The qubit systems we have today are a tremendous scientific achievement, but they take us no closer to having a quantum computer that can solve a problem that anybody cares about. It is akin to trying to make today’s best smartphones using vacuum tubes from the early 1900s. You can put 100 tubes together and establish the principle that if you could somehow get 10 billion of them to work together in a coherent, seamless manner, you could achieve all kinds of miracles. What, however, is missing is the breakthrough of integrated circuits and CPUs leading to smartphones—it took 60 years of very difficult engineering to go from the invention of transistors to the smartphone with no new physics involved in the process.
There are in fact ideas, and I played some role in developing the theories for these ideas, for bypassing quantum error correction by using far-more-stable qubits, in an approach called topological quantum computing. Microsoft is working on this approach. But it turns out that developing topological quantum-computing hardware is also a huge challenge. It is unclear whether extensive quantum error correction or topological quantum computing (or something else, like a hybrid between the two) will be the eventual winner.
Physicists are smart as we all know (disclosure: I am a physicist), and some physicists are also very good at coming up with substantive-sounding acronyms that stick. The great difficulty in getting rid of decoherence has led to the impressive acronym NISQ for “noisy intermediate scale quantum” computer—for the idea that small collections of noisy physical qubits could do something useful and better than a classical computer can. I am not sure what this object is: How noisy? How many qubits? Why is this a computer? What worthy problems can such a NISQ machine solve?
A recent laboratory experiment at Google has observed some predicted aspects of quantum dynamics (dubbed “time crystals”) using 20 noisy superconducting qubits. The experiment was an impressive showcase of electronic control techniques, but it showed no computing advantage over conventional computers, which can readily simulate time crystals with a similar number of virtual qubits. It also did not reveal anything about the fundamental physics of time crystals. Other NISQ triumphs are recent experiments simulating random quantum circuits, again a highly specialized task of no commercial value whatsoever.
Using NISQ is surely an excellent new fundamental research idea—it could help physics research in fundamental areas such as quantum dynamics. But despite a constant drumbeat of NISQ hype coming from various quantum computing startups, the commercialization potential is far from clear. I have seen vague claims about how NISQ could be used for fast optimization or even for AI training. I am no expert in optimization or AI, but I have asked the experts, and they are equally mystified. I have asked researchers involved in various startups how NISQ would optimize any hard task involving real-world applications, and I interpret their convoluted answers as basically saying that since we do not quite understand how classical machine learning and AI really work, it is possible that NISQ could do this even faster. Maybe, but this is hoping for the best, not technology.
