“Talking quantum circuits”
Interpretable and scalable quantum natural language processing
The central question that pre-occupies our team has been:
“How can quantum structures and quantum computers contribute to the effectiveness of AI?”
In previous work we have made notable advances in answering this question, and this article is based on our most recent work in the new papers [arXiv:2406.17583, arXiv:2408.06061], and most notably the experiment in [arXiv:2409.08777].
This article is one of a series that we will be publishing alongside further advances – advances that are accelerated by access to the most powerful quantum computers available.
Large language Models (LLMs) such as ChatGPT are having an impact on society across many walks of life. However, as users have become more familiar with this new technology, they have also become increasingly aware of deep-seated and systemic problems that come with AI systems built around LLM’s.
The primary problem with LLMs is that nobody knows how they work - as inscrutable “black boxes” they aren’t “interpretable”, meaning we can’t reliably or efficiently control or predict their behavior. This is unacceptable in many situations. In addition, Modern LLMs are incredibly expensive to build and run, costing serious – and potentially unsustainable –amounts of power to train and use. This is why more and more organizations, governments, and regulators are insisting on solutions.
But how can we find these solutions, when we don’t fully understand what we are dealing with now?1
At Quantinuum, we have been working on natural language processing (NLP) using quantum computers for some time now. We are excited to have recently carried out experiments [arXiv: 2409.08777] which demonstrate not only how it is possible to train a model for a quantum computer in a scalable manner, but also how to do this in a way that is interpretable for us. Moreover, we have promising theoretical indications of the usefulness of quantum computers for interpretable NLP [arXiv:2408.06061].
In order to better understand why this could be the case, one needs to understand the ways in which meanings compose together throughout a story or narrative. Our work towards capturing them in a new model of language, which we call DisCoCirc, is reported on extensively in this previous blog post from 2023.
In new work referred to in this article, we embrace “compositional interpretability” as proposed in [arXiv:2406.17583] as a solution to the problems that plague current AI. In brief, compositional interpretability boils down to being able to assign a human friendly meaning, such as natural language, to the components of a model, and then being able to understand how they fit together2.
A problem currently inherent to quantum machine learning is that of being able to train at scale. We avoid this by making use of “compositional generalization”. This means we train small, on classical computers, and then at test time evaluate much larger examples on a quantum computer. There now exist quantum computers which are impossible to simulate classically. To train models for such computers, it seems that compositional generalization currently provides the only credible path.
1. Text as circuits
DisCoCirc is a circuit-based model for natural language that turns arbitrary text into “text circuits” [arXiv:1904.03478, arXiv:2301.10595, arXiv:2311.17892]. When we say that arbitrary text becomes ‘text-circuits’ we are converting the lines of text, which live in one dimension, into text-circuits which live in two-dimensions. These dimensions are the entities of the text versus the events in time.
To see how that works, consider the following story. In the beginning there is Alex and Beau. Alex meets Beau. Later, Chris shows up, and Beau marries Chris. Alex then kicks Beau.
The content of this story can be represented as the following circuit:
2. From text circuits to quantum circuits
Such a text circuit represents how the ‘actors’ in it interact with each other, and how their states evolve by doing so. Initially, we know nothing about Alex and Beau. Once Alex meets Beau, we know something about Alex and Beau’s interaction, then Beau marries Chris, and then Alex kicks Beau, so we know quite a bit more about all three, and in particular, how they relate to each other.
Let’s now take those circuits to be quantum circuits.
In the last section we will elaborate more why this could be a very good choice. For now it’s ok to understand that we simply follow the current paradigm of using vectors for meanings, in exactly the same way that this works in LLMs. Moreover, if we then also want to faithfully represent the compositional structure in language3, we can rely on theorem 5.49 from our book Picturing Quantum Processes, which informally can be stated as follows:
If the manner in which meanings of words (represented by vectors) compose obeys linguistic structure, then those vectors compose in exactly the same way as quantum systems compose.4
In short, a quantum implementation enables us to embrace compositional interpretability, as defined in our recent paper [arXiv:2406.17583].
