Google has achieved quantum supremacy but not really

Table of contents:

Google has achieved quantum supremacy but not really
Google has achieved quantum supremacy but not really
Anonim

Quantum computers, in theory, could revolutionize computing. But this will not happen in the foreseeable future, and Google's statements about achieving "quantum supremacy" are a hype for the sake of hype.

Quantum computer: the cryostat contains a cooled 54-qubit Sycamore processor / © Forest Stearns, Google AI Quantum Artist in Residence

Quantum computers will be able to perform tasks that take non-quantum computers thousands of years to complete. However, while the technical capabilities of new products are limited. Attempts to show their advantages over classical computers - the so-called quantum superiority - do not stand up to serious criticism. Let's try to understand why.

What a quantum computer gives

When working, a quantum computer does not rely on bits, as a normal computer, but on qubits (quantum bits). The bit has the value 0 or 1, but the qubit can be in the states | 0⟩ and | 1⟩, as well as in their superposition. That is, with a certain probability, its value can simultaneously be either an analogue of zero, or an analogue of one. In other words, instead of the usual state, zero or one value is described by a continuous variable, the so-called quantum amplitude. A typical computer with a dozen working bits can have 2 to the tenth power of clear simple states (on the order of a thousand). A quantum will have the same number of continuous variables: that is, its "content" will be radically more complex.

This shows that even a basic qubit is much more complex than a regular bit. Equally complex is the interaction of quantum bits relative to ordinary ones. If one bit can with some probability have two different values ​​at once, in one operation with it it is possible to process both of these possible states at the same time. Due to this, groups of qubits should achieve enormous computational superiority over groups of ordinary bits. Of course, superiority can be achieved only with the use of special algorithms that can take into account the new capabilities of quantum computers.

The greatest effect quantum computing will give in the field of neural networks, which today are associated with the possibility of creating artificial intelligence, comparable to natural. Shor's quantum algorithm allows you to quickly factor large numbers. This means that it can "break" most of the existing powerful cryptographic systems.

So you can hack a credit card, and even compete for someone else's wallet with cryptocurrency. There are many more possible applications of quantum computing, but artificial intelligence and breaking cryptography are currently the most understandable theoretically.

Why haven't they taken over the world yet

Despite all this, there are still no sufficiently efficient quantum computers. We emphasize: at the existing technical level, it is generally unknown when they may appear. Moreover, among scientists, including Russian ones, there are those who consider this problem to be fundamentally unsolvable for really large quantum computers.

The fact is that the smallest of practically useful quantum computers must have between a thousand and a hundred thousand qubits. This means that there will be at least 2 thousandth continuous variables in it - or about 10 to the three hundredth power. The number of all particles in the universe is less than 10 to the hundredth power. That is, the number of continuous states in a quantum computer of useful power will be such that its operation will become almost impossible to control and make it sufficiently error-free.

The number of continuous states in a quantum computer of useful power will be such that its operation will become almost impossible to control and make it sufficiently error-free.

If a thousand bits of an ordinary computer can contain a small number of errors due to the incorrect operation of one bit (transistor), then this can be easily corrected by duplication: the processor works "bypassing" the incorrectly operated bit. However, it is impossible to constantly run calculations bypassing the "wrong" continuous variable. A variable is much more complex than a simple zero or one. This is both the strength of a quantum computer and its weakness. Due to this very complexity of the value of the qubit, it is devilishly difficult to control possible errors in it.

To solve the problem, an error correction option is proposed. If the probability of an error when switching a qubit is not higher than a certain value, then you can split one logical qubit into several physical ones and try to correct the errors "stepwise", since it is easier to do this by dividing the task into stages. This looks like a good solution. However, in the end, a useful quantum computer will start not from a thousand, but from a million cubits. That is, the task of controlling his mistakes will again become much more difficult.

Because of all this, the Commission for Combating Pseudoscience of the Russian Academy of Sciences has long published material stating that large and therefore practically useful quantum computers will not be developed in any foreseeable future. That is, yes, theoretically they are possible (as well as, for example, moving faster than the speed of light), but in fact there are no practically conceivable ways to this.

