器人技术是研究机器人的设计、制造和运行的技术分支。
A robot has three characteristics you should be able to list.
机器人有三个特征，你应该能列出来。
A physical structure — a mechanical body, such as arms.
一个物理结构——机械的身体，比如手臂。
Electrical components — motors, sensors and wiring.
电子元件——电机、传感器和线路。
And programmable instructions: it follows a program, and that program can be changed — which is also its weakness, a lack of independent decision-making, since it only does what it is programmed to do.
还有可编程指令：它按照一个程序运行，而这个程序是可以更改的。
Robots do two kinds of job in the syllabus: factory equipment, like building cars, welding and moving heavy parts; and domestic appliances, like a robotic vacuum cleaner.
大纲里机器人做两类工作：工厂设备，比如造汽车、焊接和搬运重物； 以及家用电器，比如扫地机器人。
They do dull or dangerous work, quickly and accurately, and they never get tired or bored.
它们做那些枯燥或者危险的活，又快又准，而且从不疲倦、从不厌烦。
But they are expensive, they can replace human workers, and here is the key limit: a robot has no independent decision-making.
但它们价格昂贵，可能取代人类工人，而且这里有一个关键局限：机器人缺乏独立决策能力。
It only does what it was programmed to do.
它只会做被编程去做的事。
When something unexpected happens, it cannot choose for itself.
一旦发生意料之外的情况，它没办法自己作出选择。
Robots buy you three things: they do dull or dangerous jobs, they work quickly and accurately, and they do not get tired or bored.
机器人给你三样东西：它们做枯燥或危险的活，它们又快又准，而且从不疲倦、从不厌烦。
They cost three things back: they are expensive to buy and maintain, they can replace human workers, and a robot has a lack of independent decision-making.
它们也要你付三样代价：购买和维护都很贵，它们可能取代人类工人，而且机器人缺乏独立决策。
It only does what it is programmed to do.
它只会做被编程去做的事。
Artificial intelligence is the simulation of human intelligence by computer systems — a computer doing tasks that normally need human thinking.
人工智能，是用计算机系统去模拟人类智能——也就是让计算机去做那些通常需要人来思考的任务。
AI systems have five features you should know.
人工智能系统有五个你该掌握的特征。
They collect data, and the rules for using that data.
它们收集数据，以及使用这些数据的规则。
They can reason — work things out using those rules.
它们能够推理——用这些规则把事情推算出来。
They can draw conclusions, which may be approximate, a best guess, or definite, certain.
它们能够得出结论， 这个结论可能是近似的、也就是一个最佳猜测，也可能是确定的。
They can learn from data.
它们能够从数据中学习。
And they can adapt — change their behaviour as new data arrives.
而且它们能够适应——随着新数据的到来改变自己的行为。
Three examples the syllabus names: expert systems, which give advice like a human expert; natural language processing, understanding human speech or text; and self-driving cars.
大纲点名了三个例子： 专家系统，像人类专家一样给出建议；自然语言处理，理解人说的话或写的字； 还有自动驾驶汽车。
An expert system stores human expertise and reasons from it.
专家系统把人类的专业知识存起来，并在此基础上进行推理。
It has four parts, and the exam asks for all four.
它有四个部分，考试四个都会问。
The knowledge base — the stored facts about the subject.
知识库——关于这个主题的已存事实。
The rule base — the if-then rules a human expert would apply.
规则库——人类专家会运用的那些"如果……那么……"规则。
The inference engine — which applies those rules to those facts to reach a conclusion.
推理引擎——把这些规则应用到这些事实上，从而得出结论。
And the user interface — which asks the user questions and shows the result.
还有用户界面—— 它向用户提问，并把结果显示出来。
Put it together: the interface collects facts from the user, the inference engine runs the rule base against the knowledge base, and the system outputs a diagnosis or a probability.
把它们串起来：界面从用户那里收集事实， 推理引擎拿规则库去跑知识库，然后系统输出一个诊断或者一个概率。
That is exactly how one works when it is helping a doctor diagnose an illness, or finding a fault in a car.
当它在帮医生诊断疾病、或者在查一辆车的故障时，运作方式正是如此。
Machine learning is a program that improves its own performance from experience — without being reprogrammed by anybody.
机器学习是一种能从经验中自我改进的程序——而且不需要任何人重新给它编程。
That phrase matters: without being reprogrammed.
"不需要重新编程"这句话很关键。
It finds patterns in data and adapts by itself.
它自己从数据里找出规律，并作出调整。
Three everyday examples.
三个日常的例子。
A search engine gets better at ranking results.
搜索引擎排结果排得越来越准。
A voice assistant recognises spoken commands more accurately.
语音助手把口令认得越来越对。
And a robot vacuum gradually learns the layout of a room.
而扫地机器人会慢慢摸清一个房间的布局。
Nobody rewrote any of those programs — they got better from the data they saw.
这几个程序都没有被谁重写过—— 它们是靠自己看到的数据变好的。
Finally, three levels of AI — and this is a favourite question.
最后是人工智能的三个层次——这也是一道很受欢迎的考题。
Narrow AI can do only one task, or a small set of tasks: a chess program, a spam filter, a face unlock.
弱人工智能只能做一件任务， 或者一小组任务：一个下棋程序、一个垃圾邮件过滤器、一个人脸解锁。
General AI could do any task a human can do, switching freely between many different ones.
通用人工智能则能做人能做的任何任务，并且可以在很多不同任务之间自由切换。
Strong AI would think and be aware, like a real human mind.
强人工智能会像真正的人类心智那样思考，并具有自我意识。
Now the part people get wrong.
现在说人们最常搞错的地方。
Every AI that exists today is narrow AI.
今天存在的每一个人工智能，都是弱人工智能。
General AI and strong AI do not exist yet.
通用人工智能和强人工智能都还不存在。
If a question asks which level a chess computer is, the answer is narrow — no matter how good it is at chess.
如果题目问一台下棋计算机属于哪个层次，答案就是弱人工智能——不管它棋下得多好。
Four marks to lock in.
四个要拿稳的分。
First, the control loop: sensor, microprocessor, actuator — and say that it compares with a stored value and that the loop repeats.
第一，控制循环：传感器、微处理器、执行器—— 而且要写出"与存储值比较"以及"这个循环会重复"。
Second, do not mix the parts up: a sensor sends data in, an actuator causes movement out.
第二，别把部件搞混： 传感器把数据送进来，执行器把动作送出去。
Third, a robot cannot decide for itself when something unexpected happens.
第三，遇到意料之外的情况时，机器人没法自己作决定。
Fourth, all AI today is narrow AI; general and strong AI do not exist yet.
第四，今天所有的人工智能都是弱人工智能；通用和强人工智能都还不存在。
Nail these, and this topic is yours.
掌握这些，这个专题就是你的了。