Robin Li: AI will not take people's jobs. Stopping innovation is the biggest threat to mankind.

"The most dangerous and unsustainable thing for mankind is to turn off the innovation engine. Not inventing, creating or progressing is the biggest threat to mankind."

In the magical land of Africa, black brothers have all kinds of strange means of transportation, such as the Alto with 18 hands, the Jiefang truck with half the length of the front, and the 28-inch bicycle handle used to pull bananas. These black brothers are Netizens affectionately call him "Oude Biao".

Going forward a few decades, our fathers were not "Ode Biao" who made a living by relying on backward tools. Why are we no longer the same today? Advanced tools release more productivity and create more job opportunities.

Currently, the most advanced tool is AI.

After large models such as ChatGPT and Wen Xinyiyan set off a wave of AI for all, everyone has a question: Will my job be replaced by AI?

"Everyone can see that the current jobs will disappear, but we can't see any new job opportunities will be created. Just like those people 100 or 200 years ago, they can't see the new jobs that will be created later. The chances are the same. I am personally an optimist. I am not worried that large models will lead to fewer job opportunities and worsening of human life. " At the Tianjin Intelligence Conference held on May 18, Robin Li gave Answer.

AI will definitely make some people look like "Ode Biao pulling bananas", but it will also allow more people to get on the fast train of new technologies and bring about new growth miracles.

The result of the collision between new tools and old tools has been verified countless times by history.

When cars replaced horse-drawn carriages, carriage drivers lost their jobs in large numbers, but the replacement profession of drivers emerged. In addition, the huge automobile industry chain also created hundreds of millions of jobs: In China, there are 3,000 Thousands of people are engaged in automobile-related jobs.

If cars were not popularized, could horse-drawn carriage-related jobs support the employment of 30 million Chinese people? If nothing else, manpower in the manufacturing process will be completely eliminated by machines, just like no Chinese company today hires workers to grind embroidery needles, but what about abroad? "Oudebiao the Banana Puller" is the answer.

Looking at the world, backward productivity has always existed. In the future, industries and occupations that will be eliminated by AI will still be popular in countries and regions with backward productivity, because This is a Ferrari coexisting with a 28-year-old. world.

When more advanced tools are born, they will be used regardless of whether they will cause unemployment for some people or affect the interests of some people.

In developed European and American countries, card payment is still the mainstream. This is simply a joke in the eyes of the Chinese people. However, when card payment has become a way of life in Europe and the United States, our country is still using the original method of banknotes and IOUs on a large scale. , it’s just that China gave birth to and popularized advanced tools such as mobile payment, allowing us to overtake in corners.

In this AI competition, can we overtake in corners again? Baidu has a heavy burden.

Baidu is the first company among the world's major manufacturers to make generative AI products. On March 16, Baidu Wenxinyiyan officially launched the invitation-only beta test. There are already 2 million users queuing up and hundreds of thousands of companies. Enterprise application test Wen Xinyiyan.

At the same time, Baidu is also the only company in the world that has a full-stack layout at the chip layer, framework layer, model layer, and application layer. The full-stack layout can achieve layer-by-layer feedback and end-to-end optimization, which means Wen Xin Yi Yan naturally has great "acceleration". More than a month after its release, Wen Xin Yi Yan's efficiency has increased by nearly 10 times.

At the market level, the advantages brought by the full-stack layout have begun to appear.

Not long ago, Baidu Smart Cloud showed the world's first one-stop enterprise-level large model production platform "Wenxin Qianfan Large Model Platform" that is under internal testing to hundreds of corporate partners. As of now, has More than 300 ecological partners participated in the internal testing of Wenxinyiyan, and achieved test results in more than 400 internal enterprise scenarios.

For a country that is developing rapidly, industry cannot be static, and employment must be dynamically balanced. We should be aware of this from the moment the iron rice bowl was broken in the 1990s. There will always be people who will become the "banana puller" when new tools come, whether the tool is a steam engine, an internal combustion engine, a computer or AI, but on the other hand, new tools will create new industry opportunities and new industries. Opportunities can create many jobs.

