Sabine
patreon
Quantum Winter Is Coming
🕑 Added 2022-11-05 13:00:08 +0000 UTCQuantum technology current attracts a lot of attention and money, which might explain why they’re missing on my end. But it’s not just governments who think quantum everything is where your taxes should go, business investors and companies are willing to putting in big money, too. This has had a dramatic impact on quantum physics research in the past decade. It’s also created a lot of hype, especially around quantum computing. But if so much of quantum computing is hype then why are companies like Google and IBM pouring so much money into it, what’ll happen when the investment bubble bursts, what’s the “quantum winter”, and what does it mean for all of us? That’s what we’ll talk about today.
There are several different quantum technologies, and as a rule of thumb, the fewer headlines they make, the more promising they are. Take for example, quantum metrology. That’s not a mispronunciation of meteorology, that’s making better measurements with quantum effects. You basically never read anything about this. But it’s really promising and already being used by scientists to improve their own experiments. You can learn more about this in my earlier video.
On the other hand, you have those quantum things that you read a lot about but that no one needs or wants, like the quantum internet. And then there is quantum computing which according to countless headlines is going to revolutionize the world. Quantum computers are promising technology, yes, but the same can be said about nuclear fusion and look how that worked out.
A lot of physicists, me included, have warned that quantum computing is being oversold. It’s not going to change the world, it’ll have some niche applications at best, and it’s going to take much longer than many start-ups want you to believe.
Though I admire the optimism of the quantum computing believers and also the vocabulary. For example, there’s a startup called “multiverse” that is “Working with customers in more than 10 verticals”. Some of them don’t find the way back from the bathroom.
Or here’s one called “universal quantum” which has a “fault tolerant team” that “embraces entanglement” and helps everyone find a “super position”. If you look at them, do they collapse?
This company also recently published a blogpost titled “Six reasons Liz Truss needs a quantum computer” which explains for example, “Quantum computers are suited to modelling complex systems, making them better at forecasting both near-term weather patterns and the long-term effects of climate change.”
Last time I looked, no one had any idea how to do a weather forecast on a quantum computer. It’s not just that no one has done it, no one knows if it’s even possible, because weather is a non-linear system whereas quantum mechanics is a linear theory.
Here is an example of some recent quantum hype, from a website called “Investors Chronicle” in an article titled “Quantum computing: a new industrial revolution”. After creating a lot of fog about superpositions and interference, they explain that quantum computers are close to showing “quantum advantage” and that “Quantum advantage, therefore, can be interpreted as being when problem-solving power can be applied to real world issues, which makes it much more interesting for investors.”
That’s just wrong. Quantum advantage has indeed been demonstrated for some quantum computers but that just means the quantum computer did something faster than a conventional computer, not that this was of any use for real world issues. They just produced a random distribution that would take a really long time to calculate by any other means. It’s like this this guy stapling 5 M&M. That’s a world record, hurray, but what are you going to do with it?
Problem is, a lot of CEOs in industry and the financial sector can’t tell a bra from a ket and believe that quantum computing is actually going to be relevant for their business, and that it’s going to be relevant soon. For example, at a recent Quantum Computing conference in London, the managing director of research at Bank of America said that quantum computing will be “bigger than fire”. The only way in I can see this coming true is that it’ll produce more carbon emissions.
This bubble of inflated promises will eventually burst. It’s just a matter of time. This’ll cause a sudden decline of investment in quantum tech overall and be the start of a difficult time for research and development. This scenario has been dubbed the “quantum winter”. And winter is coming. But before we get to the quantum winter, let me briefly summarize how quantum computers work and what their problems are.
A conventional computer works with bits that take on one of two discrete values. A quantum computer instead uses qubits that can be in arbitrary *superpositions of two states. A quantum computer then works by entangling the qubits and shifting this entanglement around. Entanglement is a type of correlation, but it has no analogy in conventional computers. This is why quantum computers can do things that standard computers can’t do.
