Friday, November 22, 2019

What can artificial intelligence do for physics? And what will it do to physics?

Neural net illustration. Screenshot from this video.

In the past two years, governments all over the world have launched research initiatives for Artificial Intelligence (AI). Canada, China, the United States, the European Commission, Australia, France, Denmark, the UK, Germany – everyone suddenly has a strategy for “AI made in” whatever happens to be their own part of the planet. In the coming decades, it is now foreseeable, tens of billions of dollars will flow into the field.

But ask a physicist what they think of artificial intelligence, and they’ll probably say “duh.” For them, AI was trendy in the 1980s. They prefer to call it “machine learning” and pride themselves having used it for decades.

Already in the mid 1980s, researchers working in statistical mechanics – a field concerned with the interaction of large number of particles – set out to better understand how machines learn. They noticed that magnets with disorderly magnetization (known as “spin glasses”) can serve as a physical realization for certain mathematical rules used in machine learning. This in turn means that the physical behavior of these magnets shed light on some properties of learning machines, such as their storage capacity. Back then, physicists also used techniques from statistical mechanics to classify the learning abilities of algorithms.

Particle physicists, too, were on the forefront of machine learning. The first workshop on Artificial Intelligence in High Energy and Nuclear Physics (AIHENP) was held already in 1990. Workshops in this series still take place, but have since been renamed to Advanced Computing and Analysis Techniques. This may be because the new acronym, ACAT, is catchier. But it also illustrates that the phrase “Artificial Intelligence” is no longer common use among researchers. It now appears primarily as an attention-grabber in the mass media.

Physicists avoid the term “Artificial Intelligence” not only because it reeks of hype, but because the analogy to natural intelligence is superficial at best, misleading at worst. True, the current models are loosely based on the human brain’s architecture. These so-called “neural networks” are algorithms based on mathematical representations of “neurons” connected by “synapses.” Using feedback about its performance – the “training” – the algorithm then “learns” to optimize a quantifiable goal, such as recognizing an image, or predicting a data-trend.

This type of iterative learning is certainly one aspect of intelligence, but it leaves much wanting. The current algorithms heavily rely on humans to provide suitable input data. They do not formulate own goals. They do not propose models. They are, as far as physicists are concerned, but elaborate ways of fitting and extrapolating data.

But then, what novelty can AI bring to physics? A lot, it turns out. While the techniques are not new – even “deep learning” dates back to the early 2000s – today’s ease of use and sheer computational power allows physicists to now assign computers to tasks previously reserved for humans. It has also enabled them to explore entirely new research directions. Until a few years ago, other computational methods often outperformed machine learning, but now machine learning leads in many different areas. This is why, in the past years, interest in machine learning has spread into seemingly every niche.

Most applications of AI in physics loosely fall into three main categories: Data analysis, modeling, and model analysis.

Data analysis is the most widely known application of machine learning. Neural networks can be trained to recognize specific patterns, and can also learn to find new patterns on their own. In physics, this is used in image analysis, for example when astrophysicists search for signals of gravitational lensing. Gravitational lensing happens when space-time around an object is deformed so much that it noticeably distorts the light coming from behind it. The recent, headline-making, black hole image is an extreme example. But most gravitational lensing events are more subtle, resulting in smears or partial arcs. AIs can learn to identify these.

Particle physicists also use neural networks to find patterns, both specific and unspecific ones. Highly energetic particle collisions, like those done at the Large Hadron Collider, produce huge amounts of data. Neural networks can be trained to flag interesting events. Similar techniques have been used to identify certain types of gamma ray bursts, and may soon help finding gravitational waves.

Machine learning aids the modeling of physical systems both by speeding up calculations and by enabling new types of calculations. For example, simulations for the formation of galaxies take a long time even on the current generation of super-computers. But neural networks can learn to extrapolate from the existing simulations, without having to re-run the full simulation each time, a technique that was recently successfully used to match the amount of dark matter to the amount of visible matter in galaxies. Neural networks have also been used to reconstruct what happens when cosmic rays hit the atmosphere, or how elementary particles are distributed inside composite particles.

For model analysis, machine learning is applied to understand better the properties of already known theories which cannot be extracted by other mathematical methods, or to speed up computation. For example, the interaction of many quantum particles can result in a variety of phases of matter. But the existing mathematical methods have not allowed physicists to calculate these phases. Neural nets can encode the many quantum particles and then classify the different types of behavior.