What Google is proud of

Researchers at the American Internet giant published an article in Nature in which they said they showed - for the first time in world history - "quantum superiority." That is, they presented such a quantum computer that can solve a problem that is practically unsolvable for ordinary supercomputers.

Sycamore 54-qubit processor / © Erik Lucero, Research Scientist and Lead Production Quantum Hardware

To do this, they used the Sycamore quantum processor with 54 qubits at once, of which 53 can be used simultaneously. The very creation of this processor is an outstanding achievement. The total number of its continuous variables is 9,007,199,254,740,992 (2 to the power of 53). That's nine quadrillion (million billion). Controlling computational errors in such a processor is incredibly difficult. That Google was able to make this processor at all is a tremendous achievement, at the forefront of what is possible for humanity today.

But, as we noted above, a practically useful quantum computer starts from a thousand qubits without error correction and from a million qubits with it. Thus, Sycamore cannot even come close to providing a practically useful quantum computer.

How did Google use it to demonstrate "quantum superiority" over classical computers? Simple: the company's researchers specially selected a task for him in which a quantum computer should cope much better than usual. A sequence of commands was entered into the computer, after their execution, lines of 53 numbers were read, each of which corresponded to the state of the processor's qubits. This task was performed many times - according to the same principle as in the benchmark program on a regular computer.

The company's researchers specially selected a task for him, in which a quantum computer should cope much better than usual.

After the results were recorded, they were compared with the statistics expected to perform such a test. Since the benchmark performed well-known sequences of commands, the statistics of the results can be predicted with fairly high accuracy.

The test itself described above is pure "benchmark for the sake of benchmark". It has no conceivable practical application. But the authors of the corresponding article in Nature who worked for Google, using this benchmark, were able to state the following:

“Our Sycamore processor [completed the test task] in 200 seconds … Our benchmarks indicate that a similar task for a classic supercomputer would take about 10 thousand years. This jump in speed, an experimental realization of quantum supremacy."

The area of ​​quantum superiority is shown in red: where the repetition of computations of a quantum computer will take more than 1-10 thousand years from the usual / © Frank Arute et al

Of course, if someone completes one task in 200 seconds and another in ten thousand, then the superiority is evident.In this case, “quantum superiority”. After all, it is defined as the ability of a quantum computer to do what an ordinary computer is practically unable to do. Modern supercomputers will break down before ten thousand years of continuous operation, that is, they cannot complete the Sycamore task at all.

Why Google is proud of this in vain

It looks simple. The folks who work for Google built a computer with no practical use and picked up a task for it that a quantum computer - even useless in practical computation - should still do better than a regular supercomputer. It's as if we took the one-legged champion and let him race against Usain Bolt, forbidding him to use the other leg. It seems a little dishonest, but formally - yes, the one-legged showed an advantage over the two-legged.

Not really. As noted by researchers from IBM, the authors of the work in Nature "a little" played along with the brainchild of Google. They evaluated the execution of the test program by Sycamore, assuming that a classic supercomputer would count on the same instructions using RAM. However, in real life, supercomputers and computers generally have more than just random access memory.

The IBM folks figured that if a supercomputer used both RAM and hard disks when performing the same test, it could do it in 2, 5 days, or a couple of hundred thousand seconds. This is a thousand times slower, that is, we are no longer talking about "quantum superiority" in the literal sense.

Of course, a quantum processor performed a task specialized for it a thousand times faster than a classical one. But what is the point if such problems do not occur at all outside the benchmark for "proving" the superiority of quantum computers?

But what is the point if such problems do not occur at all outside the benchmark for "proving" the superiority of quantum computers?

To summarize: Google tried to cut the hype on the "proof of quantum superiority" and involved Nature in its dubious venture. In fact, there was no evidence of quantum superiority.

Moreover, both Russian researchers and IBM doubt in general that quantum computers will ever show practical superiority over classical ones in the conceivable future. They will add - yes, but you should not expect any miracles from them. In this century, artificial intelligence will not be made on their basis and your debit card or crypto wallet will not be hacked. Be careful not to trust sensational headlines without checking everything underneath to the very end.

Popular by topic