For some mathematical problems, a quantum computer can give you an answer much faster than any conventional computer possibly could. This is called “quantum advantage”. Those problems include things like factorizing large numbers into prime factors. But also calculating properties of molecules and materials without having to chemically synthesize them first. Putting these questions on a quantum computer could speed up material design and drug discovery. Quantum computers can also solve certain logistic problems or optimize financial systems. This is why, if you’re Bank of America, you think it's bigger than fire.
And like fire, quantum computing is not magic, it’s an application of standard quantum mechanics. There is no speculative new physics involved. It’s rather to the contrary. Claims that quantum computers will *not work rest on speculative new physics. But it’s one thing to say if you could build them, they’d be useful. It’s another thing entirely to actually build them.
So what does it take to build a quantum computer? First of all you need qubits. Then you have to find a way of entangling many of those qubits. And, like a conventional computer, a quantum computer needs an algorithm, that tells it how to move the entanglement around. Eventually, you make a measurement which collapses the wave-function and you read out the final state. This final state should be one that answers your question correctly with high probability. This means most importantly, a quantum computer isn’t a stand-alone device, it needs other devices for the programming and the readout. The quantum part is really just a small piece of the whole thing.
Now let’s talk about the problems. First there’s the qubits. Producing them is not the problem, indeed there are many different ways to produce qubits. I went through the advantages and disadvantages of each approach in an earlier video, so check this out if you want to know more. But a general problem with qubits is decoherence, which means they lose their quantum properties quickly.
The currently most widely developed systems are superconducting qubits and ion traps. Superconducting qbits are used for example by IBM and Google. For them to work, they have to be cooled to 10-20 milli Kelvin, that’s colder than outer space. Even so, they decoherence within 10s of micro-seconds.
Ion traps are used for example by IonQ and Honeywell. They must “only” be cooled to a few Kelvin above absolute zero. They have much longer coherence times, up to some minutes, but they’re also much slower to react to operations, so it’s not a priori clear which approach is better. I’d say they’re both equally bad. The cooling isn’t only expensive and energy-intensive, it requires a lot of equipment and it’s difficult to scale to larger quantum computers. It seems that IBM is trying to do it by breaking world records in building large cryogenic containers. I guess if the thing with quantum computing doesn’t work out, they can rent them out for people to have their heads frozen.
There are some qubits that operate at room temperature, the most promising ones of those are currently nitrogen vacancy systems and photonics on a chip. However, for both of those, no working quantum computer exists to date and it’s unclear even what the challenges may be, let alone how to overcome them.
The next biggest problem is combining these qubits. Again, the issue is that quantum effects are fragile, so the quantum computer is extremely sensitive to noise. The noise brings in errors. You can correct for those to some extent, but this error correction requires more qubits.
More qubits bring problems by themselves, for example, they tend to be not as independent as they should be, an issue known as “crosstalk”. It’s kind of like if you’re trying to write while moving your feet in circles. It gets really difficult. The qubits states are also drifting if you leave them unattended. Indeed it’s somewhat of a mystery at the moment what a quantum computer does if you don’t calculate with it. It’s like it’s difficult to calculate what a big quantum system does. Maybe we can put it on a quantum computer?
And finally there’s the issue of the algorithms: Few algorithms for quantum computers are known, an issue that goes often unmentioned because everyone is focused on the technology. Wikipedia helpfully has a list of quantum algorithms. It’s short. Several of those algorithms don’t compute anything of practical use, and for some it's not known if they lead to any speedup.
As this brief summary makes clear, the challenges are enormous. But how far along is the technology? The largest current quantum computers have somewhere between 50 and 100 qubits, though IBM has a roadmap saying they want to make it to a thousand next year. Two different approaches have demonstrated a “quantum advantage”, that is, they have performed a calculation faster than the currently fastest conventional computer could have done. However in those demonstrations of quantum advantage, the devices were executing algorithms that did not calculate anything of use.
The record breaking “useful” calculation for quantum computers is the prime-number factorization of 21. That’s the number, not the number of digits. Yes, the answer is 3 times 7, but if you do it on a quantum computer you can publish it in Nature. In case you are impressed by this achievement, please allow me to clarify that doing this calculation with the standard algorithm and error correction is way beyond the capacity of current quantum computers. They actually used a simplified algorithm that works for this number in particular.
