Fudan Artificial Intelligence Professor: Which jobs will not be replaced by AI in the next 3-5 years?

What to watch With the rise of ChatGPT, the trend of many jobs being changed or even replaced by artificial intelligence has become increasingly clear. So which professions will be hardest hit? What characteristics do you have to avoid being replaced by artificial intelligence? Listen to the insights of Professor Zhang Junping, an artificial intelligence expert from Fudan University, and see if your job is safe in the future?

The article has been authorized by "Yitiaotv" (ID: yitaotv), please contact the original author for reprinting

Readme丨 Zhang Junping Editor丨Liu Yameng

Editor丨May

On International Labor Day, May 1 this year, the first wave of AI unemployment arrived. The technology giant IBM announced that it would suspend the recruitment of 7,800 people, saying that these jobs would be replaced by AI.

At the end of March, Goldman Sachs Group released a report predicting that 300 million jobs worldwide will be replaced by generative AI, with lawyers and administrators being the most affected.

AI generates pictures of beautiful women, which are fake and real

AI Surreal Creation: Office workers look at goldfish, melon farmers and Trump in the subway

On Chinese websites, because of ChatGPT and Midjourney, the first batch of unemployed designers and copy editors also appeared one after another.

What kind of jobs will be replaced by AI in the next 3-5 years?

Which industries are relatively safe?

What kind of abilities are needed if you want to become an AI engineer?

Can liberal arts students switch to AI?

The article interviewed Professor Zhang Junping, an artificial intelligence expert at Fudan University, and answered the above questions.

Next 3-5 years

Which jobs will be replaced by AI?

Professor Zhang Junping walking on Fudan campus

The emergence of ChatGPT-4 is shocking. We who do AI research knew that something like this would come out sooner or later, but we didn’t expect it to be so fast and the performance would be so good.

Since March, many people in my circle of friends have been posting chat screenshots of ChatGPT-4, very enthusiastically. Together with Midjourney V5, everyone is worried about whether their jobs will be replaced by AI?

Human-machine coexistence scenario, an editorial department generates it through Midjourney

This concern is reasonable.

The most amazing thing about ChatGPT-4 is its "emergent function" , that is, when the amount of data it trains is large enough, this complex system will be born. Attributes that some do not have - is close to human "thinking mode" and "intellectual performance".

There is a thinking chain in it, which helps ChatGPT-4 to "think in a chain". Just like sometimes when we do homework and reach a certain point and can't do it, the parents say "think about it again". In fact, they don't say anything, but the student feels that I may still have some things that I haven't mastered. , through slow thinking and a little guidance, I suddenly came up with a correct answer.

So if you ask ChatGPT-4 to "think again" in the dialog box, it will also give you an improved answer, and everyone will be surprised.

Because AI improves production efficiency, an excellent talent can do a lot of work. The phenomenon of a small number of people running a large market capitalization company may become more and more common in the future. You see Midjourney is a typical example. It has only 11 employees but annual revenue of US$100 million.

AI generates “fire” white-collar workplace

Looking closely, there are two criteria for jobs that are likely to be replaced in the next 3-5 years: Mental work and Simple and easy to repeat. It is true that white-collar workers are more affected.

In my own life, many of the courier calls I receive now are robots. Domestic scientific researchers need to translate their papers into English. In the past, they may have to find foreign native language translators. In the future, they may try ChatGPT-4 translation. It is fast and can limit the proper nouns in the field, which should be very good.

ChatGPT is embedded in Office365,

Can automatically generate briefings and forms

The next most dangerous ones are office clerks, human resources, and those who make financial statements.

Microsoft Office 365 has embedded ChatGPT into Word, PPT and Excel, which can automatically generate presentations, PPT and tables. The value of these Office skills you have spent so much time learning in the past has declined.

There is a joke that "Finance will not be replaced by AI because it cannot be used as a scapegoat." Although it makes sense, the increase in production efficiency means that the company's demand for financial talents is compressed, and your employment space will become smaller.

There is also the lawyer industry. We know that a very important part of a lawyer's work is to be proficient in legal regulations and search for past cases. The search process is very time-consuming, and there should be a dedicated group of people in the law firm to do this work.

