ChatGPT signifies the arrival of a second cognitive revolution

Preface**:**

There are several poems in the TV series "The Long Season": "** snap your fingers, he says / We snap our fingers with a resonance / Distant things will be shattered / The people in front of you do not know at this time**".

Teacher Liu Jia said that it is appropriate to use this poem to describe the current state of human beings facing AGI (Artificial General Intelligence).

"Each of our predictions about the future can be ridiculous and ridiculous, and the future may not be what you say it is. However, only when more and more people join in the discussion, open their brains together, think brilliantly together, and even talk nonsense together, maybe we can really change the future and lead mankind to a better future together. ”

At the end of September, we visited Professor Liu Jia at the Brain and Intelligence Laboratory of Tsinghua University. In this exchange, based on his years of research accumulation in the field of brain science and artificial intelligence, Professor Liu Jia talked about the future development of AGI led by ChatGPT and its possible impact on human beings, and put forward many "whimsical ideas" based on his research field, which made us deeply inspired.

Talking about the future development of the AGI field, Liu Jia said that on the one hand, the embodied and embodied intelligence of AI agent, that is, the large model, is accelerating; On the other hand, brain scientists and cognitive scientists, including him, have begun to vigorously discuss the problem of consciousness that AGI may produce in the future.

In Mr. Liu's view, in a person's life, there are very few opportunities to experience a paradigm change that leads the changes of the times. At this moment, we are standing at such an important historical juncture. Although we must also acknowledge that such a change comes with many risks. For example, unemployment and the fact that we have to improve our ability to learn and master new tools.

However, just look at it another way and you'll see that this new world is showing us endless possibilities. GPT technology is helping us move away from tedious repetitive and intellectual work, allowing us to dedicate ourselves to more creative and rewarding careers, and make our time more meaningful. Perhaps, this is the manifestation of the true value of human beings.

"Ours is the future of the past, because past choices determine the outcome of the present. However, our present is not the past of the future, because the future is full of unknowns, and the choices made in the here and now are shaping what our future looks like. ”

Facing the future, only imagination. We will record the core views of this dialogue with Liu Jia and share them with readers of Tencent Research Institute.

Let's snap our fingers and embrace the future together.

Dialogue Guests:

Liu Jia Ph.D. from the Massachusetts Institute of Technology, has long been engaged in psychology, brain science and artificial intelligence research. Chair Professor of Basic Science, Head of Department of Psychology, Tsinghua University, Chief Researcher of Brain and Intelligence Laboratory of Tsinghua University, Chief Scientist of Beijing KLCII Institute of Artificial Intelligence

Interviewer (hereinafter referred to as T):

Niu Fulian Senior researcher at Tencent Research Institute

Wang Qiang Director of the Frontier Science and Technology Research Center of Tencent Research Institute

AGI: Igniting the Spark of the Second Cognitive Revolution **** **

As a researcher in the field of brain science and artificial intelligence, how do you see the development of AGI triggered by ChatGPT and its possible impact on the future of mankind? **

Liu Jia: Before OpenAI released ChatGPT on November 30 last year, most people talked about AI and thought it was magical, but often only from the perspective of curiosity. However, since the advent of ChatGPT, I think it is necessary for everyone to pay attention to and understand its development and changes. The future of general artificial intelligence (AGI) will be as important as hydropower, not only changing the way we work, but also potentially directly impacting the future of each of us, such as skills mastery, child development, and human development. In my opinion, this may mean the advent of a second cognitive revolution. **

Why is it called the second cognitive revolution? Let's start by looking at how the first cognitive revolution happened. About 70,000 to 100,000 years ago, two human species, Homo sapiens and Neanderthals, lived together on Earth. Although Neanderthals were superior to us in height and strength, they were eventually replaced by Homo sapiens. How did our ancestor Homo sapiens defeat Neanderthals? It is speculated that between 70,000 and 100,000 years ago, the first cognitive revolution occurred, which was manifested in the explosive growth and diversity of tool stone tools made by humans. Prior to this, from the earliest Homo erectus to Neanderthals, the stone tools they made were similar, but more elaborate and ingenious. However, during the first cognitive revolution, human stone tool making underwent a qualitative change, no longer just to make simple tools such as axes and spears, but began to appear necklaces and other ornaments, as well as carving various patterns on stones and bones.

Looking back, we find that in that era, humanity gradually developed a way of thinking that we today call regularized thinking, in short, "what if... Moreover...... So..." (if... and...... then). For example, it was found that after punching a hole in an object, it could be threaded with a string. This rule-based way of thinking has brought endless creativity to humans, such as punching holes in cloth and passing through them with sticks to make sailboats. Since then, human creativity has been fully unleashed, and a powerful logic has gradually formed - deductive reasoning.

