Liu Jia, a professor at Tsinghua University: Five key abilities we need in the era of artificial intelligence

Liu Jia/Text

The core of general education in ancient Greece was to cultivate the political and cultural abilities of the aristocracy, which had nothing to do with the living needs and social functions of ordinary people. Jobs related to daily life, such as farming crops and raising livestock, were exclusive to ordinary citizens and slaves. In ancient Hebrew, the words for "slave" and "work" were the same. During the Roman Empire, the practicality of general education was greatly enhanced - even the speculative nature of general education for the aristocracy was significantly weakened, while practical disciplines such as law, architecture, and rhetoric were vigorously developed.

In modern times, the Industrial Revolution and modernization required a large number of laborers with basic education to meet the needs of the new economic structure and technological development. Therefore, the compulsory education system in Prussia in the 19th century became the model for modern education, aiming to cultivate the work skills of all social classes.

In the era of AGI, when transforming the specialized education that originated in the Industrial Revolution into general education, we should not only inherit the tradition of ancient Greek general education that emphasizes extensive knowledge and rational thinking but also integrate modern science and technology and interdisciplinary thinking. We should emphasize the balance between the practicality and ideological nature of education and incorporate the perspectives of globalization and cultural diversity. Therefore, in my opinion, in modern general education, we need to cultivate the following five abilities:

» Research: Ask the right questions.

» Statistics: Explore the relationships between all things.

» Logic: Deduce the unknown from the known.

» Psychology: Understand yourself and perceive others.

» Rhetoric: Persuade others and lead innovation.

Among the five abilities in the AGI era, "logic" and "rhetoric" come from ancient Greek general education, while "research", "statistics", and "psychology" are the crystallization of the progress of modern science and technology and the development of human civilization.

Research: Ask the right questions

The management expert Peter Drucker once said, "The most serious mistakes are not due to wrong answers but due to asking the wrong questions."

In the AI era, when a computer has learned the rules and assertions of this world, does it then have intelligence? This touches on the most fundamental question: What is the essence of intelligence? Many domestic educators believe that the essence of intelligence is memory. This is why from regular classroom tests to the ultimate college entrance examination, all are exams based on academic knowledge rather than academic abilities - those who can remember the most knowledge and formulas can get high scores and be admitted to good universities.

Connectionist AI researchers represented by Hinton asked a completely different question: "How can we enable a computer to simulate the information - processing mechanism of the human brain to have intelligence?" Hinton believes that "the design inspiration for modern AI comes from the understanding of how the brain works. The brain consists of a network of a large number of brain cells. Input will trigger a series of activities in the neural network, ultimately producing an output. The output result depends on the strength of the connections between brain cells. If these connection strengths are changed, the output result corresponding to each input will also change. Currently, the way AI works is that instead of programming the computer directly, we show it a large number of examples. Through these examples, it will adjust the connection strengths on its own, thus learning to generate the correct answers without our explicit programming." In this passage, the core word is "learn". That is to say, in the eyes of connectionist AI researchers, the essence of intelligence is learning.

Therefore, the development process of artificial neural networks is a process of continuously enhancing the learning ability of AI by imitating the information - processing mechanism of the brain. The first AI with autonomous learning ability was the perceptron proposed by psychologist Frank Rosenblatt in 1958. In this model, Rosenblatt proposed an automatic weight adjustment mechanism by simulating the strengthening process of brain synapses, enabling the artificial neural network to automatically adjust weights to complete learning tasks without manual setting.

Knowledge is only the product of intelligence, while learning is the cause of intelligence.

However, how did Hinton blaze a new trail in the booming wave of symbolic AI and ask a question that was heretical at that time but later proved to be extremely correct? I believe that the ability behind this comes from Hinton's "research" ability. In scientific research, the most crucial step is to ask high - quality questions through literature review and critical thinking. Literature review helps to build a comprehensive knowledge framework and find the blind spots in previous research, while the core of critical thinking is to question existing assumptions, break traditional frameworks, and find the essence of the problem through logical reasoning and systematic analysis, thus fundamentally revealing new research directions.

Interestingly, we can transfer the "research" ability to any other field. For example, when buying a school - district house (literature review: check policy documents to understand whether there is a risk of changes in the local school - district division policy; critical thinking: the return on investment in a school - district house may be lower than diversified education investment), or when seeing a doctor (literature review: check the professional rankings of different hospitals in corresponding departments; critical thinking: whether high - level hospitals tend to over - treat). At this time, literature review means extensive reading, and behind critical thinking is an "equal" mindset.

Sam Altman said, "In terms of human intelligence, the future will not be as important as it is now. AI can make up for human intelligence. In the future, the ability to ask the right questions is more important than the ability to find answers." Therefore, to ask the right questions, we need research ability based on reading and criticism.

