AGI Paradigm Revolution and Humanity’s Future

(This article is based on Professor Liu Jia’s keynote speech at the “Tencent Academy Classmates Meeting – Smart Future: AI Hefei Study Tour”)

Author | Liu Yanni

Since the advent of ChatGPT on November 30, 2022, generative artificial intelligence has emerged like the morning star, heralding a technological revolution that leads a new era. The wave of this revolution has quickly swept across the globe, impacting people from all walks of life who feel its potentially disruptive effects. In this unprecedented change in a century, understanding the deep logic of generative artificial intelligence and its potential paradigm shifts in human society and the business world has become particularly important.

01 AGI vs AI: The Difference Between Humans and Apes

The Tencent CHO Private Board has organized five thematic learning sessions since its first event in 2021, with 45 partner companies’ CHOs joining us. They come from different industries and development stages, making each of our collisions and exchanges very rich and interesting.

When we talk about artificial intelligence, you might think of various achievements in AI over the past decade: facial recognition, text recognition capabilities, big data analysis capabilities, and even AlphaGo defeating Lee Sedol. However, how is the AGI we are discussing now different from the AI of the past?

We can categorize artificial intelligence using the four quadrants shown in the diagram below. One dimension of the quadrants represents the environment (X-axis), which can be either open or closed. In a closed environment, the tasks that AI has to handle are defined, while in an open environment, the tasks are uncertain. The other dimension represents the nature of the tasks (Y-axis), which can be static or dynamic. Static tasks are similar to analytical work, defined and specific, such as facial recognition; while dynamic tasks change according to circumstances, such as teaching activities where a teacher needs to flexibly adjust the lecture content and interaction with students based on their responses.

The two dimensions of environment and task can divide artificial intelligence into four quadrants. Tasks like facial recognition and text recognition belong to the third quadrant (CV), where AI needs to complete simple static tasks in a closed environment, primarily involving identification and classification. The “Go master” AlphaGo, which defeated Lee Sedol in 2016, belongs to the second quadrant (RL), where AI processes dynamic tasks in a closed environment while responding flexibly to the opponent’s moves. The fourth quadrant (Cybernetics) is exemplified by Boston Dynamics’ robotic dog, which needs to perform static tasks (exploration and feedback) in an open environment (rooms/outdoors/flat ground/stairs, etc.), representing a simple control-type feedback.

In these three quadrants, the previous generation of artificial intelligence has made tremendous progress and even surpassed human capabilities in certain aspects. However, in the first quadrant (AGI), humans still maintain an absolute advantage in completing open tasks in an open environment. Before the emergence of GPT, artificial intelligence in the first quadrant only existed in science fiction. But the advent of GPT has made machines that are as intelligent as humans gradually become a reality.

Thus, we can say that if we refer to the task processing from the second to the fourth quadrants as AI, and the first quadrant as AGI, the difference between them is as vast as the difference between apes and humans.

02 Where Does ChatGPT Come From?

“Language is the Carrier of Thought”

If we want artificial intelligence to think and operate like humans, the primary condition is to give it human-like thinking capabilities. So what is thinking? Philosopher Holmes once said: “Language is the blood of the soul, and thoughts run and grow within it.” The carrier of thought is language, and our understanding of the world is realized through the description of language. Language encompasses all our human thoughts, all our logic, and all our rationality; the boundaries of language are the boundaries of the world, which applies to both the human brain and machines.

Therefore, having language capabilities is an important path to achieving thinking abilities for AI. Geoffrey Everest Hinton, the father of deep learning, has made a romantic description: “The caterpillar extracts nutrients and then transforms into a butterfly; GPT-4 has extracted billions of nuggets of thought, absorbed our nutrients, and finally transformed into a butterfly.”

This is why large models are often referred to as large language models (LLM).

Understanding the Three Elements of AGI from Human Evolution: Massive Computing Power, Large Parameters, and Big Data

Looking back at the evolution of humans from capable individuals to wise individuals, if we consider the tools invented by humans as an important marker of human intelligence, from about 3 million years ago to about 100,000 years ago, the types of tools invented by humans did not show significant changes. This indicates that human intelligence did not evolve significantly over this long period of 3 million years, a time referred to in human evolutionary history as “historical garbage time.”

However, around 70,000 to 100,000 years ago, remarkable things happened. The types of tools invented by humans exploded, and there were even meaningful inscriptions on some tool fossils. For example, the fossil of a baboon bone from 43,000 years ago has 28 scratches; what does this mean? Archaeologists discovered that these scratches were used to record time, representing 28 days of a month. This indicates that humans at that time not only invented a rich variety of tools but also began to record what was happening in the world through scratches. Later, cuneiform writing emerged, leading to a leap in human cognition, a period referred to academically as “the first cognitive revolution of humanity.”

(The fossil of a baboon bone from 43,000 years ago)

Why did such an explosion and breakthrough occur? A reasonable inference is that although human intelligence did not have significant breakthroughs during the past 3 million years of “garbage time,” the size of the brain continued to grow, resulting in more neurons, more complex connections, and more advanced brain structures. This is a remarkable miracle in evolutionary history because, from an evolutionary perspective, no other organ has increased in size by nearly three times in such a short period of 3 million years.