Similar ideas underlie neural networks that seek to classify the properties of materials, such as conductivity or compressibility. While the theory for the materials’ atomic structure is known in principle, many calculations have so-far exceeded the existing computational resources. Machine learning is beginning to change that. Many hope that it may one day allow physicists to find materials that are superconducting at room temperature. Another fertile area for applications of neural nets is “quantum tomography,” that is the reconstruction of quantum state from the measurements performed on it, a problem of high relevance for quantum computing.

And it is not only that machine learning advances physics, physics can in return advance machine learning. At present, it is not well understood just why neural nets work as well as they do. Since some neural networks can be represented as physical systems, knowledge from physics may shed light on the situation.

In summary, machine learning rather suddenly allows physicists to tackle a lot of problems that were previously intractable, simply because of the high computational burden.

What does this mean for the future of physics? Will we see the “End of Theory” as Chris Anderson oracled in 2008?

I do not think so. There are many different types of neural networks, which differ in their architecture and learning scheme. Physicists now have to understand which algorithm works for which case and how well, the same way they previously had to understand which theory works for which case and how well. Rather than spelling the end of theory, machine learning will take it to the next level.

[You can help me keep my writing freely available by using the donate button in the top right corner of the page.]

Wednesday, November 20, 2019

Can we tell if there’s a wormhole in the Milky-Way?

This week I got a lot of questions about an article by Dennis Overbye in the New York Times, titled “How to Peer Through a Wormhole.” This article says “Theoretically, the universe may be riddled with tunnels through space and time” and goes on to explain that “Wormholes are another prediction of Einstein’s theory of general relativity, which has already delivered such wonders as an expanding universe and black holes.” Therefore, so Overbye tells his readers, it is reasonable to study whether the black hole in the center of our Milky Way is such a wormhole.

The trouble with this article is that it makes it appear as if wormholes are a prediction of general relativity comparable to the prediction of the expansion of the universe and the prediction of black holes. But this is most definitely not so. Overbye kind of says this by alluding to some “magic” that is necessary to have wormholes, but unfortunately he does not say it very clearly. This has caused quite some confusion. On twitter, for example, Natalie Wolchover, has put wormholes on par with gravitational waves.

So here are the facts. General Relativity is based on Einstein’s field equations which determine the geometry of space-time as a consequence of the energy and matter that is in that space-time. General Relativity has certain kinds of wormholes as solutions. These are the so-called Einstein-Rosen bridges. There are two problems with those.

First, no one knows how to create them with a physically possible process. It’s one thing to say that the solution exists in the world of mathematics. It’s another thing entirely to say that such a solution describes something in our universe. There are whole books full with solutions to Einstein’s field equations. Most of these solutions have no correspondence in the real world.

Second, even leaving aside that they won’t be created during the evolution of the universe, nothing can travel through these wormholes.

If you want to keep a wormhole open, you need some kind of matter that has a negative energy density, which is stuff that for all we know does not exist. Can you write down the mathematics for it? Yes. Do we have any reason whatsoever to think that this mathematics describes the real world? No. And that, folks, is really all there is to say about it. It’s mathematics and we have no reason to think it’s real.

In this, wormholes are very, very different to the predictions of the expanding universe, gravitational waves, and black holes. The expanding universe, gravitational waves and black holes are consequences of general relativity. You have to make an effort to avoid that they exist. It’s the exact opposite with wormholes. You have to bend over backwards to make the math work so that they can exist.

Now, certain people like to tell me that this should count as “healthy speculation” and I should stop complaining about it. These certain people are either physicists who produce such speculations or science writers who report about it. In other words, they are people who make a living getting you to believe this mathematical fiction. But there is nothing healthy about this type of speculation. It’s wasting time and money that would be better used on research that could actually advance physics.

Let me give you an example to see the problem. Suppose the same thing would happen in medicine. Doctors would invent diseases that we have no reason to think exist. They would then write papers about how to diagnose those invented diseases and how to cure those invented diseases and, for good measure, argue that someone should do an experiment to look for their invented diseases.

Sounds ridiculous? Yeah, it is ridiculous. But that’s exactly what is going on in the foundations of physics, and it has been going on for decades, which is why no one sees anything wrong with it anymore.