Statistics: Explore the relationships between all things

Now, quantitative funds are using AI and big data to analyze the sentiment on Twitter, Reddit, and news websites in real - time and formulate trading strategies by combining market data. The quantitative trading model of "social sentiment + big data" is reshaping the financial market. So, when we see a certain stock becoming a hot topic on social media, remember that a quantitative trading algorithm has discovered this trend before us and is already executing trades.

Through modern technology, we can find patterns in complex and disordered multimodal big data to predict the future. In fact, the ancients also had a similar thinking paradigm. The "Book of Changes", regarded as the "chief of all classics", was their "data analysis tool". The ancients' thinking is similar to modern big data, that is, through complex data (changes), discover the changing laws of things (simplicity), and predict future trends (constancy).

In modern society, from predicting the spread path of epidemics to the dynamic evolution of financial markets and the spread of emotions on social networks, the statistics of big data have become a key method for us to explore the way the world works. More importantly, statistics is not just a "tool" for predicting the future; it is profoundly influencing the way we understand the world and even shaping the future of artificial intelligence.

In addition, due to the emergence of the Internet and the interconnection of all things, data has also undergone a fundamental change. In short, big data has four main characteristics (abbreviated as 4V): large volume, high velocity, variety, and veracity. The 4V characteristics of big data determine that big - data talents are not those who master skills such as distributed computing, streaming data processing, and data visualization but those who have data thinking.

Data thinking is different from traditional data analysis or data science. It is a higher - level cognitive way, that is, how to see the facts, reveal the laws, and optimize actions through data. Its core concepts include the following four points.

First, data - driven decision - making, that is, making decisions based on data rather than intuition. For example, when Starbucks opens stores globally, it does not choose locations based on experience but analyzes data on pedestrian flow, consumption levels, and competition to accurately predict the profitability of stores.

Second, pattern recognition, that is, discovering trends, associations, and outliers in data. For example, PayPal, one of the world's largest online payment platforms, has successfully reduced fraud - related losses by more than 70% and reduced the false - alarm rate by 50% through the detection of outliers in normal transaction patterns.

Third, causal reasoning, that is, understanding the causal relationships between data rather than just correlations. The best example is the survivor bias effect, that is, people only focus on the "surviving" or successful things and ignore those that have failed or disappeared, thus leading to judgment biases.

Fourth, maximizing data value, that is, mining value from data to create a competitive advantage. For example, Netflix analyzed data and found that people who watched political dramas also liked the movies of Kevin Spacey and David Fincher. So, it produced "House of Cards" starring Spacey and directed and executive - produced by Fincher. During the premiere of "House of Cards", Netflix added about 2 million new subscribers in the United States.

Logic: Deduce the unknown from the known

Imagine that in ancient times, our ancestor, a Homo sapiens, had good luck one day and hunted many animals. There was more meat than he could eat, so he wanted to save it for the next day. The problem facing this Homo sapiens was how to preserve the meat and prevent ants or mice from stealing it. Suddenly, he remembered the fruits hanging on the branches he saw on the hunting road, and then he came up with a solution: string the meat with a rope and hang it on the rock wall so that ants and mice couldn't reach it, and the meat could be preserved. By observing natural phenomena, summarizing laws, and then applying them to another scenario, this is inductive reasoning, also known as "statistical learning" or "associative learning".

This learning ability is not unique to humans and is also common in other animals. The reason why Homo sapiens were able to break out of the natural food chain and become the masters of all things was not because they learned inductive reasoning but because they summarized the knowledge obtained through inductive reasoning into the form of "IF - AND - THEN".

» IF: If a hole is made in an object and a rope is passed through the hole, it can be hung up.

» AND: I have meat and a rope now.

» THEN: So the meat can be hung in the air.

At first glance, this seems to be over - complicating a simple common sense. However, it was this formalized expression that brought fundamental changes to the lives of Homo sapiens, making humans change from products of the environment to creators of the environment and starting to transform the world according to their own ideas.

Aristotle called this syllogism deductive reasoning and called the content of IF the "first - principles". That is, all things in the world can be traced back to the most basic principles or causes through logic and reasoning, and then, starting from the first - principles, disruptive innovation can be achieved.

From the primitive "IF - AND - THEN" thinking mode of Homo sapiens to the AGI era where 0s and 1s are ubiquitous today, deductive reasoning, as a new thinking mode, makes humans different from animals. Obviously, compared with learning mathematical operations (such as partial differential equations) or physical rules (such as quantum computing), it is more important to master the thinking mode of first - principles and deductive reasoning. Unfortunately, in the knowledge - based education system, we always train students to directly perform calculations or reasoning after seeing a problem in order to get an answer immediately. The drawback of this straight - line thinking mode from problem to answer is that it is easy to be confined within the existing knowledge framework and fall into the trap of "local optimum".