To be fair, there have been some cute applications of quantum algorithms for simple examples in quantum chemistry and machine learning, but none of this is anywhere even close to being commercially interesting.
How many qubits do you need for a quantum computer to do something commercially interesting? Current estimates say it’s several hundred thousand to a few millions qubits, depending on what you want to calculate and how large your tolerance for errors is.
A lot of quantum computing enthusiasts claim that we’ll get there quickly because of Moore’s law. Unfortunately, I have to inform you that Moore’s law isn’t a law of nature. It worked for conventional computers because those could be miniaturized. However, you can’t miniaturize ions or the Compton wavelength of electrons. They’re already as small as it gets. Nature’s a bitch sometimes.
In the past years there’s been some noise around Noisy intermediate scale quantum computers, or NISQs for short. Those are small quantum computers in which you just accept the noise, kind of like YouTube comment sections. But no one seems to have found anything useful to do with them and the hype around them has noticeably died down recently.
I guess you understand now why I am extremely skeptical that we are anywhere close to commercially relevant applications of quantum computers. But let’s hear what some other people say.
There is for example Mikhail Dyakonov, a physics prof who has worked on quantum things much longer than I have. He’s written a book that was published in 2020 under the title “Will We Ever Have a Quantum Computer?” It has only 49 pages which is what happens if you agree to write a book but then notice half through you’d rather do something else. He finishes by answering his own question:
“No, we will never have a quantum computer. Instead, we might have some special-task (and outrageously expensive) quantum devices operating at millikelvin temperatures. The saga of quantum computing is waiting for a profound sociological analysis, and some lessons for the future should be learnt from this fascinating adventure.”
The brevity of Dyakonov’s book is balanced by another book “Law and Policy for the Quantum Age” by Chris Hoofnagle and Simson Garfinkle, who make it to a whooping 602 pages. Their book was just published earlier this year, it’s freely available online, and it has an adorable cat pic on the cover, so definitely go check it out. Hoofnagle is a professor for law and Garfinkle is a data scientist, but their book has been heavily informed by people who work in quantum computing. They look at the possible future scenarios. The most likely scenario, they say, is the “Quantum Winter” which they describe as follows:
“In this scenario (call it “Quantum Winter”), quantum computing devices remain noisy and never scale to a meaningful quantum advantage… After a tremendous amount of public and private monies are spent pursuing quantum technologies, businesses in the field are limited to research applications or simply fail, and career paths wither. If that happens, funding eventually dries up for quantum computing. Academics and scientists in the field either retool and shift, or simply appear irrelevant, even embarrassing.”
Then there is Victor Galitski, Professor at the Joint Quantum Institute at the University of Maryland who wrote in a 2021 post on LinkedIn:
“The number of known quantum algorithms, which promise advantage over classical computation, is just a few (and none of them will "solve global warming" for sure). More importantly, exactly zero such algorithms have been demonstrated in practice so far and the gap between what’s needed to realize them and the currently available hardware is huge, and it's not just a question of numbers. There are qualitative challenges with scaling up, which will likely take decades to resolve (if ever).”
Most recently, there was an opinion piece by Nikita Gourianov in the Financial Times. Nikita works on computational quantum physics at the University of Oxford. He writes “As more money flowed [into quantum computing], the field grew, and it became progressively more tempting for scientists to oversell their results… After a few years of this, a highly exaggerated perspective on the promise of quantum computing reached the mainstream, leading to… the formation of a classical bubble.”
He then points out that no quantum computing company is currently making profit and that “The little revenue they generate mostly comes from consulting missions aimed at teaching other companies about “how quantum computers will help their business”.”
I have to disagree on the final point because big companies have another way to make money from quantum computers, namely by renting them out to universities. And since governments are pouring money into research, that’s quite a promising way to funnel tax money into your business. Imagine the LHC was owned by Google and particle physicists had to pay to use it.
