What is artificial intelligence? (Popular science: What exactly is artificial intelligence?) AI didn

 

1. Artificial intelligence software chat GPT

In the morning, I posted this message: Google Photos automatically helped our little friend create a 1 minute and 10 second soundtrack video, and even volunteered to give it a Chinese name called "Baby Grows Really Fast". Then, when I was completely unprepared, I pushed it to me. To be honest, I mostly deleted the things that Google Photos had volunteered to do before, but this time, it was very infectious, valuable for preservation, and deeply moved.

2. Artificial Intelligence Movies

To be honest, artificial intelligence technology is becoming more humanized and interesting, and then someone commented, 'Is this based on rules?'? Many people's babies have been sent videos with the same name by Google Photos. Can this be considered artificial intelligence? Well, it's true. Nowadays, facial recognition is so mature. You can identify the same person from a bunch of photos, and then edit a set of photos and videos according to a fixed pattern in chronological order - all routines, all routines. It doesn't seem intelligent at all.

3. List of artificial intelligence stock leaders

On the other hand, just a few years ago, people still referred to facial recognition as artificial intelligence? Why don't we see face recognition today? Because this technology is no longer stunning? Ah, human beings who love new things and dislike old things, but when I think about it the other way around, I have uploaded thousands or even tens of thousands of photos I have taken over the years to Google Photos without anyone complaining. They also know that you don't have time to organize your baby's photos or didn't even think about creating a growth album for your baby. They took the initiative to help you do it, but it moved you to a mess, with both human and technological flavors. Isn't this both intelligence and emotional intelligence? Not even artificial intelligence?.

4. Artificial intelligence AI

Pulling it around, the question arises: What exactly is artificial intelligence? First, let's look at the definition of what children's shoes in China will look like if they want to learn artificial intelligence without climbing over the wall. On Baidu Baike, the entry of "artificial intelligence" is an entry written by such an authoritative expert contributed by several experts called "Expert Committee for the Evaluation of Scientific Terms in Popular Science Encyclopedia of China, China Communications Society". The most important first few definitions only indicate an original source, It is a promotional public relations soft article called "Artificial Intelligence, Where Are iFlytek's Goals?" I really don't know if these experts want to popularize knowledge or promote products.

5. Artificial intelligence computing power

It can be inferred that the definition of Baidu Baike is actually a copy of an old textbook. If you don't believe in searching, you can not only find the name of the textbook, but also find hundreds of thousands of web pages that use this definition as a test question for civil servants, adult education and other strange things in a technical term. The quality of the core definition is still the same. Later, the discussion on artificial intelligence will be expanded, I really don't know how many inquisitive primary and secondary school students will be affected (high-energy warning: experts later "accidentally" mixed into the promotional text of Changhong Intelligent TV).

6. Artificial Intelligence English

By the way, in Zhihuli, the definition of the topic page of "artificial intelligence" is probably taken directly from Baidu Baike, without indicating the original source, The current situation of knowledge dissemination in the network environment inside the wall can be seen (aside: can you write a mini program when you have time to automatically analyze and analyze the channels and directions of various low-level knowledge dissemination in the network inside the wall, to see if the victims are more civil servants, more white-collar workers, or more primary and secondary school students.

7. Artificial Intelligence GPT

）The book is authentic. In the textbooks I used to read, there were no awkward and scholarly definitions. According to my understanding, in history, the definition of artificial intelligence has undergone multiple transformations and some superficial ones that failed to reveal internal laws have long been abandoned by researchers. However, to this day, there are still many widely accepted definitions.

8. Artificial Intelligence and Intelligent Manufacturing

The specific definition used usually depends on the context in which we discuss the problem and the focus of our attention. Here, we will post a recently written long article on popular science, briefly listing and analyzing several historical influences, Or the currently popular definitions of artificial intelligence (Stuart Russell, Peter Norvig, Artistic Intelligence: A Modern Approach, Third edition summarizes the interrelationships between these definitions at the theoretical level, which can be referenced).

9. Ranking of the Best Schools in Artificial Intelligence

It is quite interesting to analyze and discuss these definitions, which is similar to ancient philosophers sitting around to discuss "why people are human", or science fiction fans debating Asimov's "three laws of robots". In fact, many pragmatism oppose metaphysical discussions, and they will say loudly, Hey, what's artificial intelligence? As long as machines can help people solve problems, it's okay.

