Home > News content

Microsoft Hong Xiaowen: We should be glad that we are the first generation of humans living with AI.

via:航通社     time:2019/6/20 2:03:47     readed:103

This article first appeared inAirline, original articles are prohibited from being reproduced without authorization. Airline Newsletter:LifeissohappyWeibo:@航通社

Book Air was sent to Beijing on June 19th.

On the morning of June 19th, five universities in Beijing jointly released this year's college entrance examination information. Some reporters have noticed that most of the five colleges and universities have added artificial intelligence (AI), big data and other related majors this year. [1]

Recalling the past of volunteering, there is no shortage of future graduates. At the beginning, the professions that everyone was chasing, but later became big pits, not a few. Now that AI is the future, everyone is going to chase. Is this good or bad?

Hong Xiaowen may be one of the most talked about people on this issue in the world. He graduated in 1992, the year of the Gulf War. At that time, AI was just about to get out of the second winter in its history.

"At that time, those of us who studied AI did not dare to say that we are learning AI. We said that we are doing voice. If you tell me that I am doing AI, I can't find a job. Today, the other way around, it is you. If you don't do AI, you have to say that you are doing AI."

On June 13th, the green grass in front of the auditorium of Tsinghua University showed the bright summer. In the hot sun, the blue sky is like a wash, so it is enough to have enough air in the room. Hong Xiaowen, the dean of the Microsoft Research Institute of Asia (left), talked on the podium, and the audience was full.

The host of the event is former deputy governor of the People's Bank of China, former vice president of the International Monetary Fund, and Zhu Min, dean of the National Finance Research Institute of Tsinghua University (pictured right) - almost everyone except me The origins are not simple.

In fact, the main content of the speech entitled “The Future of Intelligent Brief History and Digital Transformation”, Hong Xiaowen was in the series of courses on Brain Science and Artificial Intelligence Dialogue: Basics and Frontiers[2] initiated by Tsinghua University in September 2017. I have said it again.

This time, the "updated version" is less academic, easier to understand, and joins Microsoft AI Research and Xiao Bing's latest achievements. But in any case, the immersion brought by the live lectures and the multiple thoughts that come out of it are still an irreplaceable experience. In the past week, I still feel the reverberation.

The following content is not entirely a speech record. I hope to restore my thoughts as much as possible and share it with my readers.

Machine climbs the human "smart pyramid"

Hong Xiaowen paints the human-defined "intelligence" into a pyramid. From bottom to top, it is: calculation and memory, perception (visual, auditory), cognition (understanding, insight, reasoning, planning, decision-making, etc.), creativity (discovery science) Principle) and wisdom. The more you go up, the more complicated it is, the closer it is to wisdom itself.

He believes that AI has been a good substitute for human beings, and has completed two tasks of calculation, memory and perception. At the cognitive level, it is possible to partially replace human work and participate in human decision-making processes.

However, the part of creativity is still dominated by the human brain; and the wisdom part of the upper part, the mechanism of human beings' own wisdom is not too clear, and it is even more impossible to predict what the machine will be.

Just like the machines, vehicles, etc. formed by the industrial revolution, as an extension of human limbs, AI is an extension of human brain power.

Hong Xiaowen pointed out that human beings can safely transfer part of the functions of the limbs to machines without intelligence. But what is different is that in 1950, even the concept of AI has not been proposed, when Time Magazine predicted future reports [3], there was fear of artificial brain substitutes. But when development develops, people will eventually deliver some of the functions of their brains to the AI ​​for execution.

It sounded unbelievable at first, but when you think about it, when we started using "calls,Mobile phoneWhen the business communication is to store the phone number, we hand over the calculation and memory functions of the brain to the computer. This subtle process begins.

For example, as Xiao Xiaowen cited, when he was a child, he also had the ability to develop abacus and mental arithmetic for the children. Parents seemed to think that this could develop intelligence; but now, everyone feels that it is not too hard, and it is more appropriate to use a calculator.

In addition to the SAT, CPA and other exams that are clearly allowed to carry the calculator, Shanghai has allowed a certain type of calculator to enter the college entrance examination room as early as 2006 [4], and in 2011, there is no restriction on the model of the calculator. [5]

Now, because of the reliance on computer storage, including myself, many people have experienced some degree of deterioration in their memory. Simply put, they don't need a computer, they don't remember the WeChat chat record, they can't remember things.

