Jerry Chow on Quantum Computing, Encryption, AI and Error Correction

Matthew Berman interviews Jerry Chow, IBM Fellow and director of quantum systems at IBM Quantum, about quantum computing, encryption, hardware, AI and the path to error correction. Chow explains where quantum systems could complement classical computers and where the field still faces research challenges.

Asked for a simple definition, Chow says ordinary computers encode information in bits and process it using familiar binary rules. Quantum computers use quantum bits, or qubits, whose information follows the mathematics of quantum mechanics. This changes the representation and processing of information.

That different mathematical structure creates an opportunity to solve certain problems more efficiently, potentially with exponential advantages in particular cases. Chow emphasizes specific problem classes: a quantum computer is not automatically faster at every task.

Berman asks about dramatic comparisons with classical supercomputers. Chow says the goal is to find problems with the right structure for a quantum algorithm. One famous theoretical example is Shor’s algorithm for factoring large numbers, a problem relevant to some widely used public-key cryptography.

Another potential application is modeling molecules and materials. Electron interactions are quantum mechanical, and simulating complex systems can become extremely expensive on classical machines. Chow explains that controllable qubits could represent aspects of those interactions, with potential implications for materials and drug discovery.

Berman asks whether quantum computing would mean the end of encryption or cryptocurrency. Chow says the transition need not be sudden: quantum-resistant cryptographic methods exist, including lattice-based approaches. He points to protocols supported by NIST and says the major practical challenge is moving industries and data to new standards.

Banks and telecommunications companies are beginning to consider that migration, Chow says. Replacing established cryptographic infrastructure involves cost and coordination. His point is that organizations can plan for a future capability instead of treating every form of encryption as already broken.

Berman asks whether a lone developer could build a cryptographically powerful quantum computer unexpectedly. Chow describes the systems being developed as capital-intensive infrastructure with hardware, networking and classical controls at data-center scale. IBM publishes a roadmap and tracks progress over multiple years.

Berman asks about the iconic chandelier-like machine and its cooling requirements. Chow explains that IBM uses superconducting qubits made with materials such as aluminum, niobium and titanium nitride. Carefully designed circuits made from these materials can behave quantum mechanically and be combined on a processor.

For this superconducting approach, the chip must be cooled to roughly 10–15 millikelvin. Chow compares that with liquid nitrogen at about 77 kelvin and the cold of space. The temperatures are extreme, but dilution refrigerators using helium isotopes can reach them.

The familiar golden structure in many photographs is the inside of a dilution refrigerator, with the processor at the coldest end. During operation it is enclosed by heat shields. Its successive stages cool wiring and components as well as the chip.

Other components condition control signals and amplify signals from the processor. Classical electronics then digitize the measurements. The machinery is more than the refrigerated chip: it includes the controls and conventional computing infrastructure needed to operate it.

Berman asks why a quantum computer produces classical output. Chow explains that people program and observe the system through classical inputs and measurements; the computation inside follows quantum rules. Classical and quantum computers can pass work back and forth, including preprocessing and analysis of results.

On software, Chow recalls IBM putting a five-qubit machine on the cloud in 2016. The team needed a way for users to program it, so it exposed a graphical representation of quantum circuits rather than only the laboratory controls physicists had used.

A quantum circuit combines gate operations on qubits with measurements that return classical results. Chow compares this low-level representation to an assembly language. As with classical computing, useful applications require additional layers of abstraction above individual operations.

IBM introduced Qiskit, its quantum information software development kit, in 2017. Chow says its APIs help users map problems, optimize circuits and execute them on hardware. The project has also supported a community learning how to program quantum systems.

Developers should not need to understand every physical detail of a qubit, Chow says. IBM is building higher-level functions for application areas such as chemistry and optimization, so a user can work at a more meaningful level while the software maps the problem onto circuits.

Berman asks whether quantum computers will replace ordinary computers. Chow says no: some problems map better to quantum circuits, some to tensor operations on GPU clusters and others to conventional CPUs. IBM calls the combination of these representations “bits, neurons and qubits.”

The full set of useful quantum problems is still unknown. Researchers have mathematical evidence of speedups for some structured problems, subject to sufficient hardware and circuit resources. Chow sees a larger unexplored space of algorithms, including regimes that are difficult to simulate classically.

On AI, Chow separates two directions: AI for quantum and quantum for AI. The first includes a Qiskit code assistant that helps developers write software, as well as using AI to optimize how a circuit maps onto hardware.

Chow describes IBM’s AI transpiler service, which uses reinforcement-learning techniques to optimize circuit mapping. He says AI may help break down problems and improve quantum workflows, with further research needed to identify the most useful applications.

Quantum machine learning is more nuanced. Chow mentions research on using quantum circuits for classification and learning, while noting that the nature of any practical advantage remains under investigation. The challenge is finding datasets and problem structures that map well to quantum methods.

Looking further ahead, Chow describes quantum-centric supercomputing that places CPUs, GPUs and quantum processing units together. He expects researchers to explore which tasks belong on each type of hardware, potentially including AI workloads. Building integrated systems is part of discovering what works.

Berman asks about Google’s Willow device. Chow says its work studies quantum error correction using a surface code and evaluates its performance as a logical memory. To explain why this matters, he first describes how today’s qubits lose information through interactions with their environment.

Superconducting qubits are physical structures that can be affected by fabrication, packaging and control wiring. A computation requires gates and interactions between qubits; if errors destroy the stored information during the sequence, the result cannot be trusted. Every qubit platform needs a strategy for dealing with errors.

Chow distinguishes near-term error suppression and mitigation from error correction. Mitigation can characterize noise and use multiple circuit runs to estimate cleaner results. Error correction instead encodes logical information across multiple physical qubits and monitors errors during computation.

A classical repetition code can copy a bit and use a majority vote, but quantum information cannot simply be copied in the same way. Quantum codes therefore use a more complex arrangement that detects error information without directly destroying the encoded state.

In a surface code, a lattice contains data qubits that carry information and additional qubits measured to reveal error syndromes. Chow compares interpreting these measurements to a large game of Minesweeper: clues help locate where corrections are needed.

The surface code is widely studied, but Chow says it can require a large number of physical qubits per logical qubit. IBM has also investigated low-density parity-check codes. He presents their reduced overhead as a research direction toward more scalable error correction, not as a completed large-scale machine.

For people who want to learn more, Chow points to IBM Quantum’s learning materials, including introductory circuits and error-correction concepts. He also recommends explanatory videos by John Watrous. The resources provide a route from first experiments to deeper theory.

Berman asks about NVIDIA CEO Jensen Huang’s public comments on quantum timelines. Chow avoids attributing a precise meaning to Huang’s remarks and describes IBM’s own multiyear roadmap. He emphasizes that quantum systems are intended to augment CPUs and GPUs rather than replace them.

In the interview, Chow cites IBM’s goals for a fault-tolerant system called Starling in 2029 and a larger Blue Jay in 2033. These are roadmap targets presented in the conversation, not delivered capabilities or guarantees of future performance.

Berman asks how Chow entered the field. Chow grew up in a household that valued science and mathematics; his father was a nuclear physicist. In college he sought experimental work on nanoscale systems, approached a Harvard professor and began moving carbon nanotubes in the lab.

That early experience studying nanoscale electronic transport led Chow deeper into quantum physics and eventually to superconducting qubits during his PhD. He points listeners to his professional updates and IBM Quantum’s site for further information.