The Quantum Agent

In the cult movie classic The Matrix, the disturbingly nihilistic ’Agent Smith’—an artificially intelligent computer program who looks like an FBI agent and acts like a prison guard in a simulated reality—seems smarter and more powerful than the other agents in the system. He goes off-book to make his own decisions independent of his controller and to cunningly anticipate attacks; he can replicate himself and, just when you think he has been deleted, he sneakily reappears.

There are perhaps some eerie similarities between Agent Smith and many of the questions tossed about by quantum physicists. What properties of an agent give rise to free will, or the appearance of it? Can you make perfect copies of information? Is something really deleted when you think it is?

To a quantum physicist, an ’agent’ isn’t always of the FBI variety. It doesn’t even need to be ’intelligent’ in the way that we normally think about that word. All it needs to do is observe something, process that information, and act upon it. A simple logic gate in computing—the little switch that says "if the input is x, then do y"—is an agent. But there is still some mystery surrounding the phenomenon, particularly at small scales. Here quantum physics applies and particles can do odd things, like be in two places at once, or travel along separate paths simultaneously. Fully describing quantum-scale agents, and understanding how they might give rise to the everyday agents we see in our macroscopic ’classical’ world, is a work-in-progress.

Most of the things we typically ascribe with intelligent agency, after-all, exist in the human-scale world of classical physics, not the microscopic quantum realm. So how does that agency emerge? As University of Maryland Baltimore County physicist Sebastian Deffner puts it, "Is there some underlying physical reason why agents tend to look classical?"

A group of researchers are tackling this question from all angles as part of an FQXi grant round called ’Agency in the Physical World’—a program that has spread US$2 million amongst 24 research projects, including Deffner’s, which has been awarded over $111,000. The unifying topic for these projects: How quantum measurement might lead to the emergence of agency and (bonus points for this part) the conscious mind. Intelligence and free will, after all, are the ultimate manifestations of ’agency.’

The work is interesting not just philosophically, but should feed into efforts to make quantum computers and optimize quantum machines, which might process information in novel ways. Understanding agency is synonymous with understanding how quantum systems are controlled. "Right now, we have extremely sophisticated techniques for controlling airplanes and things like that. We have all sorts of instruments," says another FQXi grant-winner Aephraim Steinberg, a physicist at the University of Toronto, who has been awarded $138,000. "These days we want systems that control quantum devices. So we need an analogue of that."

Erasing Information

While many of the projects funded under FQXi’s Agency in the Physical World program are theoretical, Steinberg’s project is perhaps unusual in that it contains an experimental side: a set-up designed to probe, specifically, how quantum-scale agents gather information, and—importantly—how that information gets erased.

It is a standard tenet of quantum physics that if you measure something, you will interact with it such a way that you inevitably disturb it. This isn’t an issue in the classical world: look at an airplane and it doesn’t drop from the sky. But if you try to observe the properties of a tiny object, such as a particle of light—or photon—then you could seriously alter it, perhaps to the point of destruction. The more perfectly you try to pin it quantum characteristics down, the more completely you disturb it. You may even annihilate its quantum properties entirely.

The destructive power of measurements can be seen, for example, when physicists investigate the way that photons interfere. In so-called ’interferometer experiments,’ photons are made to travel along different paths in an instrument and then recombine, creating an interference pattern. This pattern is a quantum effect created because photons can behave in a wave-like manner, existing in a superposition that enables them to travel both paths at the same time. But when physicists try to peer at the system closely to see how exactly photons are able to perform this trick, traveling along both paths simultaneously, the quantum superposition is destroyed. The photons then just behave like classical particles, traveling along either one path or the other, but not both, and so the interference pattern disappears.

There is a cunning way around this problem, however. Physicists can avoid observing the photon’s path directly—and thus avoid completely destroying the interference pattern—while still learning something about the route that it takes. They do this by storing information about the path a photon has taken in the state of that photon. One possibility is to encode the path information in the direction in which the photon is polarised—that is, whether it vibrates up and down, or side to side. "It’s more gentle in its measurement," says Steinberg. "We can use a photon’s polarization as a scratch pad to write down its path. It is not an irreversible disturbance."

So why doesn’t measuring the path information in this indirect manner irrevocably change the photon’s properties? The key, Steinberg explains, is to erase the information before it "gets out" into the universe. But there are different degrees of erasure, and thus different implications for the interference pattern. "If you fail to erase the information completely, then the interference is destroyed," says Steinberg.

There are different ways of erasing quantum information, including the computer-programmers’ trick of swapping out information with a load of junk data. "We can turn the erasure on and off, or partially erase; the more information you keep the less interference you have," says Steinberg. There is a whole area of quantum physics called ’optimal quantum control’ that aims to optimize this effect: how much disturbance you can cause through observation and still retain a usable, controllable system.

Researchers can set up a system where if a photon takes path A, it is erased one way, and if it takes path B, it is erased another way. Either way, the information should be erased, right? But it is already known that superposition provides a loophole. If the photon travels both paths, this can allow some of that information to survive erasure (just like our sneaky Agent Smith).

Steinberg is probing these ideas experimentally. To begin with, his student Arthur Pang is constructing an interferometer that can be used to interrogate the destructiveness of various types of quantum measurement. "His project will interrogate where that information lives," says Steinberg.

