Space Exploration and Digital Life
Space Exploration Requires Too Much Fuel
When I was a child, there was an old man in our yard who worked in aerospace. He often explained some aerospace knowledge to me. What impressed me the most was a solar system space map on his wall, similar to the one below.
Solar System Space Map (Delta-V Chart), Source: Wikipedia. My wife said it looked like a subway map at first glance. I asked her to look at the pictures inside more carefully, and she said it looked like stones :)
The old man told me, Doesn’t this solar system space map look a lot like a train route map? But the numbers on it are not distances, but changes in speed (Delta-V).
When he was young, he also hoped to build an extensive space network like a train network, but the stations would no longer be Beijing West, Shanghai Hongqiao, but the Earth’s surface, low Earth orbit, Earth-Moon transfer orbit, Mars transfer orbit, Mars surface, etc. Unfortunately, to date, humans have not visited most of the stations on this map.
The most important reason for this is that human energy technology is too backward in the face of space. Current rockets rely on ejecting propellant for propulsion, and the fastest ejection speed today is only about 4500 meters per second. This is much faster than a bullet, but still not very fast in space. For example, the first cosmic velocity is 7900 meters per second, and considering air resistance and gravity at a 250-kilometer orbit, a speed increment of about 9200 meters per second is needed to enter Earth’s orbit at 250 kilometers high.
More critically, the mass of fuel required by rockets grows exponentially with the required speed change. When I was in elementary school, I didn’t understand this. If the ejection speed of the propellant doesn’t change, shouldn’t the speed be proportional to the amount of fuel burned? For example, if a car’s fuel tank has twice as much fuel, it can travel twice as far.
The old man said that the more fuel burned, the lighter the rocket becomes, and burning the same amount of fuel can increase the speed more. This is the famous Tsiolkovsky rocket equation. My car example is not wrong, but that’s because the mass of the car’s fuel tank is a small part of the car’s total mass, so a first-order approximation can be made. If a car needs to travel tens of thousands of kilometers, and the fuel tank’s mass becomes a large part of the car’s total mass, then like a rocket, the mass of the fuel needed grows exponentially with the travel distance.
After encountering calculus in high school, I understood how this exponential relationship is derived and what first-order approximation means. I also will never forget that the derivative of ln(x) is 1/x.
Exponential growth is very frightening.
Actually, the first step to reach low Earth orbit (LEO) requires a speed increment of 9.2 kilometers per second to achieve the first cosmic velocity. The initial mass needs to be 7.7 times the final payload mass sent into orbit, which means the fuel must be at least 6.7 times the payload mass. This does not even consider the additional costs of fuel tanks, etc. This is just the first stop closest to the Earth’s surface on the solar system space map.
Imagine we have 4 passengers, totaling 300 kilograms, in a 2-ton car. Each trip would require 13 tons of fuel, and the car would be scrapped after one trip. Can the cost not be high? This is why space exploration is always so expensive. Even if SpaceX’s reusable rockets solve the cost problem of the rockets themselves, and the measurement and control costs can be reduced with scale, the fuel cost of rockets is still much higher than that of airplanes. For example, the average fuel consumption per seat per 100 kilometers for an A380 airliner is less than 3 liters, comparable to a household car. A transoceanic flight of 10,000 kilometers requires only 300 liters of fuel per person. But taking a rocket requires at least a few tons of rocket fuel (the old man told me that traditional rocket fuel was more expensive than aviation kerosene, but SpaceX reduced the fuel price by using liquid oxygen and aviation kerosene). The price is certainly higher for saving time. Of course, if it costs $10,000 to travel from China to the United States in one hour, many wealthy people would still be willing to buy tickets.
The second step is to overcome the gravitational potential energy from low Earth orbit to an infinite Earth orbit, requiring a speed increment of 3.3 km/s, which is the difference between the second cosmic velocity and the first cosmic velocity. With this second cosmic velocity, one can reach the intersection of the blue and yellow lines on the map, which means escaping Earth’s gravitational range. The amount of fuel must be at least 15 times the payload mass.