At the meeting, the data presented by Robin Li showed that: the first industrial revolution brought about the first wave of growth in global population and per capita GDP; the second scientific and technological revolution grew even faster: in 80 years, the world’s population grew In the past 60 years, the world's per capita GDP has increased by 3.2 times; in the 30 years since the third technological revolution, the world's per capita GDP has increased by 2.8 times.

"Artificial intelligence is a major wave comparable to previous industrial revolutions and will definitely create the next growth miracle of the global economy." Robin Li gave a firm prediction.

There is a qualitative gap between economic and employment growth caused by advanced tools and backward tools. This may be why my country continues to emphasize "optimization and upgrading of industrial structure" in supply-side reform.

Most of the results brought about by AI are dynamic balance and even dynamic growth: on the one hand, the development, maintenance, and update of AI products will inevitably require a large number of professional positions. This change is like the automobile industry that replaced horse-drawn carriages; on the other hand, when After AI is fully popularized, as long as the ideas are active and the expressions are clear, machines can be transformed into personal productivity, which means that many "populations" will become "talents" because of AI. To some extent, AI is a "talent dividend" "Added a lever.

"For human beings, the biggest danger and the greatest unsustainability is not the uncertainty caused by innovation. On the contrary, we stop innovating, do not invent, create or make progress, and continue to follow inertia. The various unpredictable risks it brings are the greatest threat to mankind. " Robin Li finally said.

The following is the transcript of Robin Li’s speech:

Dear leaders and distinguished guests: The topic I bring to you today is called "Large Models Change Artificial Intelligence". Gong Ke also said just now that artificial intelligence has received much higher attention than before in the past six months, mainly because of the emergence of generative artificial intelligence. The underlying technology of generative artificial intelligence is actually a large model.

So why will large models change artificial intelligence? It is big computing power, big models, and big data that lead to the emergence of intelligence. What is the emergence of intelligence? In the past, artificial intelligence taught the machine whatever skills I wanted it to learn. Those who have been taught may be able to do it, but those who have not been taught may not be able to. After the so-called emergence of intelligence occurs in large models, it will be able to learn skills that have not been taught before. This is why some people say that we are now moving towards general artificial intelligence.

At the same time, the development direction of artificial intelligence is moving from discriminative to generative. What is discrimination? The applications of artificial intelligence that we were familiar with in the past were basically discriminative. For example, face recognition, when a person comes over, I identify who this person is or who he or she is not. This is a typical identification type. Search engines are also typical discriminative artificial intelligence. The user enters a keyword or a paragraph, and we match what he is looking for across the entire network to determine which web page and which piece of content he needs. This is discriminative artificial intelligence.

What is generative artificial intelligence? Today I want to write an application. Please write it for me. In other words, I am entertaining 10 people on the weekend, and I need to come up with a menu. This does not matter whether it is right or wrong, but it can give you some feelings, give you some creativity, and give you a good foundation to develop your ideas. In other words, paint me a picture of busy traffic. This is something that people didn’t think artificial intelligence should do in the past, but now it can do it.

So what will this lead to? As a result, people's work efficiency has been greatly improved. For example, its efficiency will be greatly improved in tasks such as content creation, customer service, and translation. Therefore, we have also seen that many research institutions believe that in the next 10 years, the efficiency of many tasks will be doubled. At the same time, it also brings some worries. Will such efficiency improvement make many people's jobs disappear? After these people lose their jobs, will it cause unpredictable problems to us humans?

Actually, I have talked about this for a long time. The best way to find out the answer is actually to look back on the past. Because many people also believe that artificial intelligence is a symbol of the fourth industrial revolution. We can look at what jobs have been replaced by previous industrial revolutions?

200 years ago, the work of carrying water from underground wells basically disappeared. The first industrial revolution was the invention of the steam engine. The first application of the invention of the steam engine was water for mining. How could it be brought up using machines? What has happened with the disappearance of jobs like carrying water underground? If we look at it, a lot of new jobs are actually being created.

