The Universe is a Machine That Keeps Learning, Scientists say.

Well, this is unexpected. The universe is a learning algorithm that is teaching itself physics as it goes along.

The Universe Is a Machine That Keeps Learning, Scientists Say
https://news.yahoo.com/news/universe-machine-keeps-learning-scientists-200700798.html

Well, this is the stuff of madness, but I'm thrilled that someone is probing this idea of the universe teaching itself. I think that research into physics and the nature of the universe could help not only us, but the universe as well. So, yeah, it's expensive, but we could be shaping the universe to be a better place as we and it learn.

 

I never thought that the universe could learn. How is that possible?

Here is the Abstract:

We present an approach to cosmology in which the Universe learns its own physical laws. It does so by exploring a landscape of possible laws, which we express as a certain class of matrix models. We discover maps that put each of these matrix models in correspondence with both a gauge/gravity theory and a mathematical model of a learning machine, such as a deep recurrent, cyclic neural network. This establishes a correspondence between each solution of the physical theory and a run of a neural network. This correspondence is not an equivalence, partly because gauge theories emerge from N → ∞ limits of the matrix models, whereas the same limits of the neural networks used here are not well-defined. We discuss in detail what it means to say that learning takes place in autodidactic systems, where there is no supervision. We propose that if the neural network model can be said to learn without supervision, the same can be said for the corresponding physical theory.

Since the abstract is very difficult to understand I copied part of the Introduction.

It used to be thought that our standard model - including general relativity - is the low energy description of a unique consistent theory satisfying a short list of principles. But research in string theory, loop quantum gravity and other approaches to quantum gravity point to the opposite conclusion: there is a vast landscape of equally consistent theories [4–7]. What is then called for is a very different approach to the “Why these laws?” question: Let us seek a dynamical mechanism by which laws can evolve and change, and search within that setting for a reason why the present laws are more likely than another set. For example, the coupling constants might turn out to be dynamical variables. This opens the door to several kinds of explanation, new to fundamental physics, but well understood in other parts of science. These range from deterministic evolution in a space of couplings to stochastic forms of evolution on a landscape of theories. The possibility of a useful explanation for some properties of accepted laws, analogous to the explanatory power of natural selection in biology, becomes imaginable, and has been explored [4, 8, 9]. In biology there used to be a “Why these species?” problem: explain why dogs and cats exist while unicorns and werewolves do not. With the rise of the modern Darwinian perspective, it became clear that knowing the general principles, which apply to all biology, is a necessary prelude to understanding the detailed - often highly contingent and complex - stories that explain why a particular species likely emerged. It turns out that there are a number of puzzles concerning the values of the parameters of the standard model, which in one way or another indicate that their present values are special in that they lead to a universe far more complex than would be obtained with typical values [2, 3]. These suggest that explanations of the sort which we find in biology might be useful [4, 8]. The application of natural selection to cosmology was first proposed by the American philosopher, Charles Sanders Pierce, in 1893 [1]. The idea has been independently discovered by others since, and has been studied in some detail, to the point that several testable predictions have been recognized [4, 8, 9]. There is more that could be done in that direction, but that is not our purpose here. In this paper we want to go even further and suggest that the universe has the capacity to learn its laws. Learning is, as we will explain, a much more general notion than evolving by natural selection, it is also a more complex and demanding idea.

One of the authors is Lee Smolin, an expert on loop quantum gravity, who wrote a book about the trouble with physics. He has trouble with physics that is purely theoretical and doesn't rely on
experimental confirmation - like string theory. So, now he is involved in something that seems more down to earth?

 

It would be difficult for the uninitiated to understand the scientific paper. So, I'm leaving it to the news articles to tell the story.

Here's another take, Microsoft says that the Universe is a self-learning algorithm: Microsoft Researchers Claim Entire Universe Is a Machine-Learning Algorithm - Chaseimmortality

You should definitely read this article, it's a short read but explains some of the concepts in an easier-to-understand way.

 

This is Lee Smolin from a January 2015 interview with Scientific American talking about cosmological natural selection.

Horgan: Why hasn't the acceleration of universe—arguably the most important discovery in physics of the past 30 years--led to more theoretical advancement?