10. Artificial Intelligence Big Model

Definition 1: AI is a computer program that makes people think it's incredible. Artificial intelligence is a machine that can accomplish things that people don't think machines can do - this definition is very subjective, but also very interesting.

Whether a computer program is artificial intelligence or not is completely defined by the behavior of the program, which makes people dumbfounded. This empiricism only definition is obviously inconsistent, and will vary with times and backgrounds The experience of judges varies and different standards are applied, but this definition often reflects the cognitive style of the largest majority of ordinary people in an era towards artificial intelligence: whenever a new AI hotspot appears, news media and the public always use their own experience to judge the value of AI technology, regardless of whether this technology is fundamentally "intelligent" or not.

The history of computer chess has clearly revealed the irony of this definition. In the early days, due to limitations in running speed and storage space, computers could only be used to solve relatively simple chess game problems, such as checkers. However, this did not prevent people from referring to a computer that could play chess as an intelligent machine at that time, In the minds of most people, an ordinary computer is nothing more than a machine that can solve arithmetic problems at a very fast speed.

In 1951, Christopher Strachey of the University of Manchester, England, wrote the first computer program that can play checkers. In the mid-1950s, the checkers program developed by Arthur Samuel of IBM could compete with amateur players.

In 1962, Arthur Samuel's program defeated a blind checker master and became a major news event (Jill Cirasella, Danny Kopec, The History of Computer Games, 2006). The vast majority of media and the public believed that similar checker programs were all about artificial intelligence.

Not long after, many skilled programmers discovered that computers are basically using search or optimization to solve game problems. Although there are various algorithmic techniques to avoid exhaustion, in the public's eyes, programs are simply finding the best chess steps step by step according to pre written search strategies. With the popularity of PCs, every personal computer can run a highly skilled checkers program, The computer that can play chess has gradually lost its mysterious aura.

People began to doubt the intelligence level of checkers programs, and many people would use chess as an example, provocatively saying, "What's so remarkable about playing checkers? When you win the world championship on the chess board, it's called artificial intelligence." Everyone is familiar with the following things. In 1996, the IBM research team worked hard to create a computer called Deep Blue to challenge the world chess king Kasparov. Although they unfortunately lost, But people have seen hope that computers will triumph over humans.

In 1997, Deep Blue made a comeback and gained a great reputation after defeating Kasparov in a six game chess match. At that time, almost everyone around the world was discussing the power and horror of Deep Blue, and no one doubted that Deep Blue was a representative of artificial intelligence. At least, the public was willing to believe that in the huge black box of Deep Blue, there was a special "brain" that was not inferior to humans in the field of chess games.

The good times did not last long, and a history similar to checkers quickly reappeared. When international chess and Chinese chess were already played proficiently by computers, even chess programs on a mobile phone or tablet could compete with human experts. The public immediately began to doubt whether such game programs could still be considered artificial intelligence.

The reason is simple. The public is always willing to prove that humans are unique in terms of intelligence, whether they truly understand the details of algorithms or not. People always say that computers are just mechanically searching or exhaustive under program control. In fact, similar things happen to computer algorithms such as OCR that have lost their freshness (Roger C. Schank, Where the AI? AI Magazine Volume 12 Number 4, 1991), The facial recognition mentioned at the beginning of this article is no exception.

After refusing to acknowledge that chess programs are artificial intelligence, the public found the last battlefield to maintain human intelligence and dignity - Go. Until early 2016, apart from a professional Go player named Fan Hui and a small research and development team from Google DeepMind, almost all Earthlings, including Go experts and many computer experts, often said, What's so great about playing chess? If you really have intelligence, come and try playing Go with the world champion? Go is an inexhaustible search that relies on human perception of the overall situation. It is the only chess game where computers cannot defeat humans.

”Unfortunately, human arrogance has once again been mercilessly mocked by the rapidly developing artificial intelligence algorithms. On March 9, 2016, Go world champion Li Shishi sat in front of AlphaGo, and fate once again fell. With AlphaGo's 4-1 victory in the fifth round, the heat and panic about artificial intelligence spread worldwide, triggering a wave of artificial intelligence promotion, research and development, and investment.

Today, no one doubts that the core algorithm of AlphaGo is artificial intelligence. But when we think about checkers and chess in the past, did people not respect the program that defeated the human world champion? In a few years, when the Go program on a mobile phone can easily defeat professional chess players, and when all Go games require strict scrutiny of mobile phone cheating, will people still think that playing Go on a computer is an incredible thing? Will people still consider Go programs as representatives of artificial intelligence?.