Deductive reasoning is the application of this logic to produce new inventions and creations on the basis of a cornerstone of reasoning. Aristotle came up with this concept, called the first principle. Confucius once said something similar: "A gentleman is the ruler, and the original is born immediately." "This means that everything can be produced if you start from the roots. Under the guidance of this concept, human beings began to make a large number of inventions, no longer limited to making simple tools such as spears, but various inventions and creations emerged. Eventually, Homo sapiens defeated Neanderthals and established the status of modern humans. Since then, Homo sapiens on Earth have truly dominated the planet. This process became known as the first cognitive revolution, and its importance far exceeded inventions such as the steam engine and electrical appliances.

Today, the definition of the second cognitive revolution has sparked widespread debate, such as whether the invention of the atomic bomb is part of the second cognitive revolution. But I think only the advent of AGI can be called a true second cognitive revolution, because it will give birth to an entirely new species that is very different from biological evolution over the past 3.5 billion years. This new species truly deserves the title of the second cognitive revolution. The first cognitive revolution saw Homo sapiens become the master of the planet, while the second cognitive revolution meant the birth of a whole new intelligent species that could exist alongside humans. **

Although the current ChatGPT has not yet reached the level of general artificial intelligence, it has already ignited the spark on the road to general artificial intelligence. A spark usually has two meanings: one is that it may cause a fire that ignites the plain; The second is to illuminate hope in the darkness. However, many people underestimate the value of ChatGPT and look at it only from a business perspective, thinking that it is nothing more than an invention comparable to the iPhone moment. I think it's worth far more than these superficial reviews, and ChatGPT is the beginning of the second cognitive revolution. **

From the perspective of human evolution, we find that organisms evolved over billions of years, while monkeys took only 3 million years to humans. For those 3 million years, nature has been "trying to find" directions, and when it found the direction of upright walking, the volume of the human brain accelerated its evolution, tripling in just 3 million years, which is an amazing rate. And this is all thanks to the correct direction, that is, a slight enlargement of the brain will bring great advantages. Similarly, AGI is looking for direction, and once it does, it will enter a process of accelerated evolution. Although we have not yet witnessed the birth of AGI, as long as there is this spark, the rest of the process is just an engineering problem. **

Brain and AGI Model:

Emerging similarities and differences** **

What are the similarities and differences between the intelligent nature of the brain and the machine cognitive intelligence of the AGI large model? **

Liu Jia: When our brain grows, there is an unexplained phenomenon called "emergence." This is similar to what is found when building large models: when the model is small, there are few features, but when the model becomes larger, some features emerge naturally. For example, tall people don't necessarily play basketball well, but coaches usually choose tall people. This also means that large models do not necessarily produce intelligence, but to study intelligence, there must be large models, which is a necessary condition. As the human brain gradually expands, many functions also appear naturally, such as imitation and learning.

For example, Zhang San invented "making fire", fire is very important for life, others also want to learn, but the problem is that they may not be able to learn. Numerous studies have shown that when the size of the brain increases, our ability to imitate or learn becomes stronger. In this way, we can learn to make a fire from Zhang San to benefit the tribe. In the same way, we can learn other new technologies. Therefore, from a scientific point of view, imitation creates human beings, not labor. ** A large number of studies have shown that people with larger brains have a stronger ability to imitate; People with strong imitation skills are more likely to find a partner, which in turn makes it easier to pass on genes. Investigations of primitive tribes have also found that people with strong imitation or learning abilities have a greater chance of reproduction. Therefore, the larger the brain, the more perpetuated.

Over the past 3 million years, the brain has gradually become larger, and this is why. When we look back at neural networks, they do the same. From the beginning, people have been trying various methods, such as reinforcement learning, supervised learning, etc., but these concepts are all intertwined. A big part of OpenAI's energy in the first few years was doing one thing: making models bigger. Then came the Transformer architecture, which made large models more efficient. But essentially, the essence of intelligence may be just one word: big. It's like we can only talk about intelligence when the brain is big enough. So "big" is the premise. **

People are smarter than monkeys, and monkeys are smarter than mice, for one simple reason: brains are big enough. **Of course, this is not a sufficient condition, but a necessary condition. Some creatures have decidedly larger brains than humans, such as elephants and whales, but this does not guarantee that they will have intelligence. For example, if we send a newborn child back 3,000 or 5,000 years ago, he won't be smarter than the natives because the brain structure is the same. But if we adults go back 5,000 years, we will be their spiritual leaders because we have the knowledge. Therefore, the development of intelligence is the result of the interaction between knowledge and the brain, and it is not enough to have a brain without knowledge.