When the human brain had greater computing power, could understand more dimensions of parameters, and analyze larger amounts of data, intelligence underwent a qualitative leap, leading to the first cognitive revolution of humanity.

The same reasoning applies to artificial neural networks. If we want a neural network to possess thinking capabilities like humans, it must have richer and deeper network connections. Hinton’s lifelong research and pursuit have been to make artificial neural networks deeper, which we refer to as deep learning. He firmly believes: “The human brain connects through neurons. There is no reason why artificial intelligence cannot achieve this. If it cannot be achieved now, it is simply because we are too shallow and too small in our efforts.” Hinton’s student, Ilya Sutskever, CTO of OpenAI, also stated: “I have a very clear belief: do not doubt deep learning. If you want to create artificial intelligence, you must believe in deep learning; deep learning is the foundation of artificial neural networks, and there is no need for doubt.”

It was under this guiding belief that in mid-2020, GPT-3, characterized by massive computing power, large parameters, and big data, was released. Just like the explosion of the human brain after millions of years, the new generation of artificial intelligence reached the “pinnacle of violent aesthetics.”

From GPT-3 to ChatGPT: Even a Perfectly Engineered “Baby” Needs Training

However, after the intelligent GPT-3 was released, it initially produced many nonsensical answers, leading to much ridicule. For example, if you asked it how many eyes a person’s foot has, it would answer that a foot has two eyes. Such absurd answers occurred because it did not truly understand what an eye is; it could only base its response on the fact that humans have two eyes and that feet belong to humans, so feet should also have two eyes. In simple terms, at that time, GPT-3 only established a random probabilistic relationship and did not understand the meaning behind the language. Thus, it was humorously nicknamed a parrot that answered randomly, failing to attract much attention, and after a brief period of excitement, GPT-3 fell into obscurity.

But two and a half years later, OpenAI released ChatGPT, shocking the world. It was no longer a parrot but more like a human. What happened during those two years? Why did GPT-3 fail while ChatGPT and the later GPT-4 succeeded?

We can illustrate this with a simple hypothesis: if we send a modern baby back 3000 years, would they be better than a baby from primitive society? The answer is obviously “no” because there is no difference in brain structure between primitive humans and modern humans. However, if we send an educated modern adult back to primitive society, would they be better than the people from that era? Certainly. Why? Because under the same genetic brain conditions, modern individuals have received more advanced education.

Applying this hypothesis to GPT makes it easy to understand the underlying reason. We can consider GPT-3 as a “baby” with perfect genes but lacking education, which is why its performance is unsatisfactory. OpenAI spent two and a half years educating it, which included prompt engineering (teaching), reinforcement feedback (instructing), and aligning it with human worldviews, life views, and values (clarifying). After receiving this education, it began to exhibit powerful capabilities, which we later saw in ChatGPT and subsequently GPT-4.

03 The Paradigm Shift Brought by AGI

The concept of paradigm shift was first proposed in 1962 by the famous American philosopher of science and historian Thomas Kuhn in his book “The Structure of Scientific Revolutions.” He believed that scientific progress is not achieved through gradual accumulation of knowledge but through a series of revolutions, where one dominant paradigm is replaced by another. What is a revolution? A revolution is a fundamental change in our perceptions and understanding of the surrounding world, where the dominant way of thinking, basic assumptions, and practices undergo drastic changes, which can be termed a paradigm shift. The impact of AGI on human society today is a paradigm shift.

Business Paradigm Shift: Moore’s Law Also Applies to Services

What impact will AGI have on business? Let’s start with a story. The famous British detective novelist Agatha Christie mentioned in her autobiography that in 1919, as a middle-class individual, her annual income was £700, allowing her to own a small house in central London and hire a maid (£30/year) for daily living, as well as a professional nurse (£40/year) to care for her children. However, she thought she would never be able to afford a car, as the cheapest Ford car was £170. But looking at it now, most families can afford a car but cannot simultaneously hire a maid and a professional nurse.

This is the result of the first three industrial revolutions, where the prices of goods continue to decrease, but the prices of services become increasingly expensive. This aligns with what Intel co-founder Gordon Moore proposed as “Moore’s Law,” which states that with continuous improvements in production efficiency, the prices of goods become lower. (Moore’s Law: The number of transistors in integrated circuits will increase exponentially over time, while costs will decrease correspondingly. Although Moore’s Law was initially applied to the industrial sector, it has also been used to describe certain phenomena in commercial society.)

Sam Altman, founder of OpenAI, believes that Moore’s Law also applies to the service industry. In 2019, he quietly published an article on the company’s Weibo account titled “Everything is Moore’s Law.” The article mentioned that Moore’s Law applies not only to goods but also to services. General artificial intelligence will bring the fourth industrial revolution, which can continuously reduce the prices of services, such as lawyers, accountants, teachers, etc.

