In a remarkable article entitled Contemplating the End of Physics posted today at Quanta magazine, Robbert Dijkgraaf (the director of the IAS) more or less announces the arrival of the scenario that John Horgan predicted for physics back in 1996. Horgan argued that physics was reaching the end of its ability to progress by finding new fundamental laws. Research trying to find new fundamental constituents of the universe and new laws governing them was destined to reach an endpoint where no more progress was possible. This is pretty much how Dijkgraaf now sees the field going forward:
Confronted with the endless number of physical systems we could fabricate out of the currently known fundamental pieces of the universe, I begin to imagine an upside-down view of physics. Instead of studying a natural phenomenon, and subsequently discovering a law of nature, one could first design a new law and then reverse engineer a system that actually displays the phenomena described by the law. For example, physics has moved far beyond the simple phases of matter of high school courses — solid, liquid, gas. Many potential “exotic” phases, made possible by the bizarre consequences of quantum mechanics, have been cataloged in theoretical explorations, and we can now start realizing these possibilities in the lab with specially designed materials.
All of this is part of a much larger shift in the very scope of science, from studying what is to what could be. In the 20th century, scientists sought out the building blocks of reality: the molecules, atoms and elementary particles out of which all matter is made; the cells, proteins and genes that make life possible; the bits, algorithms and networks that form the foundation of information and intelligence, both human and artificial. This century, instead, we will begin to explore all there is to be made with these building blocks.
In brief, as far as physics goes, elementary particle physics is over, from now on it’s pretty much just going to be condensed matter physics, where there at least is an infinity of potential effective field theory models to play with.
Dijkgraaf ends with an argument indicating that human intelligence is outmoded, artificial intelligence is our future:
Science concerns all phenomena, including the ones created in our laboratories and in our heads. Once we are fully aware of this grander scope, a different image of the research enterprise emerges. Now, finally, the ship of science is leaving the safe inland waterways carved by nature, and is heading for the open ocean, exploring a brave new world with “artificial” materials, organisms, brains and perhaps even a better version of ourselves.
Along the same lines, today also brings an article in the New York Times by Dennis Overbye, Can a Computer Devise a Theory of Everything? The article discusses the new MIT Institute for Artificial Intelligence and Fundamental Interactions and Max Tegmark’s hopes that AI will “discover all kinds of new laws of physics”. My guess is that this will work just fine if you give up on the 20th century understanding of what a “law of physics” is and follow Dijkgraaf’s lead. The problem then may be not so much “will we understand the new laws of physics found by AI?”, but rather that of them not being interesting enough to be worth understanding…
Update: To clarify the point I was trying to make about the Dijkgraaf piece arguing against the “end of physics”, compare it