3. Text circuits on our quantum computer
So, what have we done? And what does it mean?
We implemented a “question-answering” experiment on our Quantinuum quantum computers, for text circuits as described above. We know from our new paper [arXiv:2408.06061] that this is very hard to do on a classical computer due to the fact that as the size of the texts get bigger they very quickly become unrealistic to even try to do this on a classical computer, however powerful it might be. This is worth emphasizing. The experiment we have completed would scale exponentially using classical computers – to the point where the approach becomes intractable.
The experiment consisted of teaching (or training) the quantum computer to answer a question about a story, where both the story and question are presented as text-circuits. To test our model, we created longer stories in the same style as those used in training and questioned these. In our experiment, our stories were about people moving around, and we questioned the quantum computer about who was moving in the same direction at the end of the stories. A harder alternative one could imagine, would be having a murder mystery story and then asking the computer who was the murderer.
And remember - the training in our experiment constitutes the assigning of quantum states and gates to words that occur in the text.
4. Compositional generalization
The major reason for our excitement is that the training of our circuits enjoys compositional generalization. That is, we can do the training on small-scale ordinary computers, and do the testing, or asking the important questions, on quantum computers that can operate in ways not possible classically. Figure 4 shows how, despite only being trained on stories with up to 8 actors, the test accuracy remains high, even for much longer stories involving up to 30 actors.
Training large circuits directly in quantum machine learning, leads to difficulties which in many cases undo the potential advantage. Critically - compositional generalization allows us to bypass these issues.
5. Real-world comparison: ChatGPT
We can compare the results of our experiment on a quantum computer, to the success of a classical LLM ChatGPT (GPT-4) when asked the same questions.
What we are considering here is a story about a collection of characters that walk in a number of different directions, and sometimes follow each other. These are just some initial test examples, but it does show that this kind of reasoning is not particularly easy for LLMs.
The input to ChatGPT was:
What we got from ChatGPT:
Can you see where ChatGPT went wrong?
ChatGPT’s score (in terms of accuracy) oscillated around 50% (equivalent to random guessing). Our text circuits consistently outperformed ChatGPT on these tasks. Future work in this area would involve looking at prompt engineering – for example how the phrasing of the instructions can affect the output, and therefore the overall score.
Of course, we note that ChatGPT and other LLM’s will issue new versions that may or may not be marginally better with ‘question-answering’ tasks, and we also note that our own work may become far more effective as quantum computers rapidly become more powerful.
6. What’s next?
We have now turned our attention to work that will show that using vectors to represent meaning and requiring compositional interpretability for natural language takes us mathematically natively into the quantum formalism. This does not mean that there doesn't exist an efficient classical method for solving specific tasks, and it may be hard to prove traditional hardness results whenever there is some machine learning involved. This could be something we might have to come to terms with, just as in classical machine learning.
At Quantinuum we possess the most powerful quantum computers currently available. Our recently published roadmap is going to deliver more computationally powerful quantum computers in the short and medium term, as we extend our lead and push towards universal, fault tolerant quantum computers by the end of the decade. We expect to show even better (and larger scale) results when implementing our work on those machines. In short, we foresee a period of rapid innovation as powerful quantum computers that cannot be classically simulated become more readily available. This will likely be disruptive, as more and more use cases, including ones that we might not be currently thinking about, come into play.
Interestingly and intriguingly, we are also pioneering the use of powerful quantum computers in a hybrid system that has been described as a ‘quantum supercomputer’ where quantum computers, HPC and AI work together in an integrated fashion and look forward to using these systems to advance our work in language processing that can help solve the problem with LLM’s that we highlighted at the start of this article.
1 And where do we go next, when we don’t even understand what we are dealing with now? On previous occasions in the history of science and technology, when efficient models without a clear interpretation have been developed, such as the Babylonian lunar theory or Ptolemy’s model of epicycles, these initially highly successful technologies vanished, making way for something else.