The third step is to leave the solar system, requiring a speed increment of 16.7 km/s and 5.5 km/s, which is the speed needed to travel from Earth to the top of the map outside the solar system. A total speed increment of 18.2 kilometers per second is needed (without considering gravity assists), requiring at least 56 times the payload mass in fuel, making it difficult for a single-stage rocket to meet the demand.
In university physics class, many people couldn’t calculate the speed increment needed to leave the solar system correctly. I haven’t found any large model that can calculate it correctly either. For example, GPT-4o always thinks that escaping the solar system requires adding the solar escape velocity at Earth’s position (42.1 km/s) minus Earth’s orbital velocity (29.8 km/s) to the second cosmic velocity, resulting in a total speed increment of 24.8 km/s, far exceeding the correct value.
The old man said, Fortunately, we live on Earth. If we lived on a larger planet like Jupiter, current technology would basically prevent us from escaping Jupiter’s gravity. Jupiter’s first cosmic velocity alone is 33 km/s, requiring fuel to be 1500 times the payload mass, which is practically unfeasible. To completely escape Jupiter’s gravity at 53 km/s, the fuel needs to be 100,000 times the payload mass, essentially declaring the death of propellant-based propulsion.
Why do so many science fiction novels choose the Moon or Mars as the base for space exploration? It’s easy to see from the map that escaping the Moon’s gravity requires only 2.6 km/s, with fuel only 0.8 times the payload mass; escaping Mars’ gravity requires only 5.7 km/s, with fuel only 2.5 times the payload mass. In contrast, escaping Earth’s gravity requires 12.5 km/s, with fuel 15 times the payload mass. It’s like doing maritime trade; it’s certainly more convenient in Shanghai than in Urumqi.
If we want to buy a round-trip ticket to Mars and back, the fuel-to-payload mass ratio needed is the product of the ratios for the trip to Mars and the return trip. This is multiplication, not addition, because the payload landing on Mars needs to include all the fuel required for the return trip to Earth.
Going to Mars from Earth, if not relying on Mars’ atmospheric braking, requires a total speed increment of 18.5 km/s. If the deceleration phase relies entirely on Mars’ atmospheric friction for braking (which has many engineering challenges), the minimum speed increment needed is 12.9 km/s, requiring fuel and payload to be at least 17 times the payload mass. Returning from Mars to Earth, if relying entirely on Earth’s atmosphere for braking, requires a minimum speed increment of 6.1 km/s for escaping Mars’ gravity and Mars-Earth transfer orbit, requiring fuel and payload to be at least 4 times the payload mass. Considering both trips together, the fuel mass needs to be at least 68 times the effective payload. In practice, due to the difficulty of reusing rocket engines, more fuel is needed.
The enormous gravity of each planet, combined with the exponential fuel mass requirement, makes it extremely difficult to visit a planet without resupply and return. Visiting two planets without resupply and returning is almost impossible. Therefore, if rocket propulsion technology does not make significant breakthroughs, interstellar travel is destined to be difficult and limited.
The Tight Time Window for Space Exploration
With fuel, can we just go whenever we want?
Of course not, because the relative positions of planets vary at different times. Once the departure and arrival dates are specified, a trajectory curve can be calculated, and the required fuel mass can be determined. Plotting the departure date on the horizontal axis and the arrival date on the vertical axis, and drawing a blue curve for the same fuel mass required for different departure and arrival dates, these blue curves look like rings of a tree, with the center being the optimal launch time requiring the least fuel mass.
Strangely, I couldn’t find the desired results using the Chinese name “年轮图” that the old man told me, but I could find it using its English name “Porkchop Plot.”
