Datawhale dry goods
Author: Wang Peng, an expert at Tencent Research Institute
Source: Shuo Shuo Vocational Education
In history, every time technology replaces human beings, it provides more and better new jobs. But we tend to forget the generation or generations that were sacrificed in the process. How and how quickly the education system responds may determine the cost of this replacement.
The "Future City Supplements" series has been written for many years, and only two scenes of education and medical care have been delayed. Other urban systems are extensions of human capabilities, while education and medical care determine the boundaries of human spiritual and physical capabilities. These two, especially the mode of basic education, in a sense almost determine the spatial scale and shape of the community, as well as the layout of basic public facilities.
Since ChatGPT became popular, its impact on educational concepts and methods has aroused widespread discussion, and it seems that people can see some trends in future education under the influence of artificial intelligence. Although the educational impact of AI does not seem to be a space issue in the short term, only by knowing the boundaries and capabilities of AI can we know what irreplaceable things can be left in the physical space.
What will ChatGPT and AIGC be able to do?
First of all, what is the ability to distinguish ChatGPT? What trends does it illustrate? What needs to be clarified is that the ChatGPT we are currently trying is limited in its ability to update data in real time, and it obviously seals the domain and scale of many questions that can be answered. Many criticisms of it, such as inaccuracy, poor mathematics, lack of opinion, and often serious nonsense, will quickly evolve and improve in larger-scale training and applications, and will also be used in NLP-based models. Add more mathematics and other abilities. It is foreseeable that in a short period of time, large models will have more and more comprehensive and general capabilities. (This article was written when ChatGPT was just released. The current evolution rate is obvious to all, and these problems are improving rapidly.)
AIGC tools represented by Vincent graph and ChatGPT are transforming the production method of knowledge and content from brain thinking to machine generation + brain screening. From text, image, video to 3D space, AIGC can automatically generate creative content such as articles, music, film and television works, design, games, etc. by simulating the human creative process, and even control robots, and a large number of new formats are emerging and patterns. In this process, the substitution of people cannot be avoided.
In history, mankind has experienced such a situation several times, but each time it survived. When industrialization came, most farmers lost their land and they found jobs in factories. When mass production came, most workers lost their jobs, but they found new jobs in the service industry instead. Each previous replacement has provided more and better new jobs. However, this time machines look set to replace the vast majority of jobs in all fields, mental and physical, in the short term, and we don't yet see where these hundreds of millions of new jobs will come from. . What's more , every previous industry update has cost at least a generation. The premise is that the public education system can respond in a timely manner and cultivate talents who can adapt to new technologies.
ChatGPT has passed exams in law, medicine and other fields with high scores. Of course, this also shows that on a global scale, "rote memorization + limited reasoning" is still the basic orientation of education and talent selection. Cultivating a large number of industrial workers and engineers needed by the industrial society is the main task of our existing education system, and it seems that there will be no changes in a short time. So as an anxious parent, what can you do for your child's future? What public education can't do in time can only be supplemented by family education.
Where is education going?
In the process of large-scale labor replacement, education always plays the most important guiding role. Soon we'll all have a personal assistant who knows everything, so what else do we still need to learn on our own? How will the education system, which was originally devoted to teaching knowledge and skills, be adjusted?
As for what kind of talents the future society needs more and the new requirements for basic education, ChatGPT gave me a well-regulated answer: "It is necessary to cultivate talents with more comprehensive and innovative abilities, rather than simply imparting knowledge. In addition, it is necessary to Pay more attention to educating people to have the ability to adapt to the future society, such as learning ability, creativity, collaboration ability, etc.” However, these seemingly simple keywords actually need to be understood dialectically.
About teaching students in accordance with their aptitude
First of all, AIGC technology will change the way and mode of teaching, making it possible to teach students in accordance with their aptitude. For example, through intelligent content generation and recommendation systems, and even interactive learning tools and games, students can obtain a more personalized and diverse learning experience, and teachers can more efficiently generate teaching materials, test questions and homework, etc. Thereby improving teaching efficiency and quality. In fact, combined with XR technology, each student can receive a completely different personalized education. At this time, the meaning of the school's centralized imparting of knowledge disappears, and communication and collaboration with people may become the meaning of the school's physical space. However, such a school can be completely surrounded by nature.
About Programming and Math
Since it seems that software is what replaces human beings, it is easy to think of letting children learn programming and become artificial intelligence engineers, but programming and artificial intelligence are actually two different things. Encoding seems to be the first thing to be replaced.
Although the training and debugging of large models is still a high-cost technical work, in the future we will be able to communicate with MaaS (Model as a Service) through multi-modal API interfaces, and directly obtain personalized results through voice or even brain-computer interfaces. service without the need for a coding process. The current GPT, due to digesting the massive code of GitHub, already has the ability of basic coding. Copilot X, driven by GPT-4, already has functions such as dialogue, text generation code, speech generation code, automatic code bug repair, and code interpretation. With the help of ChatGPT, I am starting to learn to use Python to generate code to complete some simple tasks. Although there are still many small problems for beginners, he can take the trouble to help you explain error messages and help you modify and debug.
So from this point of view, the ability to program is actually the ability to describe requirements in language. Of course, whether it is a Vincent diagram or controlling a robot, the core of human-computer interaction returns to the language of human-to-human communication .
