Quantum Computers Should (Also) Be Useful
The quantum computing field has been around for 25+ years and we still don’t have a single use case where it makes sense to use a QC vs the alternatives. Yes, QCs are amazing testbeds for fundamental science. But they should also be useful! And right now, they … aren’t. At least not in the clearcut way we all want them to be.
When I first got into QC back in 1999, it was mostly because of an interest in fundamental physics. Are there actually many worlds, is the collapse of the wavefunction a thing or not, things like that. That journey was amazingly cool. I learned a lot and met many unique and awesome people.
I’m still fascinated by the science of the quantum world. But I’m also interested in whether the economic promise of QCs can be delivered. The field has been around for 25+ years and we still don’t have a single use case where it makes sense to use a QC vs the alternatives. Yes, QCs are amazing testbeds for fundamental science. But they should also be useful! And right now, they … aren’t. At least not in the clearcut way we all want them to be.
A few months ago I saw a preprint by the D-Wave folks that contained a claim of quantum supremacy for a quantum simulation problem. (You can see a definition for quantum supremacy here). They simulate what happens when you drive a bunch of magnets through a quantum phase transition between paramagnetic and spin glass phases by quenching a transverse magnetic field. If this sounds exotic, it is, but it’s related to many other less esoteric problems, including in areas related to Sanctuary’s work in embodied AI.
Like all current quantum supremacy claims, (a) it’s unclear whether the claim will hold up (someone could come up with a better classical approach that erases the claim), and (b) even if it does hold up, it’s not clear how to use the underlying capability to build applications that make sense to run on a QC.
There’s nothing to be done about (a). There are a lot of smart people working to build better QCs and better classical approaches to erase quantum supremacy claims. I think eventually the QC folks will win. I think they already have. But we’ll see.
Independent of (a), can progress can be made on (b)? Assuming a quantum supremacy claim holds up, regardless of what QC is being used or what the problem it’s solving is, is there a way to turn that into something useful? This is surprisingly hard. Our experience with conventional computers is that two machines can be directly compared by asking which is faster. But that’s not how it works in this case. Even if a QC is much faster on some exotic problem, turning that speed into solving a useful problem seems to always erase any benefit you started with.
Here’s an analogy. Let’s say we see trees blowing in the wind. Their leaves are moving all over the place, sometimes they fall off, it’s all very complicated. Now imagine that what you’re watching is a computation (it actually is — nature is constantly computing the solutions to the laws of physics). It’s super fast and accurate — the trees do an exceptional job of computing themselves.
But now ask how you’d use that computation to do a different computation, say multiplying two matrices together. How would you use the trees’ inherent computation to multiply two matrices? I have no idea. Presumably it’s possible. But this is kind of like the issue with quantum supremacy. A QC can be very good at doing something (like being itself, solving the equations of the laws of physics that describe it, like the tree), but then using that to solve a different problem is not easy.
The world is constantly computing solutions to the laws of physics, and it’s really good at it!
I have an idea for how to start picking at the threads of this problem. It’s a connection between quantum supremacy and a peculiar type of difficulty in building game-playing AIs.