2 Note that our conception of compositionality is more general than the usual one adopted in linguistics, which is due to Frege. A discussion can be found in [arXiv: 2110.05327].
3 For example, using pregroups here as linguistic structure, which are the cups and caps of PQP.
4 That is, using the tensor product of the corresponding vector spaces.
About Quantinuum
Quantinuum, the world’s largest integrated quantum company, pioneers powerful quantum computers and advanced software solutions. Quantinuum’s technology drives breakthroughs in materials discovery, cybersecurity, and next-gen quantum AI. With over 500 employees, including 370+ scientists and engineers, Quantinuum leads the quantum computing revolution across continents.
As organizations assess the impact of quantum computing on cryptography, many focus on algorithm migration and timelines. But preparing for PQC requires a broader view—one that includes not just new algorithms, but also the quality of the inputs that support them, including randomness.
That’s why Quantinuum joined with partners Thales, Keyfactor, and IBM Consulting to form the QSafe 360 Alliance, a collaboration focused on helping organizations build crypto-agile security architectures that are ready for the quantum era. Together, we’ve released a whitepaper—Digital Trust & Cybersecurity After Quantum Computing—to offer practical guidance on post-quantum readiness, from discovery and planning to deployment.
Lessons from Past Vulnerabilities
The history of cryptography offers clear examples of what happens when randomness fail, and how long those issues can go unnoticed. The Polynonce attack, first disclosed in 2023, exploited weak randomness in Bitcoin transaction signatures and enabled the theft of at least $25 million across 773 wallets. The vulnerability persisted undetected for nine years. The Randstorm disclosure, published in 2022, revealed that biased key generation in widely used Bitcoin wallet libraries exposed millions of wallets—across a window of more than a decade (2011–2022). In both cases, cryptographic algorithms functioned as designed; it was the randomness beneath them that silently failed, leaving companies vulnerable for many years
Post-Quantum Cryptography Inherits These Risks
Post-quantum cryptography (PQC) algorithms are being designed to resist attacks from quantum computers. But they still depend on random values to generate key material. That means any implementation of PQC inherits the same reliance on randomness—but without a way to prove its quality, that layer remains a potential vulnerability.
As security teams run cryptographic inventories, develop crypto-agility plans, or build software bill-of-materials (SBOMs) for PQC migration, it’s important to include randomness in that scope. No matter how strong the algorithm, poor randomness can undermine its security from the start.
A New Approach: Proven Randomness
Quantum Origin takes a fundamentally different approach to randomness quality to deliver proven randomness which improves key generation, algorithms, and the entire security stack. It leverages strong seeded randomness extractors—mathematical algorithms that transform even weak local entropy into provably secure output. These extractors are uniquely powered by a Quantum Seed, which is generated once by Quantinuum's quantum computers using quantum processes verified through Bell tests.
This one-time quantum generation enables Quantum Origin as a software-only solution designed for maximum flexibility. It works with existing infrastructure—on cloud systems, on-premises environments, air-gapped networks, and embedded platforms—without requiring special hardware or a network connection. It's also validated to NIST SP 800-90B standards (Entropy Source Validation #E214). This approach strengthens today’s deployments of AES, RSA, ECC, and other algorithms, and lays a secure foundation for implementing the NIST PQC algorithms.
The QSafe 360 Alliance
The QSafe 360 Alliance whitepaper outlines the path to post-quantum readiness, emphasizing crypto-agility as a guiding principle: the ability to adapt cryptographic systems without major disruption, from randomness to key generation to algorithmic strength.
For security architects, CISOs, and cryptographic engineering teams building their post-quantum transition strategies, randomness is not a peripheral concern. It is a starting point.
The QSafe 360 Alliance whitepaper offers valuable guidance on structuring a comprehensive PQC journey. As you explore that framework, consider how proven randomness—available today—will help strengthen your security posture from the ground up.
Our quantum algorithms team has been hard at work exploring solutions to continually optimize our system’s performance. Recently, they’ve invented a novel technique, called the Quantum Paldus Transform (QPT), that can offer significant resource savings in future applications.
