The Real Threat of AI: How Generative Models Are Destroying the Middle Class

I read two articles recently, which are really frightening. One is the ChatGPT replacement employment report made by the National Development Institute of Peking University and Zhaopin Recruitment. The conclusion of the report is that the large language model headed by ChatGPT will replace most white-collar and knowledge-based jobs.

Another article is an article published by the data emperor Chen Qin in Caixin. He analyzed 1,639 occupations in China and found that the occupation with the highest replacement rate by generative AI is translation, followed by insurance underwriters and playwrights. More than 90% of the work content of this profession can be replaced by AI.

Secondly, occupations related to art, graphics, and video, such as visual communication designers , decorative artists, art editors, and advertising designers, have been replaced by AI for more than 80%.

The replaced occupations are not only liberal arts, but now the most profitable programmers in the field of science and engineering also have 75% of their work at risk of being replaced by AI. In extreme cases, companies may lay off about 70% of their programs in the future members, leaving only the most capable and innovative people.

Both articles concluded that this round of generative AI has the greatest impact on white-collar jobs, especially jobs involving “information processing” and “data processing.” It has the least impact on blue-collar jobs and the service industry, among which occupations such as greening, cleaning staff, masseurs, and pastry chefs are almost unaffected.

Scholars who study AI have found that the more difficult the skills learned by human beings, the easier it is to be replaced by AI, while the natural abilities of human beings are not easy for AI to master.

For example, masseurs and cleaners, whose actions are too complicated for machines to imitate, and jobs such as secretaries and managers that require long-term dealings with people, cannot be replaced by AI.

People’s social ability and body movements have evolved over millions of years, and it is difficult for AI to penetrate the secrets. Most of today’s occupations were created after the Industrial Revolution. For these human beings, they are created rather than born AI has shown a strong learning ability.

Those occupations that require high growth are now facing the threat of AI, and these high-growth occupations are precisely the cornerstone of the middle class.

In many sci-fi works, people always imagine that low-level jobs are most likely to be replaced by machines. The emergence of ChatGPT subverts this cognition. The evolution of technology is completely different from what sci-fi writers expected. Until today, the United States has not been able to make a humanoid worker who can replace porters, but GPT can already do as well as clerks, translators, lawyers, accountants, and programmers.

The destruction of the middle class does not seem to be a fantasy now. This process has already happened in the United States.

After the launch of GPT-4, OpenAI published a paper to describe the impact of this technological revolution. They predict that after the popularization of large language models, 80% of the American workforce will be affected, 10% of their jobs will be replaced by AI, and 19% of the workforce will have 50% of their jobs replaced by AI.

They also found that occupations with higher wages were more likely to be replaced, a trend that only reversed after annual income was greater than $100,000. It is precisely the middle class that is affected by the big language model.

At present, there have been two live cases. For example, a data labeler, originally a new profession, only appeared in the “China Occupation Code” in 2022. The main job content is “labeling and processing raw data of pictures, text, voice and other services.”

However, after the emergence of GPT, manual data labelers were quickly replaced. OpenAI’s experiments found that the similarity between GPT’s data labeling results and manual labeling results reached 81%, which is basically the same. The cost of GPT marking is much lower than labor.

GPT-4 labels 40,000 pieces of content, and 100 pieces of content are labeled each time, only 400 times are required. Using the GPT-4 model, each labeling of 100 pieces requires only $0.12. In other words, it only costs 48 US dollars in total, and human data labelers need at least 10,000 yuan a week to complete the labeling of 40,000 pieces of content. GPT only takes half an hour, and the money for a cup of milk tea. The quality of the two is almost the same, and GPT can quickly replace human labelers.

More recently, strikes in Hollywood have also been linked to the application of AI. This is the first simultaneous strike by the Screenwriters Guild of America and the Screen Actors Guild of America in 63 years. The scale of the strike is unprecedented. The core of this strike is that AI affects employment. Screenwriters are unwilling to “feed” their scripts to AI, nor do they want to be reduced to The role of auxiliary AI “correcting homework”. Actors are afraid that they will be replaced by “digital people”, especially the living space of group performers will be greatly squeezed. AI has literally begun to destroy the middle class in America .