Definition 2: AI is a computer program with a similar way of thinking to human beings. This is a very popular definition in the early development of artificial intelligence and another similar way. The same definition from the origin of the way of thinking is that AI is a computer program that can think according to the logical laws of thinking. Fundamentally, it is an intuitive idea similar to bionics.

Since it is called artificial intelligence, it is the most straightforward way to use programs to simulate human intelligence, but historical experience has proved that the idea of bionics is not necessarily feasible in the development of science and technology. One of the best and most famous examples is that aircraft humans have been dreaming of flying into the sky in the way that birds flutter their wings for thousands of years, but ironically, they really fly with humans in the sky and break the speed of birds The record of flight altitude is a fixed wing aircraft with vastly different flight principles from birds.

How do humans think? How do people actually think? This is itself a complex technical and philosophical issue to understand the way humans think. Philosophers attempt to find the logical laws of human thinking through reflection and speculation, while scientists use psychological and biological experiments to understand the laws of physical and mental changes in human thinking.

These two paths have played an extremely important role in the development history of artificial intelligence. In other words, logic is the highest criterion to determine whether a person's thinking process is rational. From the ancient Greek sages, formal logic, mathematical logic, language logic, cognitive logic and other branches have summed up a large number of regular rules in the course of thousands of years of accumulation and development, And successfully provided methodological guidance for almost all scientific research.

It is the greatest pursuit of many early AI researchers to let AI programs in computers follow the basic laws of logic to perform operations, induction or deduction. Dendral, the first expert system program in the world, is an example of successfully solving a problem in a specific field with human expert knowledge and logical reasoning rules.

This is a program written by Stanford University researchers in Lisp language to help organic chemists infer unknown organic molecular structures based on material spectra. The Dendral project achieved remarkable success in the mid-1960s, deriving a large number of intelligent programs () to infer material structures based on material spectra.

Dendral) The reason why Dendral can solve problems in a limited field is that it depends on the experience and knowledge accumulated by chemists about what molecular mechanism may produce what spectrum, and on the success of Dendral, which relies on a large number of decision rules in line with human logic reasoning laws, in fact, has driven the extensive application of expert systems in various fields related to artificial intelligence, from machine translation to speech recognition, from military decision-making to resource exploration.

For a while, expert system seemed to be synonymous with AI, and its popularity was no less than today's in-depth learning. But people soon discovered the limitations of building AI systems based on human knowledge base and logic rules. An expert system that solves problems in specific and narrow fields is difficult to be extended to a slightly broader knowledge field, let alone to daily life based on world knowledge.

A famous example is the dilemma when people used grammar rules and vocabulary lists to achieve machine translation in the early days. After the Soviet Union launched the world's first satellite in 1957, the U.S. government and military were eager to use machine translation systems to understand the technological trends of the Soviet Union, but the Russian English machine translation system realized by grammar rules and vocabulary lists was full of jokes, Stuart Russell, Peter Norvig, Artistic Intelligence: A Modern Approach, Third edition.

In the face of emerging statistical models, machine learning, and other technologies, expert systems have no advantage. Since the 1990s, they have been neglected by research institutions and even had to lay off outdated linguists to keep up with technological progress. On the other hand, from the perspectives of psychology and biology, scientists have attempted to understand how the human brain works and hope to construct computer programs based on the principles of brain operation, Realize 'true' artificial intelligence.

This road is also full of thorns and the most ups and downs of examples. Biologists and psychologists have been studying the workings of the human brain for a long time, and the most important part is the processing and propagation of information (stimuli) by brain neurons. Long before the emergence of general electronic computers, scientists had proposed hypothetical models of neurons processing information, namely, The large number of neurons in the human brain collectively form a collaborative network structure. Information (stimuli) is processed through several layers of neurons to enhance, attenuate, or mask, and serves as the output signal of the system to control the human body's response (actions) to environmental stimuli.

In the 1950s, early artificial intelligence researchers used neural networks for pattern recognition, using computer algorithms to simulate the processing process of input signals by neurons, and correcting algorithm parameters based on the output results obtained after passing through multiple layers of neurons. Early neural network technology did not develop for too long and fell into a trough. There were two main reasons for this. Firstly, artificial neural network algorithms at that time had inherent limitations in dealing with certain specific problems, There is an urgent need for theoretical breakthroughs. Secondly, the computing power of computers at that time could not meet the needs of artificial neural networks.