Is there at least something new that would explain why the NYT reports on this? What’s new is that two physicists have succeeded in publishing a paper which says that if the black hole in the center of our galaxy is a traversable wormhole then maybe we might be able to see this. The idea is that if there is stuff moving around the other end of the wormhole then we might notice the gravitational influence of that stuff on our side of the wormhole.

Is it possible to look for this? Yes, it is also possible to look for alien spaceships coming through, and chances are, next week a paper will get published about this and the New York Times reports it.

On a more technical note, a quick remark about the paper, which you find here:
The authors look at what happens with the gravitational field on one side of a non-traversable wormhole if a shell of matter is placed around the other side of the wormhole. They conclude:
“[T]he gravitational field can cross from one to the other side of the wormhole even from inside the horizon... This is very interesting since it implies that gravity can leak even through the non-traversable wormhole.”
But the only thing their equation says is that the strength of the gravitational field on one side of the wormhole depends on the matter on the other side of the wormhole. Which is correct of course. But there is no information “leaking” through the non-traversable (!) wormhole because it’s a time-independent situation. There is no change that can be measured here.

This isn’t simply because they didn’t look at the time-dependence, but because the spherically symmetric case is always time-independent. We know that thanks to Birkhoff’s theorem. We also know that gravitational waves have no monopole contribution, so there are no propagating modes in this case either.

The case that they later discuss, the one that is supposedly observable, instead talks of objects on orbits around the other end of the wormhole. This is, needless to say, not a spherically symmetric case and therefore this argument that the effect is measurable for non-traversable wormholes is not supported by their analysis. If you want more details, this comment gets it right.

Friday, November 15, 2019

Did scientists get climate change wrong?

On my recent trip to the UK, I spoke with Tim Palmer about the uncertainty in climate predictions.

Saturday, November 09, 2019

How can we test a Theory of Everything?

How can we test a Theory of Everything? That’s a question I get a lot in my public lectures. In the past decade, physicists have put forward some speculations that cannot be experimentally ruled out, ever, because you can always move predictions to energies higher than what we have tested so far. Supersymmetry is an example of a theory that is untestable in this particular way. After I explain this, I am frequently asked if it is possible to test a theory of everything, or whether such theories are just entirely unscientific.

It’s a good question. But before we get to the answer, I have tell you exactly what physicists mean by “theory of everything”, so we’re on the same page. For all we currently know the world is held together by four fundamental forces. That’s the electromagnetic force, the strong and the weak nuclear force, and gravity. All other forces, like for example Van-der-Waals forces that hold together molecules or muscle forces derive from those four fundamental forces.

The electromagnetic force and the strong and the weak nuclear force are combined in the standard model of particle physics. These forces have in common that they have quantum properties. But the gravitational force stands apart from the three other forces because it does not have quantum properties. That’s a problem, as I have explained in an earlier video. A theory that solves the problem of the missing quantum behavior of gravity is called “quantum gravity”. That’s not the same as a theory of everything.

If you combine the three forces in the standard model to only one force from which you can derive the standard model, that is called a “Grand Unified Theory” or GUT for short. That’s not a theory of everything either.

If you have a theory from which you can derive gravity and the three forces of the standard model, that’s called a “Theory of Everything” or TOE for short. So, a theory of everything is both a theory of quantum gravity and a grand unified theory.

The name is somewhat misleading. Such a theory of everything would of course *not explain everything. That’s because for most purposes it would be entirely impractical to use it. It would be impractical for the same reason it’s impractical to use the standard model to explain chemical reactions, not to mention human behavior. The description of large objects in terms of their fundamental constituents does not actually give us much insight into what the large objects do. A theory of everything, therefore, may explain everything in principle, but still not do so in practice.

The other problem with the name “theory of everything” is that we will never know that not at some point in the future we will discover something that the theory does not explain. Maybe there is indeed a fifth fundamental force? Who knows.

So, what physicists call a theory of everything should really be called “a theory of everything we know so far, at least in principle.”

The best known example of a theory of everything is string theory. There are a few other approaches. Alain Connes, for example, has an approach based on non-commutative geometry. Asymptotically safe gravity may include a grand unification and therefore counts as a theory of everything. Though, for reasons I don’t quite understand, physicists do not normally discuss asymptotically safe gravity as a candidate for a theory of everything. If you know why, please leave a comment.