Although existing studies have shown that the big language model has not yet impacted China’s job market, this is more of a lag effect. As the country that invented AI, is the most open to AI, and has the least supervision, the United States will naturally be affected the most and the fastest. However, under the advancement of technology, no one can survive alone, and this impact will be transmitted to China sooner or later, which will also have a subversive impact on China’s future.

Talking with media person Zeng Hang this week, he put forward a worst-case scenario, that is, generative AI is applied on a large scale, and it replaces more than half of white-collar jobs, but does not create many new jobs.

Don’t think that technological progress will definitely create new jobs. Often it is not the same technology that creates jobs and destroys jobs, and the performance varies in different time periods.

For example, after the industrial revolution, the rapid development of machines replaced a large number of laborers. Many workers lost their jobs. Workers were forced to smash machines because they could not find jobs. Since the 1990s, the development of artificial intelligence (automation) has also caused many European and American workers to lose their jobs. The life of American blue-collar “one house, two children and two dogs” is shattered.

History also shows that the development of new technologies is more likely to widen the gap between the rich and the poor. As far as the history of the United States is concerned, this is indeed the case. In the 1970s, the middle class accounted for about 70% of the population in the United States, but now it is less than 50%. Wealth is highly concentrated in the top 1%. We have reason to believe that this trend will become more serious in the future.

In the past two years, American Internet companies have begun to lay off employees. Musk cut Twitter employees from 8,500 to 1,500. Musk found that even if 80% of the people were laid off, it would not affect the work at all. This shows that many employees are “Redundant population” does not create value, and the so-called high salary is just a form of corruption. (Ali and other major domestic manufacturers have also begun to learn from Musk)

OpenAI, which created ChatGPT, is also a small team of only 375 people, but the company’s market value is as high as 27 billion US dollars. In the past, a high market value must correspond to a large company, so it has the name “big factory”, but OpenAI has broken this. model, which tells the world that small companies can create miracles, and only a few “super individuals” are needed.

In the future, the mainstream of the company is likely to become a super company supported by a few “super individuals”. The role of AI is to make the strong stronger. In the past, a strong man may need the assistance of a dozen ordinary people, but now he only needs a ChatGPT , AI makes the “super individual” even more powerful, and one person can complete the work of dozens of people. The final result of this model is only “polarization”.

At that time, the society is likely to become a pyramid structure of a small number of elites and the vast majority of the poor, the middle class tends to disappear, the foundation of democracy will be weaker, and there will be a large number of idle “redundant population” in society who can only rely on government relief Survival, they are drunk and dreaming of death, walking towards the end of their lives in the midst of entertainment. Such a world has already seen signs in the United States.

As long as the materials are continuously “fed” to the AI, the AI ​​can continuously learn the skills of white-collar workers. The professional knowledge that people were once proud of may be useless in the future, and those white-collar workers who were once hired with high salaries will also become the new poor class. Just like the artisan class that was eliminated during the industrial revolution, the American blue-collar aristocracy replaced by globalization and automation, and the laid-off workers from state-owned enterprises replaced by marketization.

So who will be popular in the future? I think there will be three kinds of people, one is low-end service personnel whose movements are too complicated for AI to learn. The second is people with innovative capabilities. Such people are a minority in every era, and AI is hard to replace.

The third is “all-rounder”, the kind of person with social skills and management skills. In the context of machines being competent for professional skills, “setting goals and letting machines achieve them” will become the main work content. What it needs is imagination. And insight. At the same time, “the ability to deal with people” will be more important, and the role of emotional intelligence and connections will be more obvious.

The division of labor during the industrial revolution will be eliminated by history, AI will replace a large number of executive layers, and the future will belong to those with “leadership skills”. This path is good at “execution” and “imitation” and lacks individuality. East Asian countries are quite unfavorable.

Under the tide of history, the world will usher in a new round of reshuffle, and everyone will stand the test of the times. For those who are interested, it is time to “prepare early”, learn to use AI, aim at occupations that are not easy to be replaced, and don’t fall behind in this technological change.