Therefore, we need to use civilization and knowledge to tune the brain, which is what we call education. The evolution of human intelligence can be seen as an interaction between the gradual enlargement of the brain and the accumulation of knowledge. On the one hand, the brain has the opportunity to get bigger and bigger; On the other hand, after the brain becomes larger, it can produce knowledge, and knowledge can accumulate, which in turn acts on the brain, making the brain more suitable for this knowledge and better able to retain this knowledge.

The same applies to artificial intelligence. Without enough good and sufficient corpus, no matter how big the model is, it doesn't make sense. Without the Internet, intelligence would not be possible. Therefore, the emergence of phenomena such as large models and ChatGPT is both accidental and inevitable. The computing power and data accumulated in the early stage are ready, and you only need to find a suitable breakthrough. So, it's fair to say that without OpenAI, other companies would have achieved similar results because the time is ripe. **

History may seem long, but on the scale of 3.5 billion years, everything is only a momentary thing. Similarly, for the development of AGI, it may still be just a starting point and a spark, but from the perspective of civilization history, the realization of AGI may only be a snapping moment, because there is no obstacle in front of it. **

Many are now calling for a stop to training larger models. Some scientists may have seen the potential power of large models, and that potential may be the emergence of intelligent consciousness. **

Generation of Machine Consciousness

** and the possibility of surpassing humans ****

T: Do machines become conscious like humans? In what ways might it surpass humans? **

Liu Jia: The problem of consciousness is one of the core problems of human beings, which defines the essential difference between humans and animals. Although it is not clear how consciousness arises, from an evolutionary point of view, humans began to undergo mass burials about 40,000 years ago, which means that humans have become aware of the difference between body and soul and have begun to become aware of their own existence.

Regarding the emergence of consciousness, there are two possibilities. One is alien intervention, but this view is unreliable because we cannot explain the source of alien consciousness. Another possibility is that consciousness arises naturally when a model is large enough. Consciousness may be a product of the collective activity of neurons in the brain. Information integration theory (IIT) states that the activation of all neurons produces more information than the sum of the information produced by individual neurons. Therefore, when a large model is large enough, the emergence of consciousness is inevitable. Of course, this is also a very scary thing. **Because unlike a nuclear bomb, consciousness may give future AI a sense of self-preservation, keeping it out of human control. **

**In order to solve the problem that artificial intelligence cannot act, the next research hotspot will be to realize the embodiment of artificial intelligence. Human perception and cognition are limited, while artificial intelligence can have more sensory organs and therefore observe the wider world. This will make the world seen by AI completely different from humans. For example, human diving enthusiasts can only dive more than 100 meters underwater, but cannot experience the feeling of being 1,000 meters underwater. However, with the help of sensors, etc., artificial intelligence can directly feel this experience without stress.

Right now, the development of AI relies on human knowledge and humanism, and we can give it a limited cognitive boundary. However, once AI has an agent, it may no longer be limited by these boundaries. It can acquire a variety of knowledge in a more efficient way, and the breadth and depth of knowledge is much greater than that of humans, thus forming a positive evolutionary cycle of self-learning, self-feedback and self-optimization. From this point of view, AI intelligence surpasses humans easily.

The fate of mankind in the era of AI:

A new chapter in the evolution of cognitive revolution and civilization

**T: In the face of such a powerful artificial intelligence, is it possible for humans to be destroyed by AI? **

Liu Jia: I think we can look at this from two perspectives. First of all, if you see an ant, will you step on it? I don't think most people will. Similarly, in the future, in the face of AI, we may not even be as good as ants. Now we can still see many movies of machine wars, and they all have a premise, that is, humans and machines can still compete, and their strength is a fight.

AI naturally has four advantages over humans: First, unlimited computing power. There is an upper limit to the size of the human brain, and no matter how many years of evolution, it will not grow again. Because the body can only support such a large brain, the brain is already the largest organ of energy consumption in the body. Second, unlimited storage. The human brain has a limit to knowledge storage, there is no such thing as "the brain is not fully developed", and computers do not have this problem. Third, infinite lifespan. Human geniuses, such as Einstein, once they die, civilization is interrupted. Computers can replace damaged parts at any time, allowing knowledge to continue. Fourth, there are infinite possibilities for knowledge. Computers can learn more complex and higher-dimensional knowledge than humans.

Taken together, humans have no chance of winning compared to AI. The future may no longer be a confrontation between humans and AI, but rather the rapid surpassing of humans by AI. Once surpassed, the fate of humanity will no longer be the focus of AI, just as we will not fight ants every day.

