They seek to create a “brain” which unravel the enigmas of the Universe that Einstein and Stephen Hawking have not deciphered

Sometimes Albert Einstein he described scientific theories as “free inventions of the human mind”. But in 1980, Stephen hawking, the famous cosmologist from the University of Cambridge, had another thought. He argued that the appeal Theory of everything (one that would explain all known physical phenomena in one statement) might be achievable, but that the finishing touches would likely be done by computers. “The end may not be in sight for theoretical physics,” he said. “But it could be from the point of view of theoretical physicists.”

The theory of everything is not open yet, but with computers taking over many of life’s tasks – translating languages, recognizing faces, driving cars – it’s not so crazy to imagine them taking over. of the Hawkings and Einsteins world.

Computer programs like DeepMind’s AlphaGo continue to discover new ways to beat humans at games like go and chess, which have been studied and played for centuries.

Why couldn’t one of these wonderful learning machines launch a huge astronomical catalog, discern a set of new fundamental particles, or discover a wormhole in a distant galaxy, like the one in the movie Interstellar? At least that’s the dream. To think otherwise is to participate in what physicist Max Tegmark calls “carbon chauvinism” (a term that refers to those who believe that all life in the Universe must be the same as that on Earth, resulting from carbon-based chemicals).

By the time artificial intelligence tells you what lies beneath the laws of physics, you must be very scared or very excited, depending on your perspective.

Max Tegmark, physicist

In November, the Massachusetts Institute of Technology (MIT), where Tegmark is a professor, received a check from the National Science Foundation of the United States and opened the metaphorical doors of the new Institute for Artificial Intelligence and Fundamental Interactions.

The institute, headed by particle physicist Jesse Thaler, is one of seven created by the Foundation and the United States Department of Agriculture as part of an effort to advance work on the artificial intelligence. Each receives $ 20 million in five years.

“What I hope is to create a place where researchers from different fields of physics, as well as researchers working in computer science, machine learning or AI, can come together and talk and teach each other things. . ” Thaler said. “At the end of the day, I want to have machines that can think like a physicist,” he added.

Your tool in this business is a type of artificial intelligence (AI) known as neuronal red, which mimics the way neurons in a brain communicate.

Unlike other systems, such as IBM’s Watson, which are loaded with human and scientific knowledge, neural networks are designed to learn as you go, as humans do. By analyzing vast amounts of data for hidden patterns, they quickly learn to distinguish between cats and dogs, recognize faces, reproduce human speech, report bad financial behavior, and more.

“We hope to discover all kinds of new laws of physics,” Tegmark said. “This AI has shown us that it can rediscover the laws of physics.”

Experiences

In 2019, in what was sort of a startup test, Tegmark and a student, Silviu-Marian Udrescu, took 100 physics equations from a famous textbook (The Feynman lectures on physics, by Richard Feynman, Robert Leighton and Matthew Sands) and used them to generate data which was then fed into a neural network.

The system sifted through the data for the patterns and patterns, and retrieved the 100 formulas. “As a human scientist, you try many different strategies at once,” the researchers wrote in an article published in Scientific advances. “And if you can’t solve the whole problem in one go, try turning it and breaking it down into simpler pieces that can be handled separately, recursively running the entire algorithm again in each piece.”

In another more difficult experiment, Tegmark and his colleagues showed the network a video of rockets flying around and asked them to predict what would happen from frame to frame. Regardless of the palm trees in the background. “Ultimately, the computer was able to discover the essential equations of motion,” he said.

Finding new particles in a place like CERN’s Large Hadron Collider would be a snap, Tegmark said. AI loves big data, and collider data is reaching thousands of terabytes per second. What does it matter that a new particle does not appear in CERN data since the discovery in 2012 of the Higgs boson (a kind of missing link in the subatomic world), despite years of frantic testing.

For now, Tegmark conceded, there are limits to what can be skimmed off by the algorithm’s recursive problem-solving method, a practice known as regression. Although the machine can retrieve the fundamental laws of physics from a lot of data, it cannot yet access the deep principles (such as quantum uncertainty in quantum mechanics or relativity) that underlie these formulas.

“By the time AI tells you that, then we got to general artificial intelligence, and you must be either very scared or very excited, depending on your perspective,” Tegmark said.

“The reason I’m working on this, honestly, is because what I find most threatening is that we are building a super powerful AI and have no idea how it works, truth? “

Thaler, meanwhile, who heads the new institute at MIT, said he was once skeptical about artificial intelligence but is now an evangelist. He realized that as a physicist he could encode some of his knowledge into the machine, which would then give answers that he could more easily interpret.

“It becomes a dialogue between man and machine in a way that becomes more exciting,” he said, “instead of having a black box that you don’t understand making decisions for you.

He added, “I don’t particularly like to call these techniques ‘artificial intelligence’ because this language masks the fact that many AI techniques have rigorous foundations in math, statistics, and computer science.”

Yes, he admits, the machine can offer much better solutions than him despite all his training: “But in the end, I still have to decide which concrete goals are worth achieving, and I can aim for ever more ambitious goals knowing If I can rigorously define my goals in a language the computer understands, then AI can deliver powerful solutions.

“One of the reasons AI has been so successful in solving games,” said Thaler, “is that games have a very well defined notion of success.” He added: “If we could define what success means for the laws of physics, that would be an incredible step forward.”

Eliminate romance. A supercomputer is not a creature like a cat, it is just a working algorithm.

Jaron Lanier, computer engineer

The future that opens

Some physicists believe that the next big jump it will come with the advent of AI in quantum computers. Unlike classical computers, which manipulate bits that can be 1 or 0, so-called qubits in quantum computers can be both. According to quantum physics, this is how elementary particles behave on the smallest scales in nature, and it allows quantum computers to process large amounts of information simultaneously.

Could a machine produce the abstract and non-intuitive principles of quantum theory or Einstein’s principles of relativity? Could he produce a theory that humans cannot understand? Could we end up in the Matrix, or in a world managed by SkyNet, like in Terminator?

Jaron Lanier, a computer engineer who now works with Microsoft, said the IT field was full of romantic hype about the power and threat of super-intelligent machines. Eliminate romance. It’s not a creature like a cat, it’s just a working algorithm. “

Steven Weinberg, Nobel Prize winner and professor at the University of Texas at Austin, said it was “a disturbing thought” that humans weren’t smart enough to fully understand the final theory. “But I suspect that in this case, we won’t be smart enough to design a computer capable of coming up with a final theory, either.”