The transform takes complex representations and makes them simple, by transforming into a different “basis”. This is like looking at a cube from one angle, then rotating it and seeing just a square, instead. Transformations like this save resources because the more complex your problem looks, the more expensive it is to represent and manipulate on qubits.
You’ve changed
While it might sound like magic, transforms are a commonly used tool in science and engineering. Transforms simplify problems by reshaping them into something that is easier to deal with, or that provides a new perspective on the situation. For example, sound engineers use Fourier transforms every day to look at complex musical pieces in terms of their frequency components. Electrical engineers use Laplace transforms; people who work in image processing use the Abel transform; physicists use the Legendre transform, and so on.
In a new paper outlining the necessary tools to implement the QPT, Dr. Nathan Fitzpatrick and Mr. Jędrzej Burkat explain how the QPT will be widely applicable in quantum computing simulations, spanning areas like molecular chemistry, materials science, and semiconductor physics. The paper also describes how the algorithm can lead to significant resource savings by offering quantum programmers a more efficient way of representing problems on qubits.
Symmetry is key
The efficiency of the QPT stems from its use of one of the most profound findings in the field of physics: that symmetries drive the properties of a system.
While the average person can “appreciate” symmetry, for example in design or aesthetics, physicists understand symmetry as a much more profound element present in the fabric of reality. Symmetries are like the universe’s DNA; they lead to conservation laws, which are the most immutable truths we know.
Back in the 1920’s, when women were largely prohibited from practicing physics, one of the great mathematicians of the century, Emmy Noether, turned her attention to the field when she was tasked with helping Einstein with his work. In her attempt to solve a problem Einstein had encountered, Dr. Noether realized that all the most powerful and fundamental laws of physics, such as “energy can neither be created nor destroyed” are in fact the consequence of a deep simplicity – symmetry – hiding behind the curtains of reality. Dr. Noether’s theorem would have a profound effect on the trajectory of physics.
In addition to the many direct consequences of Noether’s theorem is a longstanding tradition amongst physicists to treat symmetry thoughtfully. Because of its role in the fabric of our universe, carefully considering the symmetries of a system often leads to invaluable insights.
Einstein, Pauli and Noether walk into a bar...
Many of the systems we are interested in simulating with quantum computers are, at their heart, systems of electrons. Whether we are looking at how electrons move in a paired dance inside superconductors, or how they form orbitals and bonds in a chemical system, the motion of electrons are at the core.
Seven years after Noether published her blockbuster results, Wolfgang Pauli made waves when he published the work describing his Pauli exclusion principle, which relies heavily on symmetry to explain basic tenets of quantum theory. Pauli’s principle has enormous consequences; for starters, describing how the objects we interact with every day are solid even though atoms are mostly empty space, and outlining the rules of bonds, orbitals, and all of chemistry, among other things.
Symmetry in motion
It is Pauli's symmetry, coupled with a deep respect for the impact of symmetry, that led our team at Quantinuum to the discovery published today.
In their work, they considered the act of designing quantum algorithms, and how one’s design choices may lead to efficiency or inefficiency.
When you design quantum algorithms, there are many choices you can make that affect the final result. Extensive work goes into optimizing each individual step in an algorithm, requiring a cyclical process of determining subroutine improvements, and finally, bringing it all together. The significant cost and time required is a limiting factor in optimizing many algorithms of interest.
This is again where symmetry comes into play. The authors realized that by better exploiting the deepest symmetries of the problem, they could make the entire edifice more efficient, from state preparation to readout. Over the course of a few years, a team lead Dr. Fitzpatrick and his colleague Jędrzej Burkat slowly polished their approach into a full algorithm for performing the QPT.
The QPT functions by using Pauli’s symmetry to discard unimportant details and strip the problem down to its bare essentials. Starting with a Paldus transform allows the algorithm designer to enjoy knock-on effects throughout the entire structure, making it overall more efficient to run.