If you switch to AI, it will collect all the cases and ChatGPT will give them to you in a conversational way, which is very fast, so the lawyers who used to do this part of the work will no longer be needed.

Programmers are working, an editorial board is generated via Midjourney

ChatGPT-4 will also generate code, Some programmers will be affected, especially the front-end. Because the front-end design is relatively modular, it does not involve very complex calculations. OpenAI has a demo, which is to draw a sketch on paper, and then ChatGPT-4 will run out a web page for you.

From the company's perspective, it is possible that it will be more inclined to write code for ChatGPT in the future. Because everyone has a different style of writing code. If an employee leaves and a new employee comes, he may have to rewrite the code because he is not comfortable with it. Then ChatGPT will be more consistent and more efficient from the company's perspective.

AI generated style illustration

I saw online that some of the illustrators and designers influenced by Midjourney have been laid off. It may take you two days to complete an illustration, but the machine comes out in a few minutes, and the effect is very good. This forces everyone to do more innovative work.

An interesting phenomenon is that some AI researchers’ own work has been taken away by AI.

Then we discuss why? In the past, scientific research institutions would produce some results in 3-5 years, and so many people were needed to do the fine details. But after ChatGPT-4 came out, it solved many problems, and the rest are very hard to crack. , then you don’t need so many teaching positions, which will lead to some positions being eliminated.

Which industries will be safe in the next 10 years?

Can liberal arts students switch to AI?

Craftsmen who make lacquerware

First of all, jobs related to entities, such as doctors, nurses, drivers, and niche craftsmen, such as guqin players and ceramic artists, are relying on personal experience, and are done by AI The probability of substitution is small.

Because AI has been mostly doing cognitive-related tasks, little effort has been put into perception. At this stage, it cannot do well with entities. Compared with humans, manipulators are relatively rudimentary, and it is still very difficult to twist a bottle cap. Difficult things.

Even cleaning is "simple and easy to repeat" for us humans, but it is a vague concept for machines and cannot be programmed or formalized.

So for white-collar jobs, there are still some industries that are relatively safe, which are industries that big data cannot enter.

Stills of "Get Out, Mr. Tumor"

Let’s think about how ChatGPT started? Its data is all Billion level, that is, more than 1 billion level, which means that so much data is likely to have no privacy before it can be used.

If an industry involves privacy, data cannot be disclosed, and model training cannot be used, then AI cannot squeeze in. For example, medical, banking, biology and other fields are relatively safe.

So some of my students will no longer look for jobs in Internet companies, but will go to some fields where data is relatively closed and more stable.

If high school students choose a major and only consider employment prospects, I think the direction of artificial intelligence is currently the best. As the saying goes, "You can't get tiger cubs without entering the tiger's den."

We have a new term called AI for Science, which uses artificial intelligence to help scientific development. In the future, all walks of life will need the assistance of AI, and it must be operated by people who understand the direction of AI, then there will be A very big talent gap.

AI researcher, an editorial generated via Midjourney

A good AI researcher or engineer needs three basic qualities: Mathematical foundation, programming ability, and English. The reason for learning English is to keep track of the most cutting-edge international technologies, read literature, and the requirements for programming skills are higher than mathematics.

Nowadays, you don’t need to have in-depth knowledge of artificial intelligence as before. If you major in computer science or other science and engineering, the threshold for switching to AI is not that high.

First of all, most of the current research is modular, and deep networks are just models, which are built like building blocks. In terms of algorithms, you can quickly know what the latest algorithms look like on ArXiv. There are many websites for the code itself. For example, the code on Github is shared. These three points make it easier for you to enter this industry now.

Liberal arts students also have the opportunity to switch to AI. We have students from the Chinese Department at Fudan who transferred to our natural language processing group and did quite well.

How is the current development of AI in China?

How long will it take to catch up with foreign standards?

Robots work with girls on the farm, an editorial generated via Midjourney

First of all, we really need to catch up, otherwise we will get stuck.

