Siddharth Pai: Can AI beat quantum computing at its own game?
Artificial intelligence (AI) has shown it can imperfectly but acceptably achieve much that was envisioned for quantum computers. And AI is moving fast. So quantum computing risks becoming a beautiful idea outpaced by a merely competence but deployable alternative.
For decades, quantum computing has been described as the 21st century’s technological lodestar—with its unfathomable computational power poised to solve problems beyond the ken of classical machines. Quantum computers promise to crack cryptographic codes, simulate the quantum dynamics of molecules in material science, aid drug discovery and more. Yet, as the quantum race drags on, an unexpected challenger has emerged, not to dethrone but outpace it in precisely those domains where it was expected to shine the brightest: AI.
To grasp the possibility of this disruption, begin with what quantum computing is. Unlike classical computers that encode information in binary bits—0s or 1s—quantum computers use quantum bits, or qubits, which can exist in a superposition of states.
Also Read: Will AI ever grasp quantum mechanics? Don’t bet on it
Through entanglement and quantum interference, quantum computers can process a vast space of possibilities in parallel. This lets them model quantum systems naturally, making them ideal for simulating molecules, designing new materials and solving certain optimization problems. Among its most touted applications is its potential to transform material science.
Advances with high-temperature superconductors, catalytic surfaces or novel semiconductors often require modelling the interactions of strongly correlated electrons—systems where the behaviour of one particle is tightly linked to that of many others. Classical algorithms falter in such simulations because the complexity of the quantum state space rises exponentially with system size. A full-fledged quantum computer would handle all this with ease.
But the practical realization of quantum computing remains vexed. Qubits, whether superconducting loops, trapped ions or topological states, are exquisitely fragile. They ‘decohere’ (lose their quantum state) within microseconds and must be kept at temperatures colder than deep space. Error correction remains an uphill battle. Most of today’s quantum machines can manage only a few hundred noisy qubits, far short of the millions needed for fault-tolerant computing.
Also Read: Is Google’s Willow really a ‘wow’ moment for quantum computers?
In the meantime, artificial intelligence (AI), particularly deep learning, has made remarkable incursions into the same spaces. A turning point came in 2017 with a paper in Science by Giuseppe Carleo and Matthias Troyer (bit.ly/43UHAXx). Startling scientists, they found a neural network-based variational method to approximate the wave-function of quantum systems. This approach employed restricted Boltzmann machines to represent complex correlations among quantum particles, modelling the ground states of certain spin systems that had been hard to simulate classically.
That paper didn’t just introduce a new tool; it signalled a paradigm shift. Researchers used it for deep convolutional and autoregressive networks, transformer architectures and even diffusion models to simulate quantum many-body systems. These neural networks run on classical hardware and do not require the brittle infrastructure of quantum machines.
It’s not merely a question of catching up. AI is beginning to demonstrate capabilities in material discovery and quantum simulation that, while not perfectly accurate at the quantum level, are good enough.
Generative models have proposed new crystalline structures with desirable thermal or electronic properties, while graph neural networks have predicted materials’ phase behaviour without recourse to first-principle calculations. Most strikingly, AI models have begun to assist in inferring effective Hamiltonians—mathematical descriptions of physical systems—from experimental data, a tough task even for top-level experts.
This acceleration has not gone unnoticed by major research labs. Google’s DeepMind, for instance, has begun integrating machine learning tools directly into quantum chemistry workflows. Startups in the quantum space often include AI-based pre-processing or error mitigation in their pipelines.
Also Read: Underdelivery: AI gadgets have been a let-down but needn’t be
A complementary field is fast becoming a competing one. AI will not make quantum computing irrelevant in the absolute sense, as there will always be quantum phenomena that only quantum devices can fully capture, but AI may take the lead in many practical problems before quantum hardware matures. If machine learning models can deliver 90% of the performance at 5% of the cost and infrastructure, industrial users may not wait for perfection.
Moreover, there’s a subtler factor at play: a shift in intellectual capital. The more investment AI-based methods attract, the more resources will flow into neural modelling over quantum error correction. By the time quantum machines mature, many of the use-cases originally envisioned for them may have been absorbed by AI tools that ironically use quantum data or theory. Quantum computing risks becoming a beautiful idea outpaced by a merely competent but deployable alternative.
There is an irony here that would not be lost on Schrödinger or Feynman: that the classical world, once deemed too simplistic in the face of quantum reality, might be reasserting itself through the statistical abstractions of machine learning. We set out to build a machine that thinks like nature. Instead, we taught our machines to imitate nature well enough to move forward without grasping it fully.
Quantum computing may still prove indispensable. But it will have to justify its place in a world where its promise is being appropriated by its upstart cousin AI.
The author is co-founder of Siana Capital, a venture fund manager.
topics