These are the large programs. Then there are a few small programs, like Garrett Lisi’s E8 theory, or Xiao-Gang Wen’s idea that the world is really made of qbits, or Felix Finster’s causal fermion systems.

So, are these theories testable?

Yes, they are testable. The reason is that any theory which solves the problem with quantum gravity must make predictions that deviate from general relativity. And those predictions, this is really important, cannot be arbitrarily moved to higher and higher energies. We know that because combining general relativity with the standard model, without quantizing gravity, just stops working near an energy known as the Planck energy.

These approaches to a theory of everything normally also make other predictions. For example they often come with a story about what happened in the early universe, which can have consequences that are still observable today. In some cases they result in subtle symmetry violations that can be measurable in particle physics experiments. The details about this differ from one theory to the next.

But what you really wanted to know, I guess, is whether these tests are practically possible any time soon? I do think it is realistically possible that we will be able to see these deviations from general relativity in the next 50 years or so. About the other tests that rely on models for the early universe or symmetry violations, I’m not so sure, because for these it is again possible to move the predictions and then claim that we need bigger and better experiments to see them.

Is there any good reason to think that such a theory of everything is correct in the first place? No. There is good reason to think that we need a theory of quantum gravity, because without that the current theories are just inconsistent. But there is no reason to think that the forces of the standard model have to be unified, or that all the forces ultimately derive from one common explanation. It would be nice, but maybe that’s just not how the universe works.

Saturday, November 02, 2019

Have we really measured gravitational waves?

A few days ago I met a friend on the subway. He tells me he’s been at a conference and someone asked if he knows me. He says yes, and immediately people start complaining about me. One guy, apparently, told him to slap me.

What were they complaining about, you want to know? Well, one complaint came from a particle physicist, who was clearly dismayed that I think building a bigger particle collider is not a good way to invest $40 billion dollars. But it was true when I said it the first time and it is still true: There are better things we can do with this amount money. (Such as, for example, make better climate predictions, which can be done for as “little” as 1 billion dollars.)

Back to my friend on the subway. He told me that besides the grumpy particle physicist there were also several gravitational wave people who have issues with what I have written about the supposed gravitational wave detections by the LIGO collaboration. Most of the time if people have issues with what I’m saying it’s because they do not understand what I’m saying to begin with. So with this video, I hope to clear the situation up.

Let me start with the most important point. I do not doubt that the gravitational wave detections are real. But. I spend a lot of time on science communication, and I know that many of you doubt that these detections are real. And, to be honest, I cannot blame you for this doubt. So here’s my issue. I think that the gravitational wave community is doing a crappy job justifying the expenses for their research. They give science a bad reputation. And I do not approve of this.

Before I go on, a quick reminder what gravitational waves are. Gravitational waves are periodic deformations of space and time. These deformations can happen because Einstein’s theory of general relativity tells us that space and time are not rigid, but react to the presence of matter. If you have some distribution of matter that curves space a lot, such as a pair of black holes orbiting one another, these will cause space-time to wobble and the wobbles carry energy away. That’s what gravitational waves are.

We have had indirect evidence for gravitational waves since the 1970s because you can measure how much energy a system loses through gravitational waves without directly measuring the gravitational waves. Hulse and Taylor did this by closely monitoring the orbiting frequency of a pulsar binary. If the system loses energy, the two stars get closer and they orbit faster around each other. The predictions for the emission of gravitational waves fit exactly on the observations. Hulse and Taylor got a Nobel prize for that in 1993.

For the direct detection of gravitational waves you have to measure the deformation of space and time that they cause. You can do this by using very sensitive interferometers. An interferometer bounces laser light back and forth in two orthogonal directions and then combines the light.

Light is a wave and depending on whether the crests of the waves from the two directions lie on top of each other or not, the resulting signal is strong – that’s constructive interference – or washed out – that’s destructive interference. Just what happens depends very sensitively on the distance that the light travels. So you can use changes in the strength of the interference pattern to figure out whether one of the directions of the interferometer was temporarily shorter or longer.

A question that I frequently get is how can this interferometer detect anything if both the light and the interferometer itself deform with space-time? Wouldn’t the effect cancel out? No, it does not cancel out, because the interferometer is not made of light. It’s made of massive particles and therefore reacts differently to a periodic deformation of space-time than light does. That’s why one can use light to find out that something happened for real. For more details, please check these papers.

