Theory of Everything
The Translation Bureau is a subordinate translation team, focusing on science and technology, business, workplace, life and other fields, focusing on introducing new foreign technologies, new perspectives, and new trends.
Editor’s note: “Theory of Everything” is a theory that encompasses all human cognition of the operation of nature, that is, the unity of all physical laws. Hawking said that the “theory of everything” may be achievable, but its final realization may be done by computers. This article is translated from the New York Times, author Dennis Overbye, the original title is “Can a Computer Devise a Theory of Everything?”, I hope to inspire you.
At one time, Albert Einstein described scientific theory as “the free invention of human thought.” But in 1980, the famous Cambridge University cosmologist Stephen Hawking had another idea. In a speech that year, he proposed that the so-called “theory of everything” may be achievable, but its final realization may be done by computers.
The theory of everything has not yet appeared, but computers have begun to take over many trivial things in life, such as translating languages, recognizing faces, driving cars, recommending dates and so on. In this way, it is not so crazy to imagine them taking over everything in the world.
Image source: Alex Eben Meyer/The New York Times
Computer programs like DeepMind’s “AlphaGo” continue to discover new ways to defeat humans in games such as Go and Chess, which have been studied by humans for centuries. Why these amazing learning machines can’t release the petabyte-level data compiled by the Large Hadron Collider, distinguish a new set of elementary particles, or discover another galaxy outside the solar system wormhole, just like the galaxy in the movie “Interstellar” What?
At least this is imaginable. If you don’t think so, you will fall into the physicist Max·Teg What Max Tegmark said “Carbon Chauvinism”. In November, the Massachusetts Institute of Technology (Massachusetts Institute of Technology) where Tegmark is a professor cashed a check from the National Science Foundation and opened the Institute for Artificial Intelligence and Basic Interactions. Artificial Intelligence and Fundamental Interactions).
The institute is one of seven institutes established by the foundation and the U.S. Department of Agriculture as part of promoting artificial intelligence work nationwide. Each location will receive US$20 million in funding within five years.
This institute at MIT is led by particle physicist Jesse Thaler and is the only institute specializing in physics research. It includes more than 20 scientists from various fields of physics, they come from Massachusetts Institute of Technology, Harvard University, Northeastern University and Tufts University.
Syler said on a Zoom conference call: “I hope to create a platform where researchers from different fields of physics, as well as researchers in computer science, machine learning or artificial intelligence can gather and have conversations. Transfer knowledge to each other. Eventually, I hope to make machines that think like physicists.”
Rediscover the basic laws
Their tool in this area is neural network. Unlike so-called expert systems, such as IBM’s Watson, which carries human and scientific knowledge, neural networks are designed to be like the human brain. By analyzing large amounts of data to find hidden patterns, they can quickly learn how to distinguish between dogs and cats, recognize human faces, copy human language, recognize financial misconduct, and so on.
Tegmark said: “We hope to discover all kinds of new laws of physics. We have proved that it can rediscover the laws of physics.”
Last year, Dr. Tegmark and a student Silviu-Marian Udrescu learned from a famous textbook (Richard Feynman, Robert Leiden and Matthew Sands’ “Feynman Physics Lectures”) extracted 100 physics equations, and used them to generate data, and then these The data is input into the neural network, and the system filters the data to find the rules.
“Like a human scientist, it will try many different strategies in turn,” the researchers wrote in a paper published in Science Advances last year. “If it can’t solve all the problems at once, it will try to convert the problem into simpler parts that can be solved individually, and recursively restart the complete algorithm on each part.”
In another more challenging experiment, Tegmark and his colleagues showed a video of a rocket flying around to the Internet and asked it to predict from the What happens from one frame to the next, regardless ofPalm trees in the view. In the end, the computer can discover the basic equations of motion.
Tegmark said that finding new particles in places like the Large Hadron Collider at CERN will be a breeze. Artificial intelligence likes big data, and the collider’s data can reach several gigabytes per second. Since the discovery of Higgs in 2012Boson, although people have been A wave crest has been madly checked, but no new particle has appeared in the data of CERN.
“These are the curves of human attention,” Tegmark said. “In the next 10 years, for studying physics, machine learning will be as important as mastering mathematics.”
He admitted that the current algorithm has limited effects in solving problems by recursive methods. Although this machine can retrieve the basic laws of physics from a large amount of data, it has not been able to derive the deep principles behind these formulas, such as quantum uncertainty or relativity in quantum mechanics.
Tegmark said: “When the artificial intelligence comes back to tell you this, we have reached the average level of artificial intelligence. You should be very scared or very excited about this. To be honest, I study this The reason is: I find that the most dangerous thing is that if we build a super powerful artificial intelligence but don’t know how it works, it’s scary.”
“Man-machine dialogue”
Seller is the head of the new institute at MIT. He said that he used to be skeptical of artificial intelligence, but now he has become a supporter. He realized that as a physicist, he could encode some of his knowledge into the machine, and the machine would give him answers that were easier to explain.
He said, “This will become a dialogue between humans and machines. It will become more exciting, not just a black box that makes decisions for you but you don’t understand.”