For most people, such reasoning may seem radical and unrealistic, but the phenomenon of general artificial intelligence “making services cheaper” has indeed begun to emerge. A statistic from Goldman Sachs indicates that after the emergence of ChatGPT, the demand for freelancers plummeted. Goldman Sachs’ research report stated: “In the United States, about 7% of jobs could be replaced by AI, 63% of jobs will be assisted by AI, and only 30% of jobs are unaffected by AI. About 300 million jobs may be replaced by AGI.”

Combining this with the content released by OpenAI’s governance team at the beginning of 2023, the idea of “making services cheaper” seems increasingly plausible.

Paradigm Shift in Human Society: New Equality, New Wealth

The advent of AGI will bring tremendous changes to business, but what profound impacts will it have on human society? The first three industrial revolutions brought fundamental changes to production methods, with the use of machines and the establishment of factory systems achieving mechanization and large-scale concentration of production processes. People realized that breaking down the production process into multiple tasks, completed by specialized workers, could further improve production efficiency, gradually forming the modern division of labor system.

However, the emergence of AGI has a revolutionary impact on the division of labor system. Taking Sora as an example, in the past, producing a video required collaboration among screenwriters, photographers, animators, post-editors, and actors. However, with the help of Sora, anyone can generate any exciting video by simply inputting a text description. Similar scenarios will continue to emerge, leading to two essential changes in modern society:

New Equality: Mental Ability Becomes a More Core Competitiveness. In ancient times, equality among people was established on “physical strength”; whoever had the strongest fist had the power. In modern society, equality is based on “skills”; those with professional skills and mastery are more competitive. However, in the AGI era, the competition of “mental ability” will gradually become the core competitiveness. A person may have physical disabilities or lack specific skills in photography or painting, but their various creative ideas and insights will become the core of competition.

New Wealth: Time Becomes an Important Component of Wealth. As mentioned earlier, there was a long “garbage time” in human evolutionary history when human intelligence did not significantly improve, and technological breakthroughs were rare, making time not very important. However, in the AGI era, technological development is rapid, and artificial intelligence technology evolves in 72-hour cycles, making time a crucial factor. Those who can better utilize time and create greater value within a unit of time will become important indicators of wealth.

Human Response: Emphasis on General Education, Innovation, and Ambition

In response to the changes of new equality and new wealth, individuals will inevitably need to make corresponding transformations. Among these, we particularly want to emphasize three points: the importance of general education in the AGI era, the maintenance of innovation, and the stimulation of ambition.

General Education: The Return of Human’s More Important Abilities to General Competencies. The first three industrial revolutions gave rise to specialized divisions of labor, leading to the formation of the modern education system that cultivates professionals based on disciplines. However, with the emergence of AGI, many technical and specialized jobs can be handled by artificial intelligence. The more important abilities of humans will return to general competencies, including: logical thinking (drawing correct conclusions from known facts), mathematical insight (understanding the meaning of data and forming one’s own analysis), research ability (discovering, studying, and solving problems), social skills (understanding others and oneself), and the ability to persuade others.

Creativity: An Important Ability in the Competition of “Mental Ability”. The two images below clearly show the difference between “knowledgeable people” and “innovative people.” The left shows an image generated by AGI, while the right shows a painting by Van Gogh. AI can imitate but finds it difficult to create. In the AGI era, we can easily access various types of knowledge, but those who can truly innovate are those who can fully utilize knowledge and resources.

Ambition: A Moment of Unprecedented Change Calls for Courage and Faith. When OpenAI was founded in 2016, it was still an inconspicuous company. Nvidia’s Jensen Huang donated the first supercomputer DGX-1 that the company had just developed to this new company that had been established for less than a year. On the machine was a line that expressed blessings to the OpenAI team: “For computing and the future of humanity, I donate the world’s first DGX-1.” The AGI era is like the dawn, containing unprecedented opportunities and possibilities. At this special and important historical moment, humanity needs more ambition, more faith, and determination to participate, create, and achieve a beautiful future for humanity.

04 Coexistence of Humans and Machines: Beauty in Harmony, Unity of the World

Google’s futurist Ray Kurzweil predicted that true artificial intelligence will emerge by 2029. As AGI continues to develop, discussions about the relationship between humans and AI, and how the two should coexist, are increasingly prevalent. Whether it is the debate between Google co-founder Larry Page’s speciesism and Elon Musk’s human prioritization, or discussions on whether AGI development should be effectively accelerated or super-aligned, it undeniably indicates that we are moving towards a new era of coexistence between humans and AI.

Perhaps AI will soon surpass humans in logic, mathematics, recognition, and other fields, but we firmly believe that in emotional cognition, innovative ability, and moral judgment, the unique advantages of humans are irreplaceable. Sam Altman has stated: “The world will undergo earth-shattering changes. We are at the beginning of a great transformation, and humanity has a rare opportunity to shape the future.” We believe that the unique values of humanity will play a crucial role in our symbiotic future with AI, guiding us to deepen our understanding and shape a harmonious coexistence with AI, allowing human civilization to shine brightly and endure.