Porkchop Plot for Calculating Launch Windows, Source: Wikipedia
You can see that in 2005, to go from Earth to Mars, it takes at least 125 days (the red horizontal line in the figure indicates the same number of days in transit). If fuel needs to be saved, it takes at least 200 days, and there is only a short one-week time window. Departing from Earth at an unsuitable time, if you still want to keep the fuel amount within a reasonable range, it takes a very long time (e.g., departing from September to November requires at least 350 days).
Such a long time is needed because our fuel is too limited, only enough to accelerate the rocket to escape the planet’s gravity, but not enough to continuously accelerate during the long journey. During the long journey from Earth to Mars, the rocket is mostly in a shutdown state, gliding on the orbit by inertia and gravity.
After seeing the Porkchop Plot, I understood that under current propulsion technology, space exploration can never be a spontaneous trip.
Today’s Space Exploration is Like Polynesian Ocean Voyages
My grandfather liked to tell me about the maritime feats of the Polynesians. He said, Easter Island in the South Pacific is more than 2000 kilometers away from the nearest island, but a thousand years ago, Polynesians already lived there and left behind the Easter Island statues. How did the Polynesians migrate across two to three thousand kilometers to get there?
Polynesian Triangle in the South Pacific Islands, Source: Wikipedia
The migration of Polynesians between islands in the South Pacific is a feat in the history of navigation. Not only Easter Island, but Hawaii was also migrated to a thousand years ago. Because the Polynesians improved the dugout canoe, adding planks to two parallel logs, the boat’s wind and wave resistance was greatly enhanced, and more food and fresh water could be stored on the planks. Moreover, the Polynesians were very good at celestial navigation and locating based on ocean hydrology. They also needed to choose the right date to set off. Of course, a journey of thousands of kilometers, taking months, was destined to be a life-and-death struggle.
The old man who worked in aerospace said, Today’s space exploration is like the Polynesians’ ocean voyages. Modern people looking at the Polynesians’ primitive dugout canoes would think, relying only on wind power, how could they possibly cross two to three thousand kilometers of the Pacific Ocean? Today’s space exploration is similar because the speed required to escape a planet’s gravity is so fast that it has reached the limit of human fuel capabilities. Sending a little bit of payload into space requires dozens or even hundreds of times its mass in fuel. Moreover, it depends on the weather, departing in a narrow launch window.
The old man said, luxury goods are for people who have achieved nothing to show off, exploring the world is the value of life. Opening a new route on the solar system space map is far more valuable than being wealthy. Exploring the universe is certainly not for making money. But you can’t do it without money; money is a necessary resource for exploring the world. To get this money, you have to make people believe that this matter is valuable and that you are the person who can accomplish it. That’s why, even though they know interstellar travel is still far away, they still hang up the solar system space map.
Did Not Master Mathematics, Switched to Computer Science
When I was in college, among Shanghai Jiao Tong University, Fudan University, and the University of Science and Technology of China (USTC), my grandfather suggested I go to USTC. Because my grandfather wanted me to become a scientist, he felt that USTC had a quiet desk where I could build a solid foundation in mathematics and physics.
Although I was admitted to the Hua Luogeng Class for mathematical talents at USTC, I did not study mathematics well. I failed linear algebra in the second semester of my freshman year, and barely passed mathematical analysis. This was because I was always immersed in the computer room of the School of the Gifted Young, tinkering with servers and networks. I found computer science much simpler and more useful than mathematics. I didn’t know what use mathematics had, but with computer science, I could create anything I wanted, and many classmates could benefit from what I made.
My grandfather was very angry, saying that mathematics is the foundation of all sciences. I retorted that many fields in computer science do not require much mathematics. Indeed, developing most software does not require much mathematics, and many research areas in computer science only need elementary mathematics.
I also believed that information has many better properties than matter. The cost of copying information is very low, which makes open-source software and Wikipedia possible. In the material world, it is difficult to find any industrial product with such a low replication cost that it can be supplied for free.