That’s how I think it’ll go with quantum computing: First all the smaller startups will falter because they don’t reach their milestones, venture capital will evaporate, and all the overeducated quantum computists in academia will use grant money to pay a few large companies who own the only workable devices. And while those devices are interesting research objects, they’ll not be useful for commercial applications.
I might be totally wrong of course. Maybe one of those start-ups will actually come up with a scalable quantum computing platform. I don’t know, I’m guessing as much as everyone else.
But if quantum winter is coming, what does it mean for you and me? Well, some people will lose a lot of money but that just means they had too much of it to begin with, so can’t say it bothers me all that much. There’ll also be fewer headlines about how quantum computing is supposedly going to revolutionize something or other, which I’d say is a good development. And we’ll see many people who worked in quantum computing going into other professions. Chances are in ten years you can have a nice chat about the finer details of multi-particle entanglement with your taxi driver. I don’t know about you, but I’m looking forward to quantum winter.
Comments
Michiel Peeraer
The Quantum Bubble Is About To Burst, or be surpassed by a 'new' kid on the block. I just finished reading Audrey Dussutour's book "Moi le blob" ISBN 978-2-3793-1558-9 The references started with "Adamatzky A., Advances in Physarum machines: Sensing and computing with slime mould, Springer 2016." Further reading up on "Physarum polycephalum", I encountered that Shor's algorithm (of Quantum computing fame) had been ported to it. Admittedly inefficiëntly, but still. Since it has a much lower costs than a Quantum computer (basically some agar-agar and oat flakes) this story may have legs... or at least pseudopods : "Slime mold on the rise: The physics of Physarum polycephalum" https://doi.org/10.1088/1361-6463/ab866c
Today IBM has a computer with 433 qubits... In 2025 they expect to have 4000 qubits? Anything useful to do with that? https://techmonitor.ai/technology/emerging-technology/ibm-quantum-supercomputer
Rad Antonov
Appreciate you taking the time to explain Tracey. I had looked up the difference between adding and multiplying interferometers and the equation for the adding ones had an extra term, which I understood to be for the electric field at a single dish, in addition to the cosφ term of the visibility correlation. I am pining for a planetary image that actually shows some detail.
Not sure.... In an adding interferometer, you get a constant term + interference term for each baseline pair where the constant term contains the single-dish brightness information from the source, but it also contains instrumental gain variations and other sources of noise or systematic bias. Is that the E_0 you're thinking of? There is no such term in a multiplying interferometer, just the complex amplitude and phase visibilities of each baseline pair. If the source is small-ish with respect to the primary beam (and you have good short baseline information), then a good guess can be made for the zero-spacing flux -- or rather, in a u-v plot, you can see where the short-baseline data would likely cross the amplitude axis at a baseline length of 0. For extended sources, you may not care about the total brightness, you may just be interested in the structure of the smallest features, such as right up close to a central black hole of a galaxy. Then, who cares about short-baseline information. But currently, if you need to make an image of an extended, complicated source where you need both very long and very short baseline information, then you have to observe with a single dish as well as an interferometer -- SN1006 is a classic example of this sort of source. It looks like for CMB studies, they're using bolometer arrays in an adding interferometer configuration. Hmm... I should do some reading on what they're doing to beat down their noise and systematics.
From Arvin's video (https://www.youtube.com/watch?v=RCj_BJ6BddM) starting at ~10:16, quantum computing appears to be more geared to specific tasks, such as finding the correct path through the maze, faster than with a conventional computer can and so would play a niche role in computing. For example, specific tasks from a conventional computing platform would be offloaded to a quantum computer, such as specific routines in a complex simulation, to speed the overall process up and reduce the need for supercomputers, such as with modeling the potential path of a hurricane faster to improve response. I have never looked into the details, such as Lov Grover's algorithm (at ~9:31), so Arvin's more constricted example makes sense, whereas the general claims that I hear make it sound as though quantum computing would replace conventional computing altogether. So, I can see the collapse from the lack of real need of what quantum computers do better than conventional computers because the power of conventional computers is adequate in most cases.
Rad Antonov
The E ₒ² term?