In the past 100 years, we have seen rapid growth in the world's population, which is much faster than the previous one or two thousand years of population growth. At the same time, GDP per capita is growing almost as fast. What does this mean? It shows that although some job opportunities have disappeared, more opportunities have appeared. The improvement of people's work efficiency can feed more people, and everyone's life becomes better than before. The job of carrying water from the well is really not a good job.

100 years ago, the coachman’s job disappeared. This picture is a picture of Fifth Avenue in New York in 1913. This picture is almost entirely filled with cars and there is only one carriage. At that time, New York was full of horse meat sellers, and horses were useless. Later, many new jobs appeared. From 1900 to 1990, which was close to 90 years ago, following the same pattern, the world's population continued to grow at a high speed, and the value created by each person continued to grow at a high speed.

Thirty years ago, our generation experienced it personally, that is, around the time I graduated from college, what jobs disappeared? The typist's job disappeared. Nowadays, the younger generation has never seen a typewriter, but many new jobs have emerged. Following the same pattern, the world's population continues to grow at a high speed, and per capita GDP continues to grow at a high speed.

Although history does not repeat itself, it does have its rules. So this time, why are so many people worried that AI will reduce job opportunities? I think it’s because everyone can see that current jobs will disappear, but we can’t see any new job opportunities being created. Just like those people 100 or 200 years ago, they could not see the new job opportunities that would be created later. Personally, I am an optimist. I am not worried that large models will lead to fewer job opportunities and worsening of human life.

How can such a large model redefine artificial intelligence? Chairman Wan Gang also said just now that the way human-computer interaction has changed. In fact, with the development of the information industry in the past few decades, the way human-computer interaction has changed three times.

We won’t talk about it earlier. The command line was my main work interface when I was in college and graduate school. Human-computer interaction is through the command line. I type a command and it gives me the desired reaction. I thought this thing was very efficient, but most people wouldn't do it this way.

What is a simpler way of human-computer interaction? Is a graphical user interface (GUI). At least many people can understand this one, and it is friendlier than the first one. But it's still not the most natural way to interact. If I want to figure out how to reset the computer's automatic sleep time, I have to go through four levels of menus, one by one, to find this location. How many people can remember where every function is after the fourth level menu?

The birth of artificial intelligence allows us to interact with computers using natural language. When I have a need, for example, I want to check last month, April 2023, for each product line of my company, which products have a gross profit margin that exceeds the pre-epidemic level? Such a subject would have probably taken my assistant half a day and a day to obtain in the past. Today, if a computer understands your natural language, it can give you a table in a second.

Big models will also redefine marketing and customer service. In fact, the reason is very simple, that is, whoever has the best way to communicate with customers will have this customer. This reason is not because of the emergence of AI or the emergence of large models. It is just that technology has made us realize many more possibilities than before. Even if you have 7 billion customers today, each of your customers can have a dedicated 24/7 assistant who knows everything to serve him.

The large model is a foundation, and there will be various artificial intelligence applications on top of the large model. What has been a hot topic recently is what do native applications in the AI ​​era look like? Let me give you a few examples:

For example, DoNotPay, what is this? A typical application scenario is artificial intelligence lawyers. For example, if you speed in the United States, the police will give you a ticket after speeding, and you will usually pay a few hundred dollars. In fact, you don’t have to pay it. If you hire a lawyer to help you with the lawsuit, you don’t have to pay it. But the cost of hiring a lawyer may be twice as much as the fine, so you don’t hire one. If you hire AI as a lawyer today, you don’t have to pay the money.

Jasper is a tool for generating marketing ideas. It can help you create whatever ideas your company wants, so it is much more efficient.

Speak is a Korean software application that actually teaches you to learn foreign languages. It simulates various scenarios. You have to go to a restaurant to order food, negotiate with the other party, and how you want to interact with the other party. It can be done in hundreds of languages. very good.

For Baidu, our big model is called Wen Xinyiyan, which was released two months ago. It should be said that it is the first to be released among major manufacturers in the world. The reason why we want to release it as soon as possible is because the market has very strong demand. Currently, there are more than 2 million users waiting in line for testing, and hundreds of thousands of companies want to access Wenxinyiyan for testing.