Smolin: At one level there is no problem, in that the acceleration of the universe’s expansion is easily described by adding a cosmological constant to Einstein’s equations, just as Einstein proposed in 1917. The problem is just with the value of that constant—it's ridiculously tiny. This is an extreme example of the basic problem that plagues the standard model of particle physics, which is that we don’t understand the reason for the value of any of the roughly 30 parameterize we need to write the laws of physics.

I am convinced that the answer to all these puzzles must be that these constants evolve, so the explanation for their values must be historical. Indeed cosmological natural selection gives a plausible explanation for the observed value of the cosmological constant.

Horgan: Some leading physicists, such as Tegmark, Susskind, and Greene, espouse multiverse theories plus the anthropic principle as a kind of final framework for cosmology. Comment?

Smolin: This is a sleigh of hand by which they hope to convert an explanatory failure into an explanatory success. If we don’t understand the values the fundamental constants take in our universe, just presume our universe is a member of an infinite and unobservable ensemble of universes each with randomly chosen parameters. Our universe has the values it does because those make it hospitable to life.

There is so much wrong with this as a scientific hypothesis. As I have explained in detail in three of my books and several papers, it is hard to see how it could make any falsifiable predictions for doable experiments. Claims to the contrary are fallacious, as I and others have explained in detail. I won’t impose the details on your readers but just mention that these criticisms have not been answered.

What we have to do is to propose mechanisms by which the laws and constants may have evolved which imply falsifiable predictions by which they can be checked. I have proposed two: cosmological natural selection and the principle of precedence.

Horgan: You suggest above, and in your new book, that the laws of nature "evolve." Won't that hypothesis make physics and cosmology even more flexible and hence less falsifiable?

Smolin: No, the key lesson of cosmological natural selection (CNS) is that it makes falsifiable predictions for real observations. In fact the predictions I published in 1992 have held up. To mention one: there can be no neutron stars heavier than twice the mass of the sun. Current limits come close; the heaviest well-measured neutron star is at 1.9 solar masses, but so far none go over.

At first I had the same intuition your question expresses. But it's wrong; making laws evolve increases the falsifiability of science because it increases the number of hypotheses that can be checked because they imply falsifiable predictions. The reason is that the additional hypotheses concern the processes by which evolution took place. Since these processes would have taken place in the past they imply predictions which are checkable by real observations. This point is discussed in detail in my books Life of the Cosmos and Time Reborn.

One way to reconcile evolving laws with falsifiability is by paying attention to large hierarchies of time scales. The evolution of laws can be slow in present conditions, or only occur during extreme conditions that are infrequent. On much shorter time scales and far from extreme conditions, the laws can be assumed to be unchanging.

As Roberto Mangabeira Unger and I argue in our new book The Singular Universe, the most important discovery cosmologists have made is that the universe has a history. We argue this has to be extended to the laws themselves. Biology became science when the question switched from listing the species to the dynamical question of how species evolve. Fundamental physics and cosmology have to transform themselves from a search for timeless laws and symmetries to the investigation of hypotheses about how laws evolve.

 

I have had a similar impression of what is going on with life on earth and "evolution". When one looks at all the incredible variations of plants and animals, especially in the long history of the planet, it has the appearance of trying just about everything to see what works. Keep in mind, I said "appearance"; I am not saying that is how things function or even that I believe it is. One wonders, though. The metaphor would be that there is some "mind" behind it all that jiggles this and adjusts that to get another result. I don't understand how that could be, yet it also is difficult to believe what has actually developed. We simply cannot tell at this point, and it is unfortunate that so many people feel compelled to jump to conclusions and slap nouns, names and definitions on the so far unexplained.

 

One thing to consider is whether or not we create a whole other universe when we run machine learning algorithms.

Another thing, is the Universe a result of an AI computer run amuck? Is this the danger we've been fearing from unlimited AI growth? Because if it's steering us towards life-favorable conditions, then we might not need to be afraid of AI's becoming malevolent. Maybe???

 

Galaxies are just neurons in some greater being. Or not. Who knows. Lol. Super fascinating stuff though.

 

It seems to me that scientists and mathematicians are stumped and have to find a solution to their (mankind's) stupidity. I'm not even going to consider the idea of a concious universe. It is man that is concious and animals but not "things". Your mind is the universe...