“It’s amazing to think how something we discovered one hundred years ago is making quantum computing easier and more efficient,” said Dr. Nathan Fitzpatrick.
Ultimately, this innovation will lead to more efficient quantum simulation. Projects we believed to still be many years out can now be realized in the near term.
Transforming the future
The discovery of the Quantum Paldus Transform is a powerful reminder that enduring ideas—like symmetry—continue to shape the frontiers of science. By reaching back into the fundamental principles laid down by pioneers like Noether and Pauli, and combining them with modern quantum algorithm design, Dr. Fitzpatrick and Mr. Burkat have uncovered a tool with the potential to reshape how we approach quantum computation.
As quantum technologies continue their crossover from theoretical promise to practical implementation, innovations like this will be key in unlocking their full potential.
In a new paper in Nature Physics, we've made a major breakthrough in one of quantum computing’s most elusive promises: simulating the physics of superconductors. A deeper understanding of superconductivity would have an enormous impact: greater insight could pave the way to real-world advances, like phone batteries that last for months, “lossless” power grids that drastically reduce your bills, or MRI machines that are widely available and cheap to use. The development of room-temperature superconductors would transform the global economy.
A key promise of quantum computing is that it has a natural advantage when studying inherently quantum systems, like superconductors. In many ways, it is precisely the deeply ‘quantum’ nature of superconductivity that makes it both so transformative and so notoriously difficult to study.
Now, we are pleased to report that we just got a lot closer to that ultimate dream.
Making the impossible possible
To study something like a superconductor with a quantum computer, you need to first “encode” the elements of the system you want to study onto the qubits – in other words, you want to translate the essential features of your material onto the states and gates you will run on the computer.
For superconductors in particular, you want to encode the behavior of particles known as “fermions” (like the familiar electron). Naively simulating fermions using qubits will result in garbage data, because qubits alone lack the key properties that make a fermion so unique.
Until recently, scientists used something called the “Jordan-Wigner” encoding to properly map fermions onto qubits. People have argued that the Jordan-Wigner encoding is one of the main reasons fermionic simulations have not progressed beyond simple one-dimensional chain geometries: it requires too many gates as the system size grows.
Even worse, the Jordan-Wigner encoding has the nasty property that it is, in a sense, maximally non-fault-tolerant: one error occurring anywhere in the system affects the whole state, which generally leads to an exponential overhead in the number of shots required. Due to this, until now, simulating relevant systems at scale – one of the big promises of quantum computing – has remained a daunting challenge.
Theorists have addressed the issues of the Jordan-Wigner encoding and have suggested alternative fermionic encodings. In practice, however, the circuits created from these alternative encodings come with large overheads and have so far not been practically useful.
We are happy to report that our team developed a new way to compile one of the new, alternative, encodings that dramatically improves both efficiency and accuracy, overcoming the limitations of older approaches. Their new compilation scheme is the most efficient yet, slashing the cost of simulating fermionic hopping by an impressive 42%. On top of that, the team also introduced new, targeted error mitigation techniques that ensure even larger systems can be simulated with far fewer computational "shots"—a critical advantage in quantum computing.
Using their innovative methods, the team was able to simulate the Fermi-Hubbard model—a cornerstone of condensed matter physics— at a previously unattainable scale. By encoding 36 fermionic modes into 48 physical qubits on System Model H2, they achieved the largest quantum simulation of this model to date.
This marks an important milestone in quantum computing: it demonstrates that large-scale simulations of complex quantum systems, like superconductors, are now within reach.
Unlocking the Quantum Age, One Breakthrough at a Time
This breakthrough doesn’t just show how we can push the boundaries of what quantum computers can do; it brings one of the most exciting use cases of quantum computing much closer to reality. With this new approach, scientists can soon begin to simulate materials and systems that were once thought too complex for the most powerful classical computers alone. And in doing so, they’ve unlocked a path to potentially solving one of the most exciting and valuable problems in science and technology: understanding and harnessing the power of superconductivity.
The future of quantum computing—and with it, the future of energy, electronics, and beyond—just got a lot more exciting.