But beyond everyday applications, the essence of algorithms is mathematics. After all, in every AI paper, the largest space is the mathematical formula and derivation process. So if you want to collaborate with AI, you have to learn language well, but if you want to define and control AI, I'm afraid you still need to learn mathematics and logical thinking. Only logical thinking can make people make judgments and even criticize from massive amounts of information, discarding the chaff and preserving the chaff.
about language
Philosophy and linguistics are hard to separate early on, so language learning can mean far more than reading, writing, and translating skills. The ability to describe the world and create the world is about to merge into one in the digital world , which will be explained in another article of mine.
Although the field of AI translation is developing rapidly, real-time mutual translation seems not far away, and many people may think that there is no need to spend a lot of time learning foreign languages. However, in recent exchanges with various AIGCs, it has been found that when dealing with complex situations, some translation tools still need to translate Chinese into English before performing calculations, especially when it comes to subtle image and word modification, the English level is still very limited obvious. Therefore, in the future, the requirements for English proficiency will not only stop at the level of daily translation, but also the ability to communicate across cultures is particularly important .
Not only that, although many domestic companies are using Chinese corpus for model training, Chinese corpus only accounts for 5% of the entire Internet scale, and the authenticity and rigor are relatively poor, so the quality of large Chinese models may be difficult for a long time Beyond English. Therefore, English proficiency and English way of thinking may still be very important for children to engage in innovative research and work in the future, and may even be more important than programming in human-computer interaction.
Considering the control of the robot, many simple tasks in the future can be performed by generating personalized programs through natural language interaction. Of course, if it is a complex task, the difficulty of only using natural language interaction is actually not small. In the process of using Vincent diagram tools to describe architectural images, I will increasingly discover that the ability to fully express design intentions in natural language can even replace the ability to sketch in previous architectural studies.
Recently, a new recruiting position, Prompt Engineer, has appeared in Silicon Valley, which means "prompt engineer". This position does not require writing code, but only needs to find the appropriate prompt words to let AI realize its potential. In other words, no technical background is required, as long as you "can chat", you can also be an engineer. Therefore, high-quality use of natural language to communicate with machines, Chinese and English are still basic abilities . Of course, prompting engineers will not exist for a long time, but will become a basic skill for each of us. After all, it is difficult to see ppt engineers or Excel engineers now. Ultimately, everyone needs to have the ability to accurately describe requirements in order to communicate and collaborate with AI.
About Learning and Innovation
The essence of the large model is a statistical model, which to some extent represents the cognition of society as a whole. For some factual questions, it is easy to get clear answers. However, in most decision-making scenarios, it is actually based on the synthesis of a large amount of information, and ultimately requires human value judgments. ChatGPT often appears as a "master of water", which is often a value judgment that it cannot make for others. .
Our current education system is primarily about imparting knowledge. On the surface, it seems that AI will soon be omniscient in various disciplines. In the greatest sense, this ability is a tool to help us learn and obtain information quickly. If a person only has the ability to deliver AI results, he is obviously also eliminated.
For any field of human knowledge, AI will have an absolute advantage, but in the exploration of new fields, AI can only be used as an assistant. These so-called innovative fields often exist in the interdisciplinary fields of various disciplines. From elementary school to university, the most common thing I do in my spare time is to read books in the library, and I generally like to pick a book from each shelf for extensive reading. In this way, although the learning is miscellaneous but not refined, a relatively complete knowledge framework can be established earlier, and the knowledge of lifelong learning can be properly stored and read in different categories. A basic understanding of multiple disciplines can be said to be the basic literacy for engaging in cross-field innovation. With the help of AI's efficient information synthesis capabilities, we can obtain rapid and extensive learning capabilities, master the basic knowledge and laws of various disciplines, and thus accelerate the process of emerging cross-field innovations.
Through the above analysis, we will find that it seems that an AI that knows everything cannot replace our learning process, but instead puts forward higher requirements for learning other than language and mathematics and even more subjects .
As the book "The Third Wave" states, "Third Wave civilization is the opposite, with many features: decentralized production, scale, renewable energy, evacuated urban population, working from home, producing Own consumption, etc. These activities are so close to the pattern of the first wave society that it seems to go back in time. This historic change represents not a linear extension of industrial society, but a sudden change in direction-often in the form of Reverse development. This transformation should at least be on par with the industrial civilization 300 years ago. At the same time, what we are facing is not just a technological revolution, but the arrival of a new civilization." The education system for training skilled workers in the industrial age , like this era is coming to an end, and many features of future civilizations, including education and people, may be more like the state of the agricultural age.
In the changes brought about by AIGC, perhaps the vast majority of workers will be eliminated, and even the labeling work will be done better by AI in the future. Of course, the greatly developed productivity may still achieve universal basic income. Only a handful of people can and need to work with AI. This may also be an unoptimistic direction for future education: most people lie flat, and a few work harder.
Perhaps different from the current simple way of reducing the burden, mathematics, Chinese, foreign languages, including basic knowledge in more fields, will only have higher requirements for people in the future, but it is not the direction of doing more questions and solving problems, but more. A comprehensive body of knowledge. For solving specific engineering problems in a certain direction, AI will easily learn human skills. But future innovations may only be competent for encyclopedic talents. With the help of AI teaching students in accordance with their aptitude, it may be possible for human beings to achieve this state again, and to use the quantum computing mechanism of the human brain to help AI evolve. It is difficult for the public education system to deal with how children will live with AI in the future, which requires each parent to make their own choices and efforts.
" Like " three times together ↓