IR and optical VLBI would be a game changer. It looks like no instrumentation has been developed in the article, so their idea might be a long ways off. For anyone who might be listening in on this conversation, in normal radio interferometry, say the VLA, you can be off in your interference by up to a quarter wavelength, then rely on the redundancy of all of your baselines to give you good phase and amplitude solutions overall. Optical interferometers cannot deal with this sort of sloppiness at all, mostly because they only have a few telescopes in their array and so N(N-1)/2 baselines is too small to beat down the noise effectively. The longer the baseline, the harder it is to measure the distance between telescopes accurately enough and therefore higher noise. I see something like the STIRAP system in the article as being useful for regular, "short baseline" interferometry as well. All interferometers are multiplying interferometers which chuck out the single telescope information in favor of lower noise. By working as adding interferometers, they could keep the single telescope information and give much better final images.
Rad Antonov
Not only is the Josephson effect remarkable to observe, but it’s turned out to have powerful applications. Now back to quantum computing and offshoots, what do you make of proposals to use QEC techniques for VLBI in the optical frequencies? https://phys.org/news/2022-05-quantum-technique-enable-telescopes-size.html
That's a future industrial-techno-glitch-noise music project, Armando. 😆
Ah, right. 😕
Armando Mistral
"Stinking Heads" -- the great Grunge band of 1985. It didn't exist, but should have.
Armando Mistral
There are two realities at work here: scientific and economic. In classical investment economics the two had some correlation, more or less in real-time (see the Chicago School of Efficient Markets.... or not). That correlation has been broken by the vast wealth awash throughout our civilization now, which enables Stupid Money to pursue objectively absurd technologies for protracted periods before the inevitable crunch. Money can now afford to be stupid, it couldn't before. There is therefore no meaningful information value in how much money is being plowed into something, indicating that something's viability in objective reality (subjective realities, see a lot of Social Media, need not apply). Results may vary.
I know, what he's quoting seems mostly explorative, speculative.
There's nothing there that proves Sabine entirely wrong, but maybe I'm misunderstanding Wang? Is everything he's discussed speculative or has any of it actually happening?
I respect Brian Wang's analysis in several areas. This one?? https://www.nextbigfuture.com/2022/11/ionq-quantum-computers-versus-skeptics.html
My first thought was that an experimentalist would not necessarily be trying to demonstrate superdeterminism, but they might accidentally find it when the statistics of their system come out wonky. Then, I reread the video transcript, and current qubit systems undergo crosstalk which makes the systems look less independent than they should be -- damn systematics!
Rad Antonov
The experimental challenge seems to me is to rule out systematic effects and an experimentalist would be reluctant to sign on for that because they wouldn’t exactly be looking for a prediction, just a hint that something anomalous is going on.
Any unlabelled heads will be cleared out each Friday afternoon.
Rad Antonov
The idea behind those experiments is to look for autocorrelation in repeated measurements where quantum mechanics predicts the outcomes will be random. Repeating them at small time intervals and low temperatures will minimize the thermal fluctuations that can change the settings between measurements.
1. remember to label your frozen head, 2. do not take anyone else's frozen head, and 3. any frozen heads that start to stink will be thrown away, no exceptions
Thanks for laying out the history. I've never been one for the instrumentation side of astronomy, but I was once involved in an X-ray polarimetry proposal in which SQUIDS were to be used in some capacity. Now I have an anchor point for their importance in general.
Would that data be obtainable from doing the quickly-repeated experiments written about, or would that experimentation come after that data? Sorry if my line of questioning is nonsensical. (I just looked up 'asymptotic freedom', I think I understand what it means. I'm glad I remembered what an 'asymptote' is from high-school maths.)
Rad Antonov
Could be, but above all it requires new data to guide the theorists. That doesn’t necessarily mean a bigger collider, not that there is any chance another one will be built in our lifetimes. It could come from working on problems that presently get swept under the rug. For example, for all its glory, the Standard Model only makes testable predictions under conditions of asymptotic freedom. QCD isn’t particularly good at predicting properties you might actually care about, like the mass of a proton.