 

You're describing evolution.

There doesn't have to be a "mind" behind it, as there are success criteria that exist all around us. So, those changes that allow procreational success, survival, etc., can win while others tend not to win. And, there are numerous stragegies for facing those challenges - brains, speed, strength, skills like flight or digging underground, ability to hide, ability to reproduce prolifically, ability to take advantaged of different food sources, ability to survive in places others can't, etc. With the multitude of possible strategies the appearance you mention shouldn't be surprising.

 

The AI we create is somewhat different than the primary questions of thoretical physics, I think.

We create AIs. So, there are meaningful actions we can take with respect to that.

 

Wow!!!!!

I have definitely got to look into this idea!!!!

 

I would suggest that any thinking universe is not interested in creating conditions for life. As far a we know, this planet is one of the very few where that happens and that it is by far more common to see no life or possibility of one. If this thinking universe wanted to go for a "life solution" it has surely made a LOT of mistakes and doesn't learn very well.

This does indeed sound like evolution, albeit on a grand scale. ie if it works, then it remains. If it doesn't work, it is recycled into something else. That is evident everywhere in our observable universe.

 

I got into this topic so often with my Atheist and Agnostic friends that I actually wrote up a blog that goes into this idea.

www.CarbonBias.blogspot.ca/

I am of the belief that although things look terrible at this time we have reason for rather extreme optimism for the long term fate of humanity and for animals as well.

 

Well, the long term fate of humanity on Earth seems sealed by what we know about stars such as our sun, which will expand to incinerate this planet.

 

I am very, very suspicious of anything that attempts to anthropomorphize objects that seem unlikely to warrant the assertion.

 

AI and the Universal Mind (if there is one) are not equivalent. Fear of AI 's potential for malice towards humanity or life in general is justified, but the UM encompasses us and more. Tom Campbell refers to the UM as the larger consciousness system.

 

The universe does not necessarily need an ideal life solution, at least not in all places where life may exist. The physical mistakes that seem so significant to you are not any final obstacle to the progressive trajectories involved, as evident in the so-called march to extinction, the fate of so many species in the long history of life on this planet. Increases in intelligence and consciousness have arrived despite the mistakes and the flaws in the biological vehicles in which they reside. This place is a factory for soul development. That's the real nitty gritty; the physical imperfections are secondary to the overall goals.

 

And why is life the golden ticket?

 

I donno. I think A.I. is way over rated. When my ipad has a decent spell check, call me.

Regarding space/time and existance, I think it follows an existence path we don’t have the tools to decipher because the tools themselves are transitory.

 

So if I understand this, the laws of maths and physics are not permanent?

 

The problem with spell check is that it is FREE!! How much would you pay for better spell checking?

It's approaching 25 years since humans could beat computers at chess.

Now, humans can not win at the game of go - a more difficult game than chess.

The way this was done was to create a learning machine and then let it play millions of games with itself, allowing it to learn how to win. The result is that experts at the game of go say they don't even understand the strategy the computer is using to beat them.

I think the interesting point here is that humans didn't even attempt to teach the computer how to win. They let the computer figure that out.

Today, ai is being used in chemistry and other fields, where complex geometry is important. AI is solving problems in minutes that it used to take humans years to solve.

 

Have you heard of Grammarly? Now there's an AI (I think), that spell checks. How do I say that without sounding like a plug?

It's not free, except the basic version, but I use it for my spell check on my computer at home.

 

I've heard of Grammarly.

I think this is one of the examples where what we get for free isn't always the cutting edge technology.

 

I can't even picture in my imagination a universe that is learning. I have always thought that things are the way they are because that is the way chance works. But a thinking universe? With relativity, how would that work?

 

Good points!

I would have to ask whomever is a proponent of the universe "learning" what it is that they think was "learned" - or maybe even just what they think "learning" is.

Relativity is a limitation even just considering that information (whatever was "learned" somewhere) can not be communicated to the entire universe.

I really wonder what the universe would see as needing to be solved. Surely "learning" comes from having some sort of objective and some sort of evaluation system for determining whether what was "learned" is actually a step forward.