He also said, “I don’t particularly like to call these technologies ‘artificial intelligence’ because this statement obscures a fact.That is, many artificial intelligence technologies have a rigorous foundation in mathematics, statistics, and computer science. “
Recently, Thaler and his colleagues input a large amount of data from the Large Hadron Collider into the neural network, which collides with protons to find new particles and forces. Protons are the building blocks of atomic matter. They are made up of quarks and gluons.small particles . When protons collide, these smaller particles are ejected. To better understand this process, his team asked the system to distinguish between quarks and gluons in the collider data.
Researchers won’t tell the computer anything about quantum field theory. I won’t tell you what quarks or gluons are on a basic level. All they are is a mess of data, and then let the computer divide them into two categories. . And it can do it.
In other words, without knowing what quarks and gluons are, the system successfully identified and distinguished quarks and gluons. Saylor said that if you ask whether there is a third type of object in the system’s data, the system will start to discover that quarks are not just an entity, but exist in different types—the so-called up quarks and down quarks.
“When you give it more flexibility to explore, it starts to learn,” he said. “It doesn’t yet know quantum field theory, but it knows to look for patterns. This is what I am surprised that machines can do. He added that this work will help collider physicists to clarify their findings .
In a Zoom talk, Thaler showed the neural network used in the Quark-Gluon project, which he called “a stupid cartoon.” It looks like a bunch of colorful rubber bands, but it actually represents several layers of processing, involving about 30,000 nodes or “neurons”, and information is collected and transmitted in this process.
He said: “If you don’t think the running time is long, you can train this small network on your laptop.”
The beginning of quantum
“AIOne of the reasons for this success in solving game problems is that,” Saylor said, “the game has a very clear winning rule. He also said, “If we can define what it means to win the laws of physics, it will be an incredible breakthrough.” In 5 to 10 years from now, I will do what you want to do and find a replacement for particles. The standard model of physics can replace the equations of Einstein’s theory of general relativity.
Some physicists believe that with the advent of artificial intelligence on quantum computers, the next major leap will come. Unlike classical computers, where the bits are 0 or 1, the so-called qubits can be simultaneously It is 1 or 0. According to quantum physics, this is how elementary particles behave on the smallest scale in nature, which allows quantum computers to process large amounts of information at the same time.
The Massachusetts Institute of Technology (MIT) mechanical engineer and quantum computing expert Seth Lloyd (Seth Lloyd) said that this type of machine is still in its infancy, but the prospects are good. He is not a member of the newly established Artificial Intelligence Institute at MIT.
“The basic idea is that quantum systems can produce patterns that are difficult for classical systems,” Lloyd said. “Therefore, perhaps quantum systems can also recognize patterns recognized by classical systems.”
Or, as Illinois BarDawei Deputy Director of Research, Yafermi National Accelerator Laboratory Joe Lucen said: “To borrow Richard Feynman’s words, if you want to use artificial intelligence to discover Quantum world, you should use quantum artificial intelligence.”
Maria Spiropulu, a physicist at the California Institute of Technology, pointed out that “the literature on quantum artificial intelligence and quantum algorithms is increasing. They can solve what we previously thought Unsolvable problem.”
“It’s just a running algorithm”
How far this matter can go depends on who you ask. Can a machine produce the profound and unintuitive principles of quantum theory, or Einstein’s theory of relativity? Will it produce a theory that we humans cannot understand? Will we end in the Matrix world like the “Terminator” series?
I randomly surveyed some theoretical physicists and asked if they were ready to be replaced. Jaron Lanier, a computer engineer now working for Microsoft, said, “The way you ask questions is even more confusing.” He said that the field of computer science is full of exaggerated claims about the power and threat of super-intelligent machines. . “We should ask questions from a computational perspective, not a literary perspective. The algorithm is not a creature like a cat, it is just a running algorithm.”
Steven Weinberg, a Nobel laureate and professor at the University of Texas at Austin, believes that humans may not be smart enough to understand the ultimate theory of everything. This is A “disturbing thought”. But I suspect that in that case,” he wrote in an email, “we would not be smart enough to design a computer that would find the final theory. “
Harvard University physicist Lisa Randall wrote: “I can easily imagine computers finding equations that we don’t know how to explain. But this is no different from many measurements that we can’t explain. “
Arkani Hamed (Arkani-Hamed) is a theorist at the Institute for Advanced Study in Princeton, New Jersey. He disagrees with the idea that computers will discover some profound things that humans cannot understand. “This does not reflect the characteristics of the laws of nature we have seen for centuries. Since then, the laws of nature we have seen are based on more abstract, simpler and deeper mathematical ideas.”
For example, if Isaac Newton came back from the dead, Akane HaMead said he It will effortlessly catch up with the progress of contemporary physics.
The cosmologist Michael Turner of the Kavli Foundation in Los Angeles says it doesn’t really matter where our ideas come from, as long as these ideas are tested before we rely on them. .
“So where do we get these theories or paradigms?” It may come from deep principles-symmetry, beauty, simplicity-philosophical principles or religion,” he said. “As machines become more intelligent , We can add them to the resource list. “
Edward Witten, also from the Institute for Advanced Study in Princeton, pointed out that although the theory of everything does not yet exist, it may appear in the next century. “If a machine shows interest and curiosity in physics, I will definitely be interested in talking to it.”
Translator: Jane
Recommended reading: The easiest way for ordinary investors to accumulate wealth is to “hold for a long time”