The thinking mode of deductive reasoning starting from the first - principles requires consciously training U - shaped thinking. The core of U - shaped thinking is: instead of directly looking for an answer, deeply explore the structure of the problem, find its most essential core elements, and then reconstruct the answer. This process includes challenging the major premise (identifying implicit assumptions), disassembling the core elements of the problem (decomposing complexity), and breaking the boundaries of reasoning (redefining possibilities).

Elon Musk, who has continuously created business miracles and become the world's richest man, said when looking back on these achievements, "First - principles is a way of thinking in physics. You have to strip away assumptions, break things down to the most basic truths, and then reason from there." This is because "the only rules are those stipulated by the laws of physics. Everything else is just advice." Similarly, deductive reasoning is one of the most core thinking abilities of humans, and we have no reason not to use and strengthen it.

Psychology: Understand yourself and perceive others

Facing a great mission and possible great success, our first reaction is not excitement but doubt and even fear: "Am I worthy?" The humanistic psychologist Abraham Maslow called this phenomenon the "Jonah complex", that is, we are not only afraid of failure but also afraid of success. Sometimes, the biggest enemy that prevents you from reaching the peak of self - actualization is not others but yourself.

When you worry that success will make you feel lonely at the top and arouse the jealousy of friends or relatives; when you worry that after success, you will be embarrassed to be in the spotlight; when you worry that the higher you climb, the harder you will fall; when you worry that success is just a flash in the pan and glory will fade quickly, you become the biggest enemy blocking your success.

This is also why in the field of video generation, most people focus on generating short videos of a few seconds rather than dozens of seconds because generating short videos of dozens of seconds involves the real simulation ability of the physical world, which is regarded as an insurmountable gap in video generation.

However, why don't the young people at OpenAI have the Jonah complex?

From the perspective of modern psychology, happiness consists of three levels: material happiness at the bottom, psychological happiness in the middle, and social happiness at the top. From the pursuit of material happiness to psychological happiness and finally to social happiness, these three levels reflect different stages of human pursuit.

In August 2016, Jensen Huang donated the first supercomputer DGX - 1 to OpenAI, which had been established for less than a year. Elon Musk, who had previously donated $100 million, was also invited to witness it. On the shell of the DGX - 1 machine, Huang wrote:

To Elon Musk and the OpenAI team:

For the future of computing and humanity, I donate the world's first DGX - 1.

Therefore, whether it is Huang or OpenAI, from the very beginning, they pursued social happiness rather than using work as a means of making a living or achieving financial freedom, nor using work as a ladder to look down on others from the peak. Instead, it was for "the future of computing and humanity", which is also the reason why OpenAI aimed to innovate in generating dozens - of - seconds videos from the start. So, the "ambition" (mission) of the young people at OpenAI is the source of their happiness.

Maslow said, "If you always try to hide your inherent brilliance, your future will surely be bleak."

Rhetoric: Persuade others and lead innovation

Rhetoric was one of the core contents of general education in ancient Greece. Aristotle pointed out in "Rhetoric" that the essence of rhetoric is "to find the most likely way to persuade others in each situation". Therefore, rhetoric is not just a speech or writing skill but a way of thinking used to influence others, convey information, and shape opinions.

For example, many successful entrepreneurs can attract investors not because their business plans are perfect but because they can use rhetorical skills to tell an attractive story. Investors are willing to invest not only because of the data support but also because they are touched and convinced. In fact, we also often use rhetorical skills in our daily lives: from deciding where to have a party or travel, to persuading the boss to support your plan or project, and to expressing opinions and issuing initiatives on social media.

The core role of rhetoric is to prompt social members to form unified action goals and values, that is, consensus. In today's society with rapid information flow and increasing globalization, the role of consensus is becoming more and more important, thus forming a consensus premium. Consensus premium refers to the phenomenon that the common recognition and preference of society, the market, or a group for an object cause the value or price of the object to be higher than its actual value for a period of time.

In the AGI era, AI can more efficiently achieve a wider - reaching consensus through rhetoric.

In terms of emotion, through sentiment analysis of social media and monitoring of public - opinion hotspots, social anxieties, and expectations, AI can more efficiently grasp and mobilize the emotions and needs of the public. In terms of logic, AI can establish a consensus system in a decentralized network rather than relying solely on a single authoritative institution. AI can be used to automatically verify, record, and update the data in the process of consensus generation, ensuring the transparency and authenticity of the data. In terms of authority, the natural language processing technology of AI can break language and cultural barriers, promote global communication and understanding, and help achieve cross - cultural and