The Quantum Clock-Maker Investigating COVID-19, Causality, and the Trouble with AI

Sally Shrapnel embodies the word "interdisciplinary."

In the midst of the global fight against COVID-19, Shrapnel has her hands full—drawing on all the strands of her astonishingly-varied academic expertise across quantum physics, artificial intelligence (AI), philosophy, and medical practice. For now, she is spending half her time on a machine-learning model that will hopefully help doctors single out COVID-19 patients at highest risk for kidney failure; and half her time helping build a quantum clock that will test our understanding of time and the limits of its measurement.

Shrapnel’s professional trajectory has been unique; most quantum physicists do not also have a 20-year career as a medical doctor, after all. "My journey has been very non-traditional," laughs Shrapnel. "I’ve been driven totally by curiosity and a drive to understand how the world works." she muses. "As an academic you can get stuck doing one thing. So, I follow my nose." To date, her nose has led her to a post at the ARC Centre of Excellence for Engineered Quantum Systems at the University of Queensland, in Brisbane, Australia.

"Sally (Shrapnel) is one of those people who have just an incredible array of expertises," says Eric Cavalcanti, a quantum physicist at Griffith University, also in Queensland. "I think because of the breadth of her knowledge she has a special way at looking at the big picture and pinning down what’s important in a problem."

For Shrapnel, the thing that ties together all her work is the problem of spurious correlations, and the curiously-human ability to string together cause and effect.

Just because two things happen one after the other, this doesn’t always mean that one caused the other. As scientists and doctors alike are fond of saying: correlation is not causation. While people are sometimes tricked into seeing causation where only correlation exists (thinking that a pill cured a cold, for example, when really it was just the passage of time), they are usually pretty good at figuring this out in the everyday world. If ice cream sales and drownings both go up when it’s hot, for example, people intuitively understand that the weather lies behind both phenomena, rather than panicking about killer ice cream. AI systems, on the other hand, are particularly bad at it: they can pinpoint correlations extremely well, but are often distracted by spurious information and can leap to all kinds of false conclusions about cause and effect. People are good at ignoring extraneous information. Today’s AI is not.

Shrapnel is driven by curiosity to explore causation in all kinds of ways. In the classical world, cause comes before effect; time flows forwards; and entropy always increases—in other words, the three fundamental arrows of causation, time, and thermodynamics all point the same way. Is that true in the quantum world too? Can we build a framework for understanding causation in the quantum realm? And can quantum computers, or AI built from ’quantum agents,’ learn to better determine if one thing causes another? It’s a line of thinking that leads to questions that are far more practical, too: what, Shrapnel is asking, causes some COVID-19 patients to suffer kidney failure while others don’t, and can AI work that out?

Big Picture Thinking

Shrapnel started out at university wanting to be an astrophysicist. "But I was very social," she remembers. "I was afraid I would end up stuck in a radio telescope on my own somewhere." She switched to a medical degree and worked as a doctor for a year. Then she missed "big picture" thinking and returned to physics. With a masters in physics and bioengineering under her belt, she worked as a general practitioner for many years, until the curiosity bug bit her again and she went back to studying, this time with kids in elementary school, for a philosophy degree. "I was constantly giving scientific explanations to patients, and I was interested in the theory of that, how we know what’s true, the nature of explanation," Shrapnel says.

It was philosophy that gently nudged her into the quantum world. "I got really interested in this idea of causality—it has to do with so much about science," says Shrapnel. But whenever she read up about the topic, "there was always this footnote saying: this is true for everything except quantum physics." In the classical world, if A causes B then A must happen before B in time; in the quantum world, there’s some fuzziness about whether that needs to hold true—entanglement muddles with theories about causation. Shrapnel was fascinated.

Shrapnel’s PhD, with co-advisors in philosophy and quantum physics, led to a post-doc in the quantum realm, which in turn led to artificial intelligence as a model for a structured way of thinking about cause-effect relations. "At that time, it was impossible not to be interested in machine learning and its implications for medicine," she adds.

Machine learning is a brand of artificial intelligence where a computer is simply fed masses of data and tries to find patterns in that data by itself, much as a baby tries to learn about the world. The results can be very useful. Machine learning systems can spot useful correlations that had escaped human notice, or help make predictions for systems where people don’t know all the rules (like the risk of extreme weather events under climate change). Other times, the results are entirely misleading. Machine learning systems fed X-ray images might conclude, for example, that sickness correlates to the time of day an image is taken, because a greater proportion of night-time scans are from emergencies rather than routine procedures. Thus far, AI doesn’t have the "common sense" to figure out when links are spurious.

Shrapnel’s medical project is feeding machine learning systems with masses of data from hospital admissions of COVID-19 patients, including patient facts like age, gender, vital statistics, co-morbidities and disease progression, and trying to work out the earliest warning signs that their illness might lead to kidney failure. Catch it quick, and doctors can target their treatments to avoid catastrophe. The hard part, she notes, will be taking data that mostly comes from developed countries and trying to apply that to other parts of the world, without having the system be tricked by biases in the data. (Such biases have proven incredibly problematic in the past, as datasets can be skewed by factors like systemic racism.) For now, Shrapnel’s group is still trawling through about 350,000 patient records. "The algorithm will be finalized by the end of this year," says Shrapnel. "The next two years is to deploy it in a clinical trial to see if it actually improves patient outcomes."