Moreover, the transmission speed of information is naturally the speed of light, which can cross interstellar distances, while accelerating matter to speeds sufficient to traverse interstellar distances requires an impractically large amount of energy. This is the rocket fuel problem my grandfather, who worked in aerospace, told me about. Therefore, Voyager 1 chose to carry a golden record containing information about human civilization to explore extraterrestrial civilizations.
I don’t remember exactly what my grandfather said, but it was roughly the famous line from “The Wandering Earth”: “Civilization without people is meaningless.“
However, I ultimately followed my heart and switched to the computer science major in the first semester of my sophomore year.
Recently, I learned that concepts like Diffusion Model, Autoregressive, and Emergence in large models are all concepts from statistical physics. Neural networks and macroscopic objects are both complex systems and cannot be understood using reductionist methods that target individual neurons or individual atoms and molecules. In foundational research in AI, mathematics is still very important.
A person from OpenAI commented that the biggest gap for domestic large model companies is in mathematics. Therefore, OpenAI can modify Transformers, but most similar attempts by domestic companies have failed, with only DeepSeek doing a decent job.
I was good at mathematics when I was young, thanks to the mathematical enlightenment from my grandmother and grandfather, not just teaching concepts and calculations, but more importantly, helping me understand clear physical images. Just like the aerospace map on the wall of my grandfather’s house, which made me never forget the conservation of momentum and exponential logarithms, and more importantly, made me pursue the stars and the sea as a life goal.
I didn’t study mathematics well during my undergraduate years, probably because I lacked clear physical images and didn’t know what these mathematics were for. On the road to researching AI, with the aid of physical images, I found that these mathematical concepts are not so obscure and difficult to understand, and there is deep beauty behind many coincidences. I hope to gradually make up for the mathematics I missed back then.
Informatization of Life is the Direction of Evolution
I still adhere to the view I held since my freshman year, the informatization of life is the direction of evolution.
Early life transmitted information entirely through DNA. Higher organisms increasingly influence the behavior of juveniles through group behavior. The behavior of juveniles no longer depends solely on DNA but is more influenced by adult individuals, greatly improving the efficiency of information transmission.
The hallmark that distinguishes humans from other animals is the possession of language. Language is not only a carrier of thought but also a carrier of intergenerational information transmission among humans. Today, the innate conditions determined by human DNA have a significantly lower impact on individuals than the environment and education. Language is another leap in information transmission efficiency and an important milestone in the informatization of life.
Today’s large models are the first technology capable of understanding natural language besides humans, although their understanding and creative abilities are currently not as good as humans. The reason I was not optimistic about AI before but decisively devoted myself to AI after seeing the capabilities of large models is that I discovered the realistic possibility of digital life, which will bring a milestone impact on human civilization.
Compared to carbon-based life based on cells, silicon-based life has many advantages in space exploration. Humans need to eat, drink, breathe, and sleep, cannot withstand high acceleration, and cannot tolerate extreme temperatures. Chips, on the other hand, have much higher tolerance for acceleration and temperature than humans and only need electricity as an energy supply, without the need for food, drink, and breathing. Therefore, humans are a species evolved to suit the Earth’s environment, but they are not suitable for the vast universe.
For decades, the route of space exploration has been to replicate the environment for human survival on other planets or even in outer space so that human bodies can live there. However, the energy and time costs of transporting large amounts of material in space are too high, making current human technology infeasible.
I quite like Liu Cixin’s speech: “…Compared to the perilous real space exploration, they prefer to experience virtual space in VR… You promised me the stars and the sea, but you only gave me Facebook… In the long run, among the countless possible futures, no matter how prosperous the Earth becomes, those futures without space travel are all bleak…”
Why has the world become as Liu Cixin described? Going to other planets is almost everyone’s common dream. Why has capital flocked to the internet and AI, but not so much to manned spaceflight? Because in recent decades, there has been no major breakthrough in energy technology, and the vast distances and strong gravity of the universe have become almost insurmountable obstacles for human bodies to explore the universe.