PLEASE DON'T STORE THE CRYOGENICALLY FROZEN HEADS WITH THE BOOZE, THANK YOU. Much appreciated.
Armando Mistral
Well, the savagery of this takedown was absolutely up to Sabine's standards. Bravo. There were several points where I had to just pause in order to collect myself, freeze their heads indeed. The cheerleaders for this dreck will soon declare that the bubble is quantum, and not classical, and so making a profit is passe and irrelevant. The 80's called and wants its money back.
We have other bubbles in physics, however also those which may be solved easier than this quantum computing. The open problems dark matter and dark energy will turn out to be a test of Einstein's GRT. I have already offered a bet over € 50'000 that this test will be negative.
Interesting.
Rad Antonov
I am intrigued and hope it leads to a distinct prediction that an experimentalist can endeavor to measure. If you allow me to speculate, I am wondering if attempts to treat spacetime as emerging from the entanglement of quantum field theories may be related. In a sense, the where and when of the detector settings are the manifestation of the entanglement that give rise to notions of cause and effect at the macro level.
@Rad: Hi, what do you think of the work on Superdeterminism by Sabine et. al.?
Rad Antonov
The hardware for quantum computing is itself an offshoot of advancements in nano fabrication techniques to build bolometer arrays for mm wave astronomy that utilize SQUIDS as amplifiers. Keating had an observational cosmology guest on one of his podcasts who indicated that graduate students who knew how to operate a dilution refrigerator were in strong demand by the quantum computing crowd. The SQUIDs were in turn originally made to study superconductivity. Ion trap quantum computing seems to me is very much the offshoot of atomic physicists figuring out how to make BECs. Point being, if we keep supporting basic research, we will always have offshoots. Let the VCs and industry worry about commercializing them and take the hit when they fail. As for Superdeterminism, unless it can make a definitive prediction that is different from garden variety quantum mechanics, it will remain untestable, no matter the engineering breakthrough. Right now, it’s still just another interpretation. P.S. I’ve mentioned it before, but it’s worth mentioning again. The quantum computing course by Brilliant is really well done: https://brilliant.org/courses/quantum-computing/
Lots of conflicting thoughts going through my head on this one. So many times, engineering problems, which quantum computing is, just need tons of money thrown in their direction for breakthroughs to happen. But without a major breakthrough in a reasonable amount of time, the bubble will burst with consequences for all research with the word "quantum" attached to it. How many cycles of "a major breakthrough is expected in the next 10 years" will it take to burst the bubble? Then, I wonder what useful engineering will be developed even if quantum computing never really comes to fruition. For example, might there be advances in cryotechnology and super low noise entangled qubit systems that will allow robust tests of superdeterminism? I also wonder if another technology will come along that will match the expectations of quantum computing, but the engineering difficulties will be more easily solved. I don't have anything in mind, just spitballing here.
So, no need for post-quantum crypto and Bitcoin is staying. "Nature’s a bitch sometimes." Yeah!
As one who is involved in cybersecurity, I wish you had discussed the applicability of quantum computing to encryption/decryption. There is a lot of FUD being spread out there and it is even affecting me. LOL
I agree with your skepticism, but you may have to write another blog post when those who are financially dependent on the funders believing the hype start attacking. I used to work for a medium-large computer company and my neighbor, who is a well-read retired engineer, had asked me several times about whether we were going to do a quantum computer. My reply was, "No. We wait until a technology works and then if we could figure out how to build and service such a device and make a profit we would jump in." We had zero people on payroll looking at that. One area I would like to see an advance is molecular dynamics. A complex molecule just finds its minimum energy state all by itself; a supercomputer would take hundreds or thousands of hours to approximate it using today's best algorithms. Will quantum computers help? Maybe, but, as you say, you gotta build one.
The stochastic processes in my brain say that proton-boron fusion would be a better investment. That does not mean it would be good investment, just better (estimated expected value) based on the odds and the potential payback.
This seems like String Theory and the Multiverse kind of thing. I'm wondering how many bottles of beer and cider could go into those fridges.