Quantum Agency

The topic of causation was also at the heart of Shrapnel’s recent theoretical project—funded with a grant of almost US$90,000 from FQXi—on quantum agency.

An ’agent’ is just a thing that takes some action or information and does something with it; it measures or creates a cause-effect relationship. In this project, led by Shrapnel’s physicist colleague and former PhD supervisor Gerard Milburn, they imagined a very simple agent: something that sends out a pulse of light and measures its return to see how it changed. Such a system needs an actuator to send out the light; a sensor to measure its return; and some kind of brain with a memory to figure out what it all means. That’s all easily achieved in the classical world, using regular machine parts.

But what about in the quantum realm? Their first task was to show, theoretically, that if a classical actuator and sensor are replaced with tiny quantum versions of the same, then a classical brain will make fewer errors and learn faster. This makes sense, says Shrapnel. Quantum states are really fragile, which is a problem if you’re trying to use them to make a computer. But it makes them exquisitely sensitive detectors. They published their theoretical proof in March (M. J. Kewming, S. Shrapnel, and G. J. Milburn, Phys. Rev. A 103, 032411 (2021)).

Their second task was to investigate whether such a quantum agent needs to have the same understanding of the passage of time as the outside world, or whether its own internal timekeeping is enough for it to learn and understand the arrow of causation. Their theoretical work on these questions hints that even inside a quantum agent, time’s arrow and the arrow of causation are in alignment (just as they are in the everyday world). Perhaps there’s something fundamental to agency, they wonder, that requires this to be so (G. J. Milburn and S. Shrapnel, arXiv:2009.04121 (2020)).

This research "could help shed light on long-standing foundational questions such as the relationship between agency and causality," says Cavalcanti, who has also delved into the knotty issue of causation in the quantum world.

Quantum Timekeeping

Meanwhile, Shrapnel’s other quantum project—backed by an FQXi grant of almost $950,000—is much more tangible, and involves building and manipulating a novel kind of quantum clock, one specifically designed to run entirely inside a quantum computer.

There are plenty of quantum clocks in the world—tiny devices that help to keep or measure time. The one being designed by Shrapnel and her co-investigators (University of Queensland’s Arkady Fedorov and Milburn) will be different in two ways. First, it is intended to operate using the same system of qubits, on a quantum chip, as would be used in a universal quantum computer. This gives it a highly useful edge: the self-contained clock might eventually be useful for keeping time within a quantum computer. "The aspirational goal is that we produce a functional clock useful for quantum computing," says Shrapnel.

Second, importantly, the team intends to have an extraordinary level of control over their clock.

The clock will be made to ’tick’ by prodding it with a microwave pulse. That energy input will compete with spontaneous energy loss, creating a self-sustaining oscillation. The clock’s resulting timekeeping will be measured by looking at the electromagnetic field and current around it. The whole set-up is about the size of kitchen refrigerator, mostly consisting of a refrigeration system to cool the chip to millikelvin temperatures and wires for monitoring the clock pulse. The chip itself is the size of a fingernail.

Quantum noise is expected to create fluctuations in the time between ticks. The theory is that in the quantum domain, the more energy the clock loses due to spontaneous emission, the smaller these fluctuations will be and thus the better the clock.

They should also be able to drive the clock not with microwave pulses, but through the very act of measuring its quantum properties via radiated energy—thus using information about the clock as a kind of fuel. That adds another layer to their interrogations about the nature of time and thermodynamics.

So far, the team has designed the clock, and is now working on fabrication. "This is a 3-year project. And we’re just at the start,” says Shrapnel. There are some big technical hurdles, she adds: "It really is an engineering problem. It’s a big optimization task at the moment, to limit noise and all those sorts of things."

The pandemic has, unsurprisingly, not helped either. It "really set back the experimental part because we can’t get parts," says Shrapnel. "So, we’re trying to work out how to do it with what we’ve got."

Again, the fundamental mission will be to probe if and how time’s arrow, causation, and thermodynamics are in lock step (or not) in a quantum system. "Are they still aligned in the quantum world? Are the relationships between them the same as in the classical world?" asks Shrapnel.

"The quantum to classical transition is a very complex process, so starting out with ’quantum agents’ that can still be fully described and accessed is a great start," says Marcus Huber at the Institute for Quantum Optics and Quantum Information in Vienna, who likes Shrapnel’s approach to these issues.

In the process of investigating all that, they will hopefully also help to establish the fundamental limit on how well time can be measured—by anyone, using any system. "Can we work out what the most accurate clock would be?" Shrapnel asks.

The project gets to the heart of some interesting and useful questions, adds Huber. "If we are to understand the nature of time, we should uncover the thermodynamic principles allowing for its measurement in the first place," he says.

Shrapnel recently brought all these results and ideas to a workshop she co-hosted in February on engineered quantum agents. Unusually, in this pandemic world, they were able to meet in person, given Australia’s success at that time in the battle against COVID-19. The goal of the conference was to define quantum-engineered agents, and to get consensus across philosophers, physicists, and social scientists about the goals and ethics of (one day) making super-effective quantum-based machine-learning systems that can unravel causation.

"By having a deeper understanding of what causality is, we have a better chance of designing machine-learning systems that can do this better," says Shrapnel. "We want to identify cause-effect relations from the get go."