Of course, Baidu’s investment in this area did not actually just start, it did not start in the past six months. It has been four years since we released 1.0 of the Wenxin large model in 2019. Earlier, we started investing in artificial intelligence around 2013.

The reason why artificial intelligence has undergone such big changes is not only the changes in its application scenarios, but also the technology stack behind it has also undergone very fundamental changes.

The IT technology stack that each of us is familiar with is these three layers. The bottom layer is the chip layer. Typical companies are Intel, AMD, and Qualcomm. Their chips are called CPUs; the middle layer is the operating system, which in the PC era was Windows. , in the mobile phone era, it was Android and iOS; above it was the application layer. In the PC era, everyone developed software for Windows, and in the mobile era, everyone developed applications for Android and iOS.

The arrival of today’s artificial intelligence era has changed this pattern. The current IT technology stack has become four layers. The bottom layer is still the chip layer, but the main chip is no longer the CPU, but represented by the GPU, a new generation of chips suitable for parallel large-scale floating point operations. The above is called the framework layer, which is the deep learning framework. Baidu's PaddlePaddle, Meta's PyTorch, and Google's TensorFlow are all in this layer. The next layer is the model layer. Today's ChatGPT, Wen Xinyiyan, etc., these belong to the model layer. In the future, native applications in the AI ​​era will be developed based on large models.

Baidu has a clear advantage in this regard. We have leading products in each of the four layers. For example, the chip layer has Kunlun core. We have been doing this for ten years. It was first used for search applications. We bought other people's chips. After all, it was too expensive, so we developed our own. The frame layer is the flying paddle, which today has the largest market share in China. At the model layer is the Wenxin large model. In fact, in addition to Wenxin Yiyan, which is benchmarked against ChatGPT, we also have many industry large models, such as transportation large models, energy large models, etc. As for the application layer, Baidu search and other applications are among our leading applications.

These four layers all have relatively leading products or technologies. What are the benefits? That is, you can carry out end-to-end optimization, and each layer can give feedback to other layers. Based on this feedback, you can comprehensively and overall consider how to optimize.

Baidu’s layout at the chip level is Kunlun core. We have two generations of products and tens of thousands of pieces deployed, both inside and outside the company. The third generation of Kunlun core will be launched early next year.

At the framework level, Feipiao’s framework has more than 5 million developers in the field of artificial intelligence in China, and it is increasingly recognized by everyone.

The model layer just talked about what Wen Xin said, and it has many applications in various usage scenarios. In the future, I believe that more applications will be developed based on the Wenxin large model.

At the application layer, everyone is more familiar with Baidu’s search. Baidu cannot do everything at the application layer. In addition to search, we have also made traffic-related applications. Traffic is also very complex and affects a very wide range of directions.

Baidu actually mainly does two things, one is autonomous driving, or driverless technology, and the other is intelligent transportation. We have been working on unmanned driving for ten years, and now we can carry out commercial unmanned operations in Wuhan and Chongqing. We have been working on smart transportation for several years, and the results have been proven in many cities. By intelligently adjusting the time when traffic lights change, the traffic efficiency of our cities can be significantly improved by 15% to 30%. On the last working day before the May Day holiday, many people also noticed that there was a huge traffic jam in Beijing. From the second ring road to the fifth ring road, and even the sixth ring road were all red. The only green one was Yizhuang. Yizhuang has artificial intelligence for intelligent transportation. The system can dynamically adjust traffic flow, so efficiency has indeed been proven to be significantly improved.

We also look forward to the future where various applications from all walks of life will be able to find good application scenarios based on large models, and achieve significant improvements in efficiency.

Finally, I would like to say that for mankind, the greatest danger and greatest unsustainability is not the uncertainty caused by innovation. On the contrary, if we stop innovating, do not invent, create or make progress, and continue to follow inertia, the various unpredictable risks brought about are the greatest threats to mankind. This is why Baidu has continued to invest in artificial intelligence for so many years, and why we defined Baidu's mission many years ago as "using technology to make the complex world simpler."

Thanks!

The above is the detailed content of Robin Li: AI will not take people's jobs. Stopping innovation is the biggest threat to mankind.. For more information, please follow other related articles on the PHP Chinese website!