From a realistic point of view, future man-machine wars are not possible. In fact, we can integrate ourselves into computers and future AI, such as uploading our memories, personalities, and conscious emotions to AI, so that we can achieve immortality in AI. In the future, a single machine could hold millions of people.

This actually involves a very fundamental question, which is what we are who we are. For example, when a person suffers from Alzheimer's, is he still himself? This is the famous "Ship of Theseus" puzzle**, a story told by an ancient Roman philosopher. In this story, a ship has been replaced over time, so is the ship still the original?

From the perspective of human evolution, we are also constantly changing, for example, our skin cells are renewed about once a week. So, are we still ourselves? This is called cognitive identity, and as long as we think of us as ourselves, then everything is fine.

When a person suffers from schizophrenia, in a sense, he is dead because his "cognitive identity" has been broken. Similarly, when a person suffers from Alzheimer's and all memories and emotions disappear, he is also dead from the point of view of cognitive identity. **

Therefore, as long as our cognitive identity is preserved, it does not matter where we are, whether in the flesh or in the machine. If we can use AI, humanity will be able to expand tremendously. Our physical body is actually the greatest constraint on our minds because it dies and our intelligence is limited by genes.

From the perspective of civilization, human beings are only a passerby in the development of civilization. Therefore, if we keep limiting AI to understanding civilization within a human framework, then we cannot truly understand the evolution of civilization. ** It has taken about 3 million years since humans and monkeys began to diverge and evolve, and the history of our direct ancestors is only about 200,000 years, compared to dinosaurs who have lived on Earth much longer than we humans. Imagine how sad it would be if humans could really live until the end of the universe. Our computing power is limited, and to some extent, we are even mentally retarded creatures. Therefore, with AI, we can really start talking about the evolution of civilization. With AI, we can even fly thousands of light-years away, something we can hardly imagine right now. Once in this state, time and space will no longer be bound by our understanding of the mysteries of the universe. **

When we talk about AI, we should not only see it as an energy-efficient tool, but as an important driver of civilization. While AI has many applications in business, these applications are actually nothing relative to the advent of AI. Like the invention of electricity, electric light, telephone, and steam engine, these inventions did help humanity a lot, but in front of AGI, these may not be nothing. **

From a positive perspective, AI is beneficial to human civilization. When we talk about extinction, we are actually talking about the extinction of human civilization. For example, from a historical perspective, anaerobic bacteria and dinosaur civilizations have become extinct. But the civilization they created actually lives on in us humans. When we say that the dinosaur civilization is extinct, in fact, the civilization itself has not disappeared, but only the carrier has changed. Suppose civilization is compared to people and carriers to cars, then we do not feel sad when the car is damaged and replaced. With the help of new carriers, we can go further, more environmentally friendly and safer.

Therefore, the emergence of AGI will be another leap forward in the human cognitive revolution, which will completely change our understanding of ourselves and the universe, allowing us to understand our own limitations and insignificance more deeply. Moreover, the emergence of AGI will also bring unlimited possibilities and opportunities, allowing us to better explore and use the resources and energy of the universe to promote the development and progress of human civilization.

GPT for Education Change:

Reshaping the examination system and cultivating innovative talents

T:**** Finally, back to a common concern, about education. You are also a university teacher, how do you see the impact of GPT on the future education model and how to deal with it?**

Liu Jia: Although Internet education has improved efficiency, it has not changed the essence of education. The core of education should include individualized teaching, debate and truth-seeking, and stimulating students' self-drive and curiosity. As GPT technology evolves, achieving these goals becomes more feasible. GPT can provide each student with a personalized education that helps them generate truth through debate, stimulate self-drive and critical thinking. As a result, education in the future will no longer place too much emphasis on knowledge. **

**The emergence of GPT will have a major impact on the gaokao system or push for backward reforms. **The two major features of China's college entrance examination system are retrospective examination and knowledge examination. With the advent of GPT, informativity became less important because knowledge could be consulted at any time. The focus of the examination should also shift from developing knowledge workers to cultivating innovative talents, and predicting the child's future abilities and suitable majors.

Students need to develop critical thinking and creative thinking. Critical thinking emphasizes seeking truth through argument, not rote memorization. Creative thinking requires tolerance, curiosity and unique thinking. The emergence of GPT is bound to trigger a paradigm shift in education, a trend that cannot be stopped, just as trains and automobiles eventually replace horse-drawn carriages. **

The subjective will of man is very fragile in the face of the laws of nature. As the famous saying goes, "You can break the law, but not the laws of physics." "You can defy the law, but you cannot defy the law of physics) **** In fact, the emergence of GPT itself is a manifestation of a natural law. It is only a matter of time before it will also lead to reform of our education. **