It is said that GPT5 has been trained, so when can we catch up with foreign players? There are currently two factions:

One group is optimistic, thinking that the problem is not big and can catch up in 2-3 months.

The other group is the pessimistic group, which thinks it will take 1 to 1 and a half years.

You may think that one year is not too long, but in fact there are some troubles here.

The current mainstream development path of AI is three blocks: Model, computing power, and big data.

The optimistic part is that the predecessors of the model framework have already done it, and it is almost open to the public. Researchers can just make it bigger and deeper.

Geoffrey Hinton, the father of deep learning

In 2006, Geoffrey Hinton proposed a deep learning model. Later, an image classification competition used the large-scale data set ImageNet. In 2012, Geoffrey Hinton led his students to create a new deep learning model for this competition. , it was shocking all of a sudden. Compared with the previous championship, the performance improved by nearly 10 percentage points.

What is the concept? If you use traditional machine learning methods, it will increase by 0.3-0.4 percentage points every year. This means that deep learning methods are about 20 years faster than traditional machine learning methods. So at that time, everyone turned to deep learning models.

But deep learning models require powerful computing power to run on specific GPU chips.

But our country currently has a bottleneck in computing power, because in December 2022, the United States banned the sale of GPUs above A100 to China. In this way, A100 cannot be used in China (there are alternatives, but the communication module is limited), but cards better than A100 can be used abroad, which is a bit troublesome.

Our research costs are very high now, also because of GPUs. In the past, publishing an article only required time and labor costs, but now the cost of a paper may be around 100,000 yuan.

Another one is big data, and the Chinese corpus cannot be promoted.

ChatGPT has more than 1 billion levels of data for pre-training. It is all in English, but each of our Chinese platforms has an entry threshold to prevent you from searching in a wide range. There are also format issues. This causes us to pile up data, which is not as convenient as abroad.

And there is no open source after ChatGPT-2, and you don’t know where the exact gap is.

Domestic AI investment is very popular now, and capital-level driving is still quite important. Moreover, our Fudan University released a Moss system some time ago, and it is also open source. Relatively speaking, it is still a relatively small model, and everyone is still working hard.

Robots on the streets of Shanghai, generated by an editorial department via Midjourney

Historically speaking, artificial intelligence is less than 90 years old. We generally believe that its beginning was the Turing machine in 1936, during which it has been experiencing ups and downs.

It experienced its first cold winter in the 1970s and 1980s. At that time, if you said you were engaged in artificial intelligence, you would not get a project. In the early 1990s, we experienced a second cold winter.

I have been fond of reading science fiction since I was a child. I came into contact with AI in 1997. At that time, it was more popular to call myself doing machine learning rather than artificial intelligence.

My feeling is that in 2012, the year when Geoffrey Hinton led his students to win the competition, artificial intelligence really took off.

2016 AlphaGo game against Korean Go player Lee Sedol

In 2016, AlphaGo defeated Lee Sedol, and then in 2017, Google developed the Transformer network. After that, there was a series of work on ChatGPT, and various fields such as autonomous driving, AI finance, and AI medical care are making progress.

But in fact, by 2022, the AI ​​industry will be on a downward trend, because everyone feels that everything that should be done has been done, and no good applications have been seen. It is obvious that some large companies are in the field of deep learning. , are already laying off employees. But suddenly ChatGPT-4 came out in March this year, which brought everyone back.

So it has a period of prosperity and a period of decline. I have been in this field for a long time, so I am relatively calm about the craze caused by ChatGPT-4. The research scope of AI is very broad. Many problems are difficult to realize in a short time. Human beings still have a long way to go in understanding intelligence.

As a researcher, the fun lies in exploring the unknown. You can find a little progress in the unknown, and the pleasure is very satisfying.

A short video every day, telling a touching story every day, selecting beautiful things in the world every day, come and live a wonderful life with me every day. One piece (ID:yitaotv)

Focus on the Bund Education

Discover quality education

The above is the detailed content of Fudan Artificial Intelligence Professor: Which jobs will not be replaced by AI in the next 3-5 years?. For more information, please follow other related articles on the PHP Chinese website!