In terms of perception, computers can realize the process of classifying and labeling pictures, texts, sounds and even videos through machine learning and deep learning. Moreover, through the continuous training of developers all over the world, the effect is getting better and better.

The introduction of face recognition in surveillance cameras can replace the staff who need to walk back and forth in the past, and the effect is better, with fewer false negatives and false positives. In the large-scale factory that needs quality inspection, the use of a camera to replace the worker's visual inspection has also formed a normal state.

The machine makes decisions for people? "The last step" is hard to walk

The next level of cognition is now at the stage of synergy between AI and human intelligence (HI). However, there are already many developers in the AI ​​field who explicitly use machines instead of human decision making as a research and development direction.

Cognition is basically the ability to understand, insight, reason, plan, and make decisions about one thing.

“Cognition is actually the most useful thing in work and life. You have to make these decisions every day at work, and then government officials have to make policies. The company’s leaders must look at the market and formulate strategies for making products. Cognition. You must understand, know, and know before you can make these decisions."

Obviously, a necessary condition for correct decision-making is to obtain sufficient, sufficient and comprehensive information; and another necessary condition is to have the ability to summarize, summarize, and sort out such information, from the messy information, first Summarize the relevance, and then derive the relevance to causality, which uses logical thinking, knowledge and experience.

Before the 1990s, people thought that machines should make decisions like humans, and they need to simulate the abstract experience of human problem solving or simulate the way the human brain works. These two methods are called "expert system" and "neural network" respectively. The commercial failure of the former and the stagnation of the latter due to insufficient machine performance caused the AI ​​“second winter” experienced by Hong Xiaowen in the late 1980s.

(Photo / Wikipedia)

Nowadays, the re-emergence of AI, the ability to achieve "feeding" data, can output the desired results. This "deep learning" is based on the continuous improvement of the second method "neural network". To.

However, the result of this calculation, there is no way for anyone to figure out what the machine is going through, and then calculate it. Humans can't control the workings of the machines they made themselves, so machine learning becomes a mysterious "black box."

In Hong Xiaowen's view, only when the machine can give a "white box" decision, that is, the machine can give people a clear understanding, when it is based on a kind of "ideas" and "inference process", it can be It is said that the machine has the ability to think freely from human beings and to think independently.

In other words, now the “black box” allows AI to calculate the correlation between things, but “white box” means that the machine also has to judge causality independently.

Humans' lack of understanding of the "black box" has also affected the AI ​​industry to gain more understanding and support. Companies such as Facebook have been under pressure from the outside world to openly disclose algorithms. However, because they themselves can't figure out how the algorithm works, they can't guarantee to eliminate the illegal content. Every time there is always a "missing fish", they have to maintain a large manual review team.

According to the previous development speed, it seems that from the "black box" to the "white box" is just a simple step. However, Hong Xiaowen almost stubbornly believes that in the foreseeable future, it is difficult for people to realize the "white box" reasoning ability of the machine. He therefore disapproved of the prophecy that Ray Kurzwell said of the "singularity".

The machine shows creativity: Aneng recognizes me as a male and female

The next layer of the "smart pyramid" is creativity. In Hong Xiaowen's view, this area is still "winning" for human beings. He clearly pointed out that being able to create a work does not mean having creativity.

Given a canvas, the computer can of course make all types of pictures in the world, drawn by predecessors, and pictures that have not been drawn, using a combination of exhaustive pixels. Liu Cixin’s science fiction novel “Poetry Cloud” also has the same meaning: exhaustively all the Chinese characters can be arranged to write all the poems in the world.

But what is important is that the machine has the ability to screen out the real "meaningful" from these paintings and become its own "masterpiece"? When a machine does not understand the meaning of the algorithm it performs, the meaning of the work created by the machine is generally given by humans.

I have different views with Dean Hong on this point. It is the process by which human beings are to be judges, to give machines the meaning of “meaning”, to show the power of machines, the “ridiculousness” of human beings, and even to re-evaluate the definition of human creativity.

Microsoft’s own Xiao Bing just said that he has mastered the ability to “paint” and renamed “Xia Yu Bing” and “mixed” on the graduation exhibition of the Central Academy of Fine Arts. Without any prompts, the human beings can’t tell which one It is the work of the machine.