Theoretical challenges of artificial neural networks from the 1970s to the 1980s得到解决1990 年代开始，随着计算机运算能力的飞速发展，神经网络在人工智能领域重新变成研究热点但直到 2010 年前后，支持深度神经网络的计算机集群才开始得到广泛应用，供深度学习系统训练使用的大规模数据集也越来越多。

神经网络这一仿生学概念在人工智能的新一轮复兴中，真正扮演了至关重要的核心角色客观地说，神经网络到底在多大程度上精确反映了人类大脑的工作方式，这仍然存在争议在仿生学的道路上，最本质的问题是，人类至今对大脑如何实现学习、记忆、归纳、推理等思维过程的机理还缺乏认识，况且，我们并不知道，到底要在哪一个层面（大脑各功能区相互作用的层面？细胞之间交换化学物质和电信号的层面？还是分子和原子运动的层面？）真实模拟人脑的运作，才能制造出可以匹敌人类智慧的智能机器。

定义三：AI 就是与人类行为相似的计算机程序和仿生学派强调对人脑的研究与模仿不同，实用主义者从不觉得人工智能的实现必须遵循什么规则或理论框架“黑猫白猫，逮住耗子就是好猫”在人工智能的语境下，这句话可以被改造成，“简单程序，复杂程序，聪明管用就是好程序。

”也就是说，无论计算机以何种方式实现某一功能，只要该功能表现得与人在类似环境下的行为相似，就可以说，这个计算机程序拥有了在该领域内的人工智能这一定义从近似于人类行为的最终结果出发，忽视达到这一结果的手段。

另一种对人工智能的近似定义则更强调人工智能的实用色彩：AI 就是可以解决问题并获得最大收益的计算机程序略懂些编程的人都知道，几乎所有程序设计语言都提供了类似 if ... else ... 的分支结构，那么，与 if ... else ... 相关的一个哲学问题是，程序根据某个条件进行判断并完成相应操作的时候，这个“判断”以及随后的“决定”是由计算机自己做出的，还是由编程序的人做出的？如果是由计算机自己做出的，那能不能说所有执行了 if ... else ... 语句的计算机程序都是人工智能？如果相反，那计算机根据运行时的情况做决策时，人又在哪里呢？

哲学思辨容易陷入这样的两难境地，但实用主义者根本不把这当回事——执行 if ... else ... 的程序是否有智能，完全要看那个程序是不是做了和人相似的有智能的事像 Dendral 这样的专家系统就是靠大量 if ... else ... 来模仿人类专家的判定规则，这当然属于人工智能的范畴，而普通的数值计算程序即便用了 if ... else ...，也不能被称作智能。

实用主义者推崇备至的一个例子是麻省理工学院于 1964 到 1966 年间开发的“智能”聊天程序 ELIZA那个程序看上去就像一个有无穷耐心的心理医生，可以和无聊的人或需要谈话治疗的精神病人你一句我一句永不停歇地聊下去。

当年， ELIZA 的聊天记录让许多人不敢相信自己的眼睛可事实上，ELIZA 所做的，不过是在用户输入的句子里，找到一些预先定义好的关键词，然后根据关键词从预定的回答中选择一句，或者简单将用户的输入做了人称替换后，再次输出，就像心理医生重复病人的话那样。

ELIZA 心里只有词表和映射规则，它才不懂用户说的话是什么意思呢这种实用主义的思想在今天仍有很强的现实意义比如今天的深度学习模型在处理机器翻译、语音识别、主题抽取等自然语言相关的问题时，基本上都是将输入的文句看成由音素、音节、字或词组成的信号序列，然后将这些信号一股脑塞进深度神经网络里进行训练。

深度神经网络内部，每层神经元的输入输出信号可能相当复杂，复杂到编程者并不一定清楚这些中间信号在自然语言中的真实含义，但没有关系，只要整个模型的最终输出满足要求，这样的深度学习算法就可以工作得很好在研究者看来，深度学习模型是不是真的跟人类大脑神经元理解自然语言的过程类似，这一点儿都不重要，重要的是，整个模型可以聪明地工作，最终结果看起来就像人做的一样。

定义四：AI 就是会学习的计算机程序没有哪个完美主义者会喜欢这个定义这一定义几乎将人工智能与机器学习等同了起来但这的确是最近这波人工智能热潮里，人工智能在许多人眼中的真实模样谁让深度学习一枝独秀，几乎垄断了人工智能领域里所有流行的技术方向呢？。

