Sabine
patreon
Will AI kill Quantum Computing?
🕑 Added 2024-01-11 16:00:10 +0000 UTC[This is a transcript with links to references.]
It’s hard to keep track of all the things that will supposedly fix climate change but won’t. Such as hydrogen, veganism, or quantum computing. Wait what? Yes, marketing quantum computers as miracle cure for climate change has not fallen out of fashion despite my many jokes about it.
For example, in a recent article in The Hill, Sam Howell, a research associate at the Center for New American Security in Washington, DC writes “If countries are serious about climate change, they should get serious about quantum computing”. Right, if we put all our money into quantum computing there’ll be none left for oil, problem solved!
But quantum computing hype isn’t new, and this is supposed to be science news, so let me tell you what’s new. New is that Yann LeCun, the AI chief of Meta, has a surprisingly sane attitude to quantum computing. This deserves celebration and also investigation because he’s making a very good point.
According to a recent report by CNBC he said that the “practical relevance”of quantum computers is unclear, that many of the problems you can solve with quantum computing, you can solve way more efficiently with classical computers, and that quantum computing has “got such a long time horizon that it’s irrelevant” to what they’re doing at Meta.
I’ve found this to be interesting, because it draws attention to the competition between artificial intelligence and quantum computing that’s becoming increasingly noticeable and that might just be the final nail in the coffin that is commercial interest in quantum computing. I thought it’d be worth looking at this a bit closer and explain what I believe he meant.
You see, quantum computing is extremely exciting from the research perspective because it’s completely uncharted territory. We’ve never used such big quantum things before. But getting commercially relevant results out of quantum computers is another thing entirely. Not only do you need the computers to be big enough and still stable enough to maintain the fragile quantum states. You then also need to demonstrate that they can do something better than a conventional computer.
And this requires two things. First, you’ll need an algorithm that you can run on the quantum computer that has a computational advantage over a conventional computer for some sort of interesting problem. And second, you’ll have to show that in reality it’ll actually come out being faster.
The best understood quantum algorithms are those for certain problems quantum chemistry – that is, basically material science – some problems in logistics and finance, and some sorts of code-cracking. It’s a rather limited class of problems. Of course, the companies who want answers to these problems already have pretty good algorithms on conventional computers. And these are now further improving with artificial intelligence, which is why the current AI boom is bad news for quantum computing.
But how is this even relevant if quantum computers are supposedly unbeatably fast? It’s because of what gives them their advantage.
Quantum computers work with quantum bits, qubits for short, and derive their advantage from the many ways you can correlate or entangle those qubits. But it’s not that operating the qubits themselves is fast. Indeed, operating qubits is usually slower than doing operations on a conventional computer.
Keep in mind that modern computers operate in the GigaHertz range or above, that’s more than a billion operations a second. How fast you can operate qubits depends on the type of qubits. But for superconducting circuits, it’s maybe 10 million per second optimistically, and ion traps are much slower still. And that doesn’t take into account the need for error correction which’ll make them even slower.
The reason people are excited about quantum computers is not because they’re fast per operation, but because they need fewer operations to get the same result. So you see if you have a small calculation on a conventional computer, one that requires few operations, it’ll beat the quantum computer because the conventional computer is faster per operation. If you increase the size of the problem, then both the conventional computer and the quantum computer will need more operations and so more time. But for the quantum algorithm, the number of operations increases less with the size of the problem. So the idea is that regardless of how slow the qubits are, eventually the quantum computer will outperform the conventional computer because it needs fewer operations.
That’s the theory. In practice the question is where this cross-over happens where the quantum computer starts to outperform the conventional one. That’s a moving target, because each time conventional computers become faster or someone comes up with a better algorithm, the point where quantum computers finally win moves further away, to a higher number of qubits.
And this is why artificial intelligence is becoming a problem for quantum computing. Because artificial intelligence lets you get more out of conventional computers. And the number of qubits at which quantum computers will bring an advantage might eventually become so large that it becomes impractical or just prohibitively expensive to even use them.
Though it’s a shame that Meta isn’t more into quantum computing, I would really like to have a button that’s thumbs up and down at the same time.
Comments
Rad Antonov
On a really good day I could get you down to 50 mK
D Brown
Yes, and No.
Rad, As long as they are not chilled to less than one degree Kelvin, sure: https://en.wikipedia.org/wiki/Superconducting_quantum_computing#:~:text=Superconducting%20quantum%20computing%20devices%20are,frequency%20synthesizers%20and%20spectrum%20analyzers. Excerpt: "Superconducting quantum computing devices are typically designed in the radio-frequency spectrum, cooled in dilution refrigerators below 15 mK (millikelvins) and addressed with conventional electronic instruments, e.g. frequency synthesizers and spectrum analyzers."