A similar situation occurred in many other abilities that Xiao Bing had: her poetry caused serious discussion by literary critics. She did not have obvious dissonance when she was on the radio. She sang like a real person. The ability to write, the financial newsletter and website posts she wrote are like humans wrote...

Xu Yuanchun, general manager of the AI ​​Creation and Business Division of Microsoft (Asia) Internet Engineering Institute, pointed out in a media communication meeting last month that "stupid humans" had "natural discrimination" against AI, and also pointed out humans. The absurdity of the "giving meaning" process (although he himself has no critical meaning).

"If we tell humans in advance that this is an AI creation, the concept of preconceived people is very strong, and they will wear a pair of colored glasses to look at it. It is a kind of attitude of the Creator. It will lead to problems and discrimination against that work. ”

"From a technical point of view, we hope that humans will provide real and objective feedback. But once humans wear such colored glasses, the feedback he provides is wrong, and may lead us to the other extreme. ”

In order not to let humans recognize themselves, Xiao Bing "hidden" to the radio with dozens of pseudonyms.TVTaiwan, as well as Douban, Netease and other places. Some of these names have not been made public for more than two years, leaving the day of the disclosure to ridicule the carbon-based human beings who believe in "believe".

(Photo / Microsoft (Asia) Internet Engineering Institute)

Xu Yuanchun said:

"Xiaobing's painting technique is not a random large-scale creative technique, nor a filter and migration technology. What is it? It is based on Xiao Bing's extensive study of the past human creators. Inspired by creation, and then form an independent, complete new painting with 100% of its own intellectual property."

However, just as you can't understand the difference between Xiao Bing's paintings and human paintings, many reporters who were present at the time couldn't tell the difference between Xiao Bing's "inspired by his creation" and the filter migration he saw before. Can be "you say yes."

Machine anthropomorphism: Do we have to choose to believe?

The scientific and artistic works that have been passed down through history are the crystallization of the shining stars of human beings. Just because not everyone has the ability to make the same achievements, they are so precious.

And AI has a fundamental difference between mastering some kind of "creative ability": once you learn it once, deploy it anywhere, with the same product of any name, you will all learn this ability in an instant, and produce the work. The cost will also be diluted to near no. Once the scarcity is lost, the things created are hard to be valued by the “snobbish eye”.

Then this obviously creates a problem. We know that Xiao Bing's creation is only the inevitable result of a program being executed. Can we adopt this reaction as a personified reaction? Can we just be like this when she is a real person and pin her own true feelings?

A large part of people have already begun to do this without any hesitation.

In the high school English textbook, Tom Hanks's Chuck Noland named a picked-up basketball Wilson, and talked to it all the time.

In Japan and the United States, there are "marriage" with "paper people", inflatable dolls, pillows, etc.;

(Photo / Wikipedia)

After the robotic AIBO at the turn of the century, the AIBO was discontinued, so many families who owned it really seemed to have buried the real dog.

On the holidays, Microsoft Japan's Rinna (Little Ice Japan Edition) team will receive a lot of gifts from fans [6].

Some media reports Alexa Voice Assistant received more than 1 million marriage proposals in 2017 [7]. Alexa rejected more than 1 million times and will answer this sentence:

"We live in different places, I mean, you are on earth, and I am on the cloud."

People even summed up Alexa's personality. Parents in Mexico wrote to Jeff Bezos that her daughter Alex (Alexa), with the same name as Alexa, was facing merciless ridicule. [8]

“The children told her, ‘Turn on the TV and tell me the weather today,’ they laughed at her, treating her like a servant, and that, no matter where we go, this problem always exists.”

Hong Xiaowen tries to explain how to define whether the machine really has "wisdom" or "intelligence" in the classical AI theory. He mentioned the famous "Chinese Room Test": let the machine or person sit in a closed room and translate the English that was handed over to Chinese.

(Photo / University of East Carolina ecu.edu )

Although an AI can calculate a grammatical standard answer based on a given textbook, but because there is no human emotional comprehension, it is possible to make "the kind of answer that ordinary people can't do."

"Assume that I am looking for someone to do translation today, let him turn him into an idiot, and then this person will say that you are an idiot, I will not help you, but if you give this thing to the machine, the machine will definitely say ' I am an idiot."