1980 到 1990 年代，人们还在专家系统和统计模型之间摇摆不定，机器学习固守着自己在数据挖掘领域的牢固阵地远远观望短短十几年过去，从 2000 到 2010 年，机器学习开始逐渐爆发出惊人的威力，并最早在计算机视觉领域实现了惊人的突破。

2010 年至今，使用深度学习模型的图像算法在 ImageNet 竞赛中显著降低了对象识别、定位的错误率，领先的算法已经达到了比人眼更高的识别准确率（ImageNet）2015 年，语音识别依靠深度学习获得了大约 49% 的性能提升（。

http://googleresearch.blogspot.ch/2015/09/google-voice-search-faster-and-more.html）机器翻译、机器写作等领域也在同一时期逐渐被深度学习渗透，并由此获得了大幅改进。

“无学习，不AI”，这几乎成了人工智能研究在今天的核心指导思想许多研究者更愿意将自己称为机器学习专家，而非泛泛的人工智能专家Google 的 AlphaGo 因为学习了大量专业棋手棋谱，然后又从自我对弈中持续学习和提高，因此才有了战胜人类世界冠军的本钱。

微软的小冰因为学习了大量互联网上的流行语料，才能用既时尚、又活泼的聊天方式与用户交流媒体上，被宣传为人工智能的典型应用大多都拥有深度学习的技术基础，是计算机从大量数据资料中通过自我学习掌握经验模型的结果。

这一定义似乎也符合人类认知的特点——没有哪个人是不需要学习，从小就懂得所有事情的人的智慧离不开长大成人过程里的不间断学习因此，今天最典型的人工智能系统通过学习大量数据训练经验模型的方法，其实可以被看成是模拟了人类学习和成长的全过程。

如果说人工智能未来可以突破到强人工智能甚至超人工智能的层次，那从逻辑上说，在所有人工智能技术中，机器学习最有可能扮演核心推动者的角色当然，机器目前的主流学习方法和人类的学习还存在很大的差别举个最简单的例子：目前的计算机视觉系统在看过数百万张或更多自行车的照片后，很容易辨别出什么是自行车，什么不是自行车，这种需要大量训练照片的学习方式看上去还比较笨拙。

反观人类，给一个三四岁的小孩子看一辆自行车之后，再见到哪怕外观完全不同的自行车，小孩子也十有八九能做出那是一辆自行车的判断也就是说，人类的学习过程往往不需要大规模的训练数据这一差别给人类带来的优势是全方位的。

面对繁纷复杂的世界知识，人类可以用自己卓越的抽象能力，仅凭少数个例，就归纳出可以举一反三的规则、原理甚至更高层次上的思维模式、哲学内涵等等最近，尽管研究者提出了迁移学习等新的解决方案，但从总体上说，计算机的学习水平还远远达不到人类的境界。

如果人工智能是一种会学习的机器，那未来需要着重提高的，就是让机器在学习时的抽象或归纳能力向人类看齐定义五：AI 就是根据对环境的感知，做出合理的行动，并获得最大收益的计算机程序针对人工智能，不同的定义将人们导向不同的研究或认知方向，不同的理解分别适用于不同的人群和语境。

如果非要调和所有看上去合理的定义，我们得到的也许就只是一个全面但过于笼统、模糊的概念维基百科的人工智能词条采用的是 Stuart Russell 与 Peter Norvig 的定义（Stuart Russell, Peter Norvig, Artificial Intelligence: A Modern Approach, Third edition），他们认为，人工智能是有关“智能主体（Intelligent agent）的研究与设计”的学问，而“智能主体是指一个可以观察周遭环境并作出行动以达致目标的系统”（

https://zh.wikipedia.org/wiki/人工智能）坦率地说，这个定义将上面几个实用主义的定义都涵盖了进去，既强调人工智能可以根据环境感知做出主动反应，又强调人工智能所做出的反应必须达致目标，同时，不再强调人工智能对人类思维方式，或人类总结的思维法则（逻辑学规律）的模仿。

基本上，偏重实证是近来人工智能研究者的主流倾向如前所述，在今天这个结果至上的时代里，没有多少人愿意花心思推敲人工智能到底该如何定义有那个时间，还不如去跑几个深度学习的新模型，发几篇深度学习新算法的论文来得合算。