Ultimately, Steinberg notes, they will use such experimental setups to assess whether some quantumly-stored information can be used to perform logical operations. Such knowledge is important, Steinberg notes, for assessing the ultimate power of a quantum computer. Physicists already think that quantum computers will be much more powerful than classical computers at tackling certain specific tasks. But if they can also exploit quantumly-stored information, they could be even more powerful. "Understanding the limits is important," says Steinberg.

No Cloning

Deffner describes Steinberg’s work as "really, really cool and very interesting." He notes that it may have wide implications for our understanding of the foundations of physics. In particular, it relates to the thorny problem of how the fuzzy quantum world, where particles can take on myriad contradictory properties, transitions to the everyday classical world we see around us, where characteristics are clearly defined. A major difference between these two realms is related to the issue of copying information. Steinberg’s work on measurement, information-storage and erasure is intimately tied to a fundamental theorem of quantum physics that says quantum information can’t be perfectly copied.

This quantum no-cloning rule holds because any attempt to exactly copy the quantum properties of a particle involves first perfectly measuring its state, destroying its properties in the process. Yet "we have no problem copying classical information," notes Deffner. Thus, Steinberg’s work should help unpick what’s going on at the transition between the quantum and classical worlds—the boundary between the realm in which copying is possible and impossible.

One of the physicists who first helped to articulate the no-cloning rule is quantum physicist Wojciech Zurek, who has been awarded $75,000 in this FQXi grant round—and is Deffner’s former post-doctoral supervisor. Zurek has his own take on how classical behaviors emerge from the quantum world: a theory inspired by biology that he calls ’Quantum Darwinism.’ At the quantum level, multiple possibilities occur simultaneously—a photon can be in a superposition allowing it travel along both path A and path B. But when observed, this superposition snaps and the photon behaves classically, travelling down only one or other of the two paths, but not both. But how does the particle decide which path to settle into? "If you look around in our world, everything seems to be classical, but we know that under the hood it’s all quantum. So how does classicality emerge?" summarizes Deffner. "That’s a big question, but Zurek has a unique approach."

One popular idea for the mechanism through which the quantum turns classical is called "decoherence." According to this idea, a quantum system can become classical through interactions with the environment. These interactions in effect serve as a series of observations. If a photon strikes a leaf, for instance, the leaf in some sense ’observes’ the photon, destroying its quantum behaviour. Zurek’s idea builds on that to try to come up with a mechanism for how one of the quantum possibilities remains while the others fizzle out. Inspired by Charles Darwin, Zurek argues that "fittest" quantum option will survive—the option for which information is more readily copied and intercepted by multiple agents. These agents will all agree on what they have seen, generating an objective classical reality. Zurek’s new project investigates how the information gathering done by agents fits in to this Darwinian theory.

Quantum Thermodynamics

Meanwhile Deffner’s own project is a purely theoretical affair. His background is in classical-scale thermodynamics: a field that was basically invented to understand things like steam engines on a statistical level, when we didn’t have access to all the information about what’s going on at a microscopic scale. Classic thermodynamics ignores all the details about what any given atom of vapor is doing inside a chamber of gas; no one needs to know that to build a functional engine. Instead, engineers work with properties that relate to large-scale groups of particles confined in a chamber—pressure, temperature and volume—to describe what a system is doing.

These days, however, scientists and engineers are building machines on a nanoscale, where quantum scale processes are harder to ignore. On these scales, physicists need to start thinking about the behavior of individual particles; large-scale concepts that apply to groups of particles, such as volume and temperature, have little meaning. "Those are probably not the right variables on the quantum scales," says Deffner. His group is trying to come to grips with a new vocabulary for writing the quantum-thermodynamic rule book.

On the quantum scale, even the basic second law of thermodynamics gets complicated. That’s the law that tells us the entropy of a system—a measure of the disorder of the particles contained within—always spontaneously increases, if there are no outside influences keeping it in check. In practice, it means that when you try to construct a machine, not all of the energy you put in will be converted to useful work. Some will always be wasted, heating up the environment, for instance. "We think entropy increases as we go forward in time, but honestly a lot of things are swept under the rug when we say that," laughs Steinberg.

Physicists have long known that an intelligent agent could use information about the state of a system to power a reversal of entropy (see "Thermo-Demonics"). Researchers have even constructed, for example, tiny machines that can use information about a few photons as energy to charge a battery. In 2015, researchers reported an "information-powered fridge" that used information fuel to cool a system.

For his FQXi-funded work, Deffner and his group will try to write a quantum field-theory set of rules for how such systems work at the sub-atomic scale. They are interested in questions like what happens if the small agent powering an entropy-reversing system experiences a delay between when it gathers information and when it acts on it.

While all these projects focus on different aspects of agency in the physical world, they highlight how we need to expand our understanding of what being an agent entails in the micro realm, and how that relates to the world around us. We tend to think of agency as something external to a system, notes Deffner—someone standing outside a machine flicks a switch, or Agent Smith acts on Neo and the other rebels in the Matrix. But, in the end, the universe is a closed system—there’s nothing outside it looking in and acting on it. So, agency (including everything from a quantum logic gate to the concept of human ’free will’) has to emerge from the basic physics rules of the universe. Deffner, Steinberg and others like them are aiming to figure out how.