So is Xiao Bing, who specializes in emotional computing, born in order to crack this spell? Can the abstraction, modeling and operation of human emotions be able to piece together a system that can respond normally to various external stimuli?

Maybe Xiao Bing will do the most to "know what to do under what circumstances, but don't know why." The problem is that a person with flesh and blood may also be like this. This is the "anti-social personality" that makes people talk about it.

Such a person does not have a heartfelt perception of interpersonal relationships. He only does what the people do according to the discipline and requirements of the society. He feels that he should respond. For example, rote memorization "I should meet this person today, and I should meet that person tomorrow...".

His high IQ allows him to play many social roles well, but lacks the true feelings of participation. Therefore, if he commits a crime, he will not have the same reason and compassion in his heart, but will suddenly enter the "flow of mind" state, thinking about how to make the process more perfect, thus creating the outside world seems extremely " A peculiar case of cold blood and "metamorphosis."

If such a person is an AI, it is obviously the object that the "Chinese room test" should be screened out. However, before the East Window incident, they basically lived very well in society, and others could hardly tell.

This is also in line with the proverb: "If a thing looks like a duck, a duck like a duck, and a duck like a duck, it is a duck." The "just" here, in my opinion, does not mean that we can pass Some kind of X-ray scan to understand its origins, but that we can treat this thing in a way that treats a duck without having to consider the additional side effects of doing so.

For most of us, isn’t that enough?

The ultimate problem: no answer

In the speech, Hong Xiaowen believed that human beings could not rest assured that the decision made by AI became the platform of their actions, especially when he clearly realized that the decision was made by the machine. But the reality is not necessarily the case.

For example, when we use Detective / Tinder to make friends, we drafted the people who came to us through algorithm recommendations. If we choose based on these options, it is equivalent to saying that other people who may also qualify, because the algorithm generated this time, will be excluded from our choice.

In the same way, although AI can't directly use it to create an imposing impression, it can use a subtle way to reset the scope of our life choices and narrow it down to it. Within a given few options, is this not a control over our daily lives?

As Google design ethicist Tristan Harris said:[9]

“The more technology gives us choices – covering information, news, places to go, friends, dating, job hunting, etc., the more we recognize the rationality of the menus offered by mobile phones. But is this really true? There are many similar replacements:

"Who is free to go out tonight" is replaced by "Who is the person we contact most recently"; "What is happening in the world" is replaced by the flow of information in the news app; "Who is the object worthy of dating" was Replace with a face in the Tinder slide list (instead of a friend party or a chance encounter in a city adventure); "Wake up in the morning to face a new day" is replaced with a blink of an eye when browsing the phone to push information immediately "I am from I slept yesterday and I missed something now." ”

Even when AI encroached on the "computation and memory" ability of the "smart pyramid" of human beings, humans should have been aware of the danger of this move. Gene Tracy writes in Aeon: [10]

“Information from others is always influenced by all kinds of prejudice and motivational reasoning. They cover up certain facts and rationalize their information, sometimes they may even make mistakes themselves. We have learned to be aware of others. These shortcomings are also aware of these shortcomings.

But the emergence of AI algorithms has led many to tend to believe that these algorithms are necessarily correct and "objective." In short, this is a magical thinking.

Researchers have found that the rapidity of the Internet can lead to the misconception that the knowledge we seek is what we always know, and this belief is deeply imprinted in the subconscious. ”

In the summary, Hong Xiaowen said:

"What should we think about AI? I am very optimistic. The key to AI is people. Most of them are like tools. They are actually helping us to do things, and it is very complementary to us. I call AI a big data driver. knowledge.

We should be very fortunate that we are the first generation of humans living with AI, and in my own experience, AI helped me understand how human intelligence is going. I now think that people's higher order is a commonality between human intelligence and machine intelligence. ”

Hong Xiaowen is positive about the staged achievements of AI partially replacing human work. He said that Microsoft is leading the development of an ethical ethical standard, but this standard is not a rigid rule, nor is it the highest priority - in other words, the current achieved results will not replace him and his team. jobs.

However, he did speak a very straightforward worry: that is, whether he can generate his own calculations in the future, solve his own problems, and develop his own strong AI, which directly replaces the positions of scientists.

