A Reality Check on Quantum Computers
wsj.com
wsj.com
The author [1] is trivializes the notion of computational machine. Calculating using a classical/quantum computer vs doing the experiment, differ in at least two important respects.
(1) Computers are fully programmable while experimental setups are usually specific to calculating a choice few parameters. The universality of various computational models implies exactly this: a well designed physical machine can simulate any other physical system, assuming you have the correct model of the system. A carbon-spectra calculating experimental setup does not have this universal programmability.
(2) A computer often yields the correct answer with a lot fewer resources, again assuming your model of the simulated system is correct. This is because, a well designed computation can ignore unneeded part of the model, which the experimental system must include. In fact, computers are precisely useful where we can get such an advantage, and useless where we can't. As the author points out this is not possible for some carbon atoms, and it also is not possible for medical drugs. But it is possible for a wide variety of other systems - say rockets to the moon.
[1] a Nobel prize winner, so take my words with a grain of salt.
That way you get sort of the best by both worlds, you get a (limited) programmability and you get easier observations than you would have access to just by for example designing a molecule encapsulating the problem you want to solve, as you would have to observe it "doing its stuff" somehow.
Just because it can't crack crypto doesn't mean a large-scale version of a setup like this can't be valuable. The field of quantum chemistry needs to use huge supercomputer clusters to simulate stuff like this.
Having a proper quantum computer with enough fully programmable bits to demonstrate quantum supremacy would be valuable, but (as the author states) the meaning of quantum supremacy is tricky, and we currently have achieved "quantum supremacy" only if we use a very limited, "cheating" definition of quantum supremacy.
Where does Wilczek claim that? The only sentence I can see where he arguably comments on the programmability of Google's device is "the computation that Sycamore performed is very specialized". But this is before he brings up the carbon example, and he never says the programmability of Sycamore is comparable to the programmability of carbon.
Indeed, Sycamore is programable. It's obviously not capable of arbitrary-length computations, because the noise level is too high to be fault tolerant, but it has a certain accessible state space in which it can be arbitrarily configured. Here's Scott Aaronson if you don't believe me (emphasis mine):
> [Question:] Even so, there are countless examples of materials and chemical reactions that are hard to classically simulate, as well as special-purpose quantum simulators (like those of Lukin’s group at Harvard). Why don’t these already count as quantum computational supremacy?
> [Answer:] Under some people’s definitions of “quantum computational supremacy,” they do! The key difference with Google’s effort is that they have a fully programmable device—one that you can program with an arbitrary sequence of nearest-neighbor 2-qubit gates, just by sending the appropriate signals from your classical computer.
> In other words, it’s no longer open to the QC skeptics to sneer that, sure, there are quantum systems that are hard to simulate classically, but that’s just because nature is hard to simulate, and you don’t get to arbitrarily redefine whatever random chemical you find in the wild to be a “computer for simulating itself.” Under any sane definition, the superconducting devices that Google, IBM, and others are now building are indeed “computers.”
https://www.scottaaronson.com/blog/?p=4317
Therefore, I think abdullahkhalids's interpretation is right: Wilczek's carbon analogy is inapplicable.
The more I think about the distinction people are drawing, the fuzzier it gets in my mind.
Frank Anthony Wilczek (/ˈwɪltʃɛk/;[2] born May 15, 1951) is an American theoretical physicist, mathematician and a Nobel laureate. He is currently the Herman Feshbach Professor of Physics at the Massachusetts Institute of Technology (MIT), Founding Director of T. D. Lee Institute and Chief Scientist Wilczek Quantum Center, Shanghai Jiao Tong University (SJTU), Distinguished Origins Professor at Arizona State University (ASU) and full Professor at Stockholm University.
It's a similar line to AI - if the people talk about it like it has the potential to (and eventually will) destroy the world, they probably don't know much about it (in that case they're conflating sentience with intelligence, and ignoring the word "artificial).
The people who are working on it right now will - like with AI - tell you that it's in its infancy, that its current applications are more limited than they are made to appear in pop-culture, and that (generally) things are harder and take longer than people pretend.
Which is exactly what the author of the WSJ article is getting at (because he's not a hawker - he knows):
> "There’s little doubt that, in the long run, computers that exploit quantum features of matter will dramatically enhance our ability to address useful problems. But we’re not there yet, nor is success guaranteed. For the foreseeable future we will have, at best, a “quantum advantage” in well-chosen applications, not “quantum supremacy” along a broad front."
Surely we can all agree that it doesn't take human or better intelligence to destroy the world.
...how about a nice game of chess?
I understand that this is the frontier and no one has any definite answers, but I hear everything from "trivial" improvements to "insta-crack AES-256" (thousands of trillion trillion trillion trillions times faster).
Does anyone have a well informed upper/lower bound estimate?
I can tell you that Shor's algorithm for factoring takes O((log n)^2 (log log n)(log log log n)) (we can round that up to O((log n)^4) if you want) to a the best known classical algorithm of
O(exp(1.9 * (log n)^(1/3) (log log n)^(2/3)))
But unless I tell you how fast each gate is, that tells you nothing about the constant time factors.
Also, all these quantum algorithms are constructed out of logical gates, but each logical gate might be built out of 10 physical gates, using some error correcting code.
Since you really need to know the physical characteristics of the particular architecture to know how much error correction you need, it's hard to even say how many gates something will use and how fast they are until we pick a particular architecture.
All we can really say is that given a big enough number, a quantum computer can factor it faster than a classic computer.
Iff quantum computing is only useful at breaking encryption, I don't see the point in funding quantum computing research.
Quantum computing is not expected to have large impacts on machine learning at this time after Ewin Tang's paper. There's an especially large amount of fluff in this area, though.
Improvements are believed to be exponential when simulating physical quantum systems.
Quantum computing have be designed to solve a single problem : the simulation of quantum systems on a classical computer is very slow, lets build a quantum computer so that it will be fast.
It might seem like a fringe use case but it matters (a lot) to industrials and researchers in a wide variety of topics.
You can get immense speedups for a very particular set of problems, turning them from "can't possibly complete ever" to solvable; so it would be like an assistive device for these niche tasks, but it would be very important for them because otherwise they're not really practical to solve.
Whether they prove to be right or wrong, their motives aren't primarily scientific.