The first direct detection of gravitational waves was made by the LIGO collaboration in September 2015. LIGO consists of two separate interferometers. They are both located in the United States, some thousand kilometers apart. Gravitational waves travel at the speed of light, so if one comes through, it should trigger both detectors with a small delay that comes from the time it takes the wave to travel from one detector to the other. Looking for a signal that appears almost simultaneously in the two detectors helps to identify the signal in the noise.

This first signal measured by LIGO looks like a textbook example of a gravitational wave signal from a merger of two black holes. It’s a periodic signal that increases in frequency and amplitude, as the two black holes get closer to each other and their orbiting period gets shorter. When the horizons of the two black holes merge, the signal is suddenly cut off. After this follows a brief period in which the newly formed larger black hole settles in a new state, called the ringdown. A Nobel Prize was awarded for this measurement in 2017. If you plot the frequency distribution over time, you get this banana. Here it's the upward bend that tells you that the frequency increases before dying off entirely.

Now, what’s the problem? The first problem is that no one seems to actually know where the curve in the famous LIGO plot came from. You would think it was obtained by a calculation, but members of the collaboration are on record saying it was “not found using analysis algorithms” but partly done “by eye” and “hand-tuned for pedagogical purposes.” Both the collaboration and the journal in which the paper was published have refused to comment. This, people, is highly inappropriate. We should not hand out Nobel Prizes if we don’t know how the predictions were fitted to the data.

The other problem is that so far we do not have a confirmation that the signals which LIGO detects are in fact of astrophysical origin, and not misidentified signals that originated on Earth. The way that you could show this is with a LIGO detection that matches electromagnetic signals, such as gamma ray bursts, measured by telescopes.

The collaboration had, so far, one opportunity for this, which was an event in August 2017. The problem with this event is that the announcement from the collaboration about their detection came after the announcement of the incoming gamma ray. Therefore, the LIGO detection does not count as a confirmed prediction, because it was not a prediction in the first place – it was a postdiction.

It seems to offend people in the collaboration tremendously if I say this, so let me be clear. I have no reason to think that something fishy went on, and I know why the original detection did not result in an automatic alert. But this isn’t the point. The point is that no one knows what happened before the official announcement besides members of the collaboration. We are waiting for an independent confirmation. This one missed the mark.

Since 2017, the two LIGO detectors have been joined by a third detector called Virgo, located in Italy. In their third run, which started in April this year, the LIGO/Virgo collaboration has issued alerts for 41 events. From these 41 alerts, 8 were later retracted. Of the remaining gravitational wave events, 10 look like they are either neutron star mergers, or mergers of a neutron star with a black hole. In these cases, there should also be electromagnetic radiation emitted which telescopes can see. For black hole mergers, one does not expect this to be the case.

However, no telescope has so far seen a signal that fits to any of the gravitational wave events. This may simply mean that the signals have been too weak for the telescopes to see them. But whatever the reason, the consequence is that we still do not know that what LIGO and Virgo see are actually signals from outer space.

You may ask isn’t it enough that they have a signal in their detector that looks like it could be caused by gravitational waves? Well, if this was the only thing that could trigger the detectors, yes. But that is not the case. The LIGO detectors have about 10-100 “glitches” per day. The glitches are bright and shiny signals but do not look like gravitational wave events. The cause of some of these glitches is known. The cause of other glitches not. LIGO uses a citizen science project to classify these glitches and has given them funky names like “Koi Fish” or “Blip.”

What this means is that they do not really know what their detector detects. They just throw away data that don’t look like they want it to look. This is not a good scientific procedure. Here is why.

Think of an animal. Let me guess, it’s... an elephant. Right? Right for you, right for you, not right for you? Hmm, that’s a glitch in the data, so you don’t count.

Does this prove that I am psychic? No, of course it doesn’t. Because selectively throwing away data that’s inconvenient is a bad idea. Goes for me, goes for LIGO too. At least that’s what you would think.

If we had an independent confirmation that the good-looking signal is really of astrophysical origin, this wouldn’t matter. But we don’t have that either. So that’s the situation in summary. The signals that LIGO and Virgo see are well explained by gravitational wave events. But we cannot be sure that these are actually signals coming from outer space and not some unknown terrestrial effect.

Let me finish by saying once again that personally I do not actually doubt these signals are caused by gravitational waves. But in science, it’s evidence that counts, not opinion.