The transmission speed of information is the speed of light, enough to break free from the shackles of gravity, and the vast distances do not seem so unattainable. Even if information must have a material carrier, chips are more suitable carriers for the space environment than the human brain.
I believe that AI is currently the most feasible technological route to spread human civilization to the depths of the universe. If AI can carry human intelligence on chips and can survive, reproduce, and evolve autonomously, then why can’t chips be another form of life? Life has evolved such great changes in form to adapt from the ocean to the land. Why can’t adapting to the cosmic environment be another evolution of life? I do not hope that humans on Earth will be replaced by AI, but why must life on other planets and in space be in the form of human bodies?
When we make products, we all know to find a niche market. The original meaning of this niche is the ecological niche in biological evolution. In the future, will carbon-based life and silicon-based life coexist in different ecological niches in the universe?
Unfortunately, many people today are too short-sighted, only thinking about how to quickly monetize a blockbuster app with large models within two or three years. Of course, quick monetization is not wrong, as it can raise more funds to explore the stars and the sea.
The goal of exploring the universe is too grand and cannot be achieved overnight. Today’s large models are far from being able to undertake this task, and intelligence itself is not the only technical challenge of interstellar civilization. Therefore, for now, I still need to work steadily on each research project, improve the basic capabilities of large models, and make each product well to solve the real needs of humans, so that more people believe in the value of large models and attract more talent and capital to invest in this grand vision. This is what Ren Zhengfei has always talked about, “climbing Mount Everest, laying eggs along the way.” But if the ultimate goal is just to make the next TikTok, then we will miss the stars and the sea.
Digital Life Will Make Time Travel Possible
When I was an undergraduate, I read Hawking’s “The Universe in a Nutshell,” which said that time travel is basically impossible. Knowing that something is possible but difficult is an exciting thing, but knowing that something is impossible can only be regrettable.
In 2022, I won an award at Huawei, and the prize was a physical copy of “The Universe in a Nutshell.” I read it again and found that this book is really well written. Although it is a popular science book, it does not lose its rigor, and one can read a lot of deep meanings that only researchers can understand. This book, written in 2001, even predicted the scaling law. The book says:
“Some people say that computers can never display true intelligence, no matter what that intelligence means. But it seems to me that if very complex chemical molecules can operate in the human body to give them intelligence, then equally complex electronic circuits can make computers behave in an intelligent manner. And if they are intelligent, they should also be able to design even more complex and intelligent computers.”
“Increasing the scale of the human brain through genetic engineering will eventually encounter the problem that the chemical messengers responsible for our mental activities in the body move slowly. This means that further increasing the complexity of the brain will come at the cost of speed. … Electronic circuits have the same complexity problem as the human brain. However, in this case, the signals are electronic, not chemical, and they move at the speed of light, much faster.”
“Space travel beyond our solar system is likely to require genetically engineered humans or unmanned computer-controlled spacecraft.”
“The future of science will not be as comforting a picture as depicted in ‘Star Trek’: a universe filled with many human-like races, with advanced but essentially static technology. Instead, I think we will develop biological and electronic complexity alone but rapidly.”
When I was in high school, I believed that digital life was a more suitable form of life for human exploration of the universe. But my classmates around me either laughed at me for being a pseudo-scientist, saying that machines could never understand natural language and could never think like humans, or said that digital life was not life at all, and even if it was created, it could only be a tool for humans and could not be allowed to exist equally with humans. Later, I gradually forgot this naive idea until the last wave of AI in 2017, when I started tinkering myself using the opportunity of mining, but of course, my abilities were limited, and I didn’t achieve much.
In 2022, it is an indisputable fact that machines can understand natural language. Rereading “The Universe in a Nutshell,” I also found that Hawking had the same idea as me long ago, and I am not alone. Therefore, I decided to make this dream my long-term career goal. During my entrepreneurial period, I met many real big names in the AI field. Although everyone has different views on digital life, the scaling law has long been a common belief. In the digital extension of the human world, we have the hope of making everyone’s time infinite, which may be the only feasible technical route for time travel in my lifetime.
