Horses for Courses: Where Quantum Computing Is, and Isn’t, the Answer
Take, for example, the university coursework timetabling problem. A university timetable without conflicts for students, staff or venues is easy to verify but hard to discover among countless possibilities. A classical computer using a brute-force trial-and error approach would need to check half of all options on average, while Grover’s algorithm could find the right one in a number of steps roughly equal to the square root of the total number of timetable options. The difference between these two methods isn’t that significant when the number of possible timetables is small, but as this number grows the quantum advantage becomes increasingly more pronounced.
The point at which quantum computers outperform classical computers isn’t determined based on whether a quantum computer can solve a particular problem, but whether it can do so with fewer resources than other options. This fact may seem obvious, but its neglect has led some to inaccurately claim that quantum computing has achieved supremacy over classical computing.
With these points in mind, one can start thinking critically about the kinds of problems that quantum computers may be able to help solve.
Drug discovery and materials design are among the most promising applications. The properties of molecules and materials arise from complex many-body quantum interactions, which become exponentially harder to simulate as the system grows. Classical computers struggle with this complexity, but quantum computers—operating on quantum states themselves—are naturally suited to the task. This could speed up the process of discovering new medicines or materials with new and useful properties, with a cascade of impacts that would be felt across industries.
That said, artificial intelligence implemented on classical computers may be able to, at least partially, solve some of these same problems. AlphaFold, an AI developed by DeepMind, is already being used to predict protein structures—a key use case for quantum simulation. Still, physics-driven quantum simulations will probably offer more robust and generalizable solutions than machine learning models which are limited by their training dataset.
Encryption-breaking is another widely discussed application of quantum computing, but this needs to be clearly grounded in an understanding that it will only be relevant for historical communications rather than ongoing vulnerabilities. Cryptographic standards are already transitioning toward quantum-resistant algorithms—schemes for which there are currently no known quantum algorithm with a compelling advantage over classical methods.
While these new standards can’t protect past communications—hence the strategy of harvesting now to decrypt later, almost certainly being used by intelligence agencies around the world—the window of vulnerability is closing. By the time quantum computers capable of breaking today’s encryption are operational, we’ll likely have returned to a new cryptographic equilibrium that will severely undermine the usefulness of quantum computers in intelligence gathering.
Optimization problems—such as designing more efficient supply chains or energy grids to minimize wastage—are another frequently-cited application of quantum computing. Computing company IBM estimates a 1 percent improvement in last-mile delivery efficiency could yield global savings of around US$400 million (A$600 million) annually. As optimization problems occur widely across industry, estimates on the potential impact are enormous: Boston Consulting Group offers an estimate of between US$100-220 billion (A$150-335 billion) in annual value creation.
These figures should be taken with caution. The quantum algorithms most likely to deliver practical value in optimization are heuristic-based, relying on rule-of-thumb strategies that simplify the search for solutions by trading off precision for computational efficiency—a common approach in classical optimization as well. However, this inexactness makes it difficult to directly compare quantum and classical methods. As the industry-led Quantum Optimization Working Group notes, ‘being practically relevant requires taking into account many details and complications that most published results in the related quantum optimization literature are lacking’. Until quantum algorithms demonstrate a clear and consistent practical advantage, industry and policymakers should not expect quantum computers to displace classical computers in solving real-world optimization challenges.
Computer scientist Scott Aaronson aptly described quantum computing as ‘one of the most mis-popularized and mis-explained topics in the history of science’. This observation isn’t to dismiss the technology. Quantum computers won’t solve every problem, nor will they revolutionize every industry. But in the domains where they do hold real promise, such as materials science and drug discovery, their impact could be significant. In the meantime, filtering out the noise will be key to ensuring that, when these machines scale to a useful size, we’re ready to make the most of them.
Stephan Robin is a data scientist at ASPI. This article is published courtesy of the Australian Strategic Policy Institute (ASPI).