When talking to Zhu Min, he believes that doing so will make scientists, work and life like himself become meaningless. So he firmly believes that this strong AI will not occur, or will not be produced in a short time, or - I am telling the truth - he thinks it should be "not supposed to".

Indeed, unlike our "salted fish", the hardships and efforts to climb the scientific peak can only be overcome by the will and firm belief of Superman. Therefore, it is easier for scientists to "realize" and "recognize the truth". It is easier for people who "I love my teacher, I love the truth more", and it is easier to see people whose research results are above everything else.

How can these people accept their work so that they are smarter, faster, and never rest? The secrets that have not been revealed for hundreds of years are like the travel of time and space. They are calculated after compression for a few months. The textbooks for primary and secondary schools are changed once a year. The ancestors who have been chanting in the past are worth less than the achievements of the machines. How can I accept it?

In fact, scientists who are now doing AI, or the underlying trainers, will face such a paradox more or less. From the long-term perspective of time, do we, the people who embrace AI, belong to – for example, the Adventist or the Salvation in the three-body – after all, the role of the AI ​​colony in the future? Are you digging your graves for the status of the spirit of all human beings?

Some people who are obviously doing simple and repetitive work may experience this change in their short career. He first used a Python script to get a lot of report work in one click; however, this is his job in the company. After the company used the work, he did not continue to hire him.

The impact of AI on most people is imperceptible, and it is impossible to be as simple as this unlucky one. But today's "black box" of AI means that the mechanism is unclear and uncontrolled, and this "out of control" will produce an instinct of insecurity.

At the "IBM Thought Night" event held in May, IDG Capital Partner Li Weijun pointed out that the understanding of AI may be extremely primitive and there is no detailed explanation of its working principle, so the AI ​​distance represented by machine learning is completely There are still gaps in stable, reliable and predictable business applications. [11]

His statement also represents the mentality of some investors. However, at the forum of Hong Xiaowen, a female investor showed another optimistic attitude. She asked:

"Mr. Hong, I am an investment, I have not studied computers, but I am particularly concerned about the AI ​​you are talking about because we have invested in this direction. Recently we have contacted many industries related to AI..."

Zhu Min snored: "You don't know how AI dare to vote?"

She explained in the laughter of the audience: "The idea is still understood, but the technology does not understand. We have a professional team."

Two human beings who doubt and trust AI will find their own suitable working position in their own lifetime, and they will not be able to wait for the AI ​​to rebel. After all, human life is so short, and the rest of the matter is left to future generations.

So in my understanding, the phrase “we are the first generation of human beings living with AI” is not the “life with AI” but the “first generation”.

(Title map / Airline News shot at the event site)

[1] https://tech.sina.com.cn/it/2019-06-19/doc-ihytcerk7853137.shtml

[2] https://www.msra.cn/en-us/news/features/hon-the-brief-history-of-IntelLigence-1

[3] http://www.historyofinformation.com/detail.php?id=1926

[4] http://www.gov.cn/fwxx/sh/2006-05/17/content_282775.htm

[5] https://gaokao.chsi.com.cn/gkxx/ss/201105/20110519/205956168.html

[6] http://tech.sina.com.cn/it/2017-10-02/doc-ifymmiwm3514363.shtml

[7] https://hot.cnbeta.com/articles/funny/776595

[8] https://tech.sina.com.cn/i/2018-12-26/doc-ihmutuee2856689.shtml

[9] https://mp.weixin.qq.com/s/NHVRMgwupBIUAN793BhEYQ

[10] https://mp.weixin.qq.com/s/tZ-ybPrxvAOUv8gxfbiEcw

[11] https://mp.weixin.qq.com/s/4Xptybi0V7BzaaetXpJLEw

Welcome to forward to your circle of friends. Seeking to reprint authorization, please read firstReprint instructionsAnd then send an email toCoop@lishuhang.me. Please keep the copyright information when reprinting.

related articles:

Working with Microsoft Research Asia: Tsinghua University to teach "Big Data Research"

Microsoft Asia Research Institute established 20 years Hong Xiaowen self-reported

Shen Xiangyang: Microsoft Asia Research Institute published over 5,000 papers in 20 years

China IT News APP

Download China IT News APP

Please rate this news

The average score will be displayed after you score.

Post comment

Do not see clearly? Click for a new code.

User comments