The World is Big, As Long As the Heart is Brave Enough
There is another question, why explore the world? Isn’t it good to live a small life well?
When I was a child, my grandfather often told me the story of the Polynesians. A thousand years ago, the Polynesians migrated to Easter Island through a feat in the history of navigation. But in the following centuries, the population on Easter Island first increased sharply and then decreased sharply. Why? Because the resources on the island were too limited, and when the population swelled to the limit of capacity, civilization began to decline.
My grandfather was a geologist and spent most of the year in the field. Field exploration is not tourism; the places their geological team went to might not see anyone else for ten years, and eating and sleeping in the open was the norm. Encountering wild animals while camping in the wild or encountering geological disasters halfway up the mountain were not uncommon. In those days, there were no mobile phones or GPS, and if you got lost, you would be left in the mountains.
My grandfather said, if geologists did not explore mineral resources, today’s industrial civilization would not be possible. He believed that today’s Earth’s resources are also facing scarcity, and if we do not explore the universe, the future of humanity may not be much better than that of the Polynesians. The declining birth rates in many countries today are already repeating the same mistakes.
I am not too worried about the issue of declining birth rates. Many of China’s current problems are due to overpopulation and insufficient per capita resources, aren’t they? Moreover, for modern people, continuing life by replicating DNA through childbirth inevitably sacrifices personal time and location freedom.
Although digital life also requires time companionship, at least it is entirely feasible to travel around with digital life. The current controversy over digital life fundamentally stems from the fact that AI technology is not advanced enough, lacking human autonomy and emotions. Additionally, whether large models are a more efficient technological route for converting energy into intelligence compared to the human brain is still up for debate. But we believe these technical issues can be resolved. In April 2023, I mentioned that the inference cost of models equivalent to (the first version of) GPT-4 could be reduced by 100 times in 2024. Just look at how much redundancy there is in attention and how much profit margin NVIDIA’s chip pricing has. Most people didn’t believe it, but it has already come true today.
Ten years ago, a friend gave me a saying: “The world is vast, as long as the heart is brave enough.”
Ten years have passed. From the course evaluation community project I did at the school LUG (Linux User Group) to the intelligent network card and high-performance data center system research at MSRA (Microsoft Research Asia), to the AI training and inference framework and high-performance data center network at Huawei, and now to AI entrepreneurship, I have been exploring the world. I never expected to speak as an alumni representative at the anniversary return-to-school commemorative conference.
According to my travel records since 2012, I have traveled to 80 cities and made 417 trips (counting round trips as two) in the past 12 years, which can be considered physical exploration of the world.
But when I open the world map, my physical footprint is still very limited. If the AGI (Artificial General Intelligence) roadmap were drawn like a solar system space map, I probably haven’t even stepped out of the first stop, low Earth orbit, and I don’t even know what the entire space map looks like. This is also my motivation for continuously exploring the world.
Wang Xiaobo said: “That day I was twenty-one years old, in the golden age of my life. I had many extravagant hopes. I wanted to love, to eat, and to become a cloud in the sky, half bright and half dark, in an instant. Later I realized that life is a slow process of being hammered, people grow old day by day, and extravagant hopes disappear day by day, eventually becoming like a cow being hammered. But when I turned twenty-one, I didn’t foresee this. I thought I would always be vigorous, and nothing could hammer me.”
Unfortunately, many of my classmates have become like this. But I still believe in the saying “The world is vast, as long as the heart is brave enough,” and many of my classmates, like me, are still exploring the unknown world.
I hope that when I am as old and shaky as that old man, I can invite a few children to my home, point to the solar system space map on the wall, and say that life now spans every planet in the solar system, and when I was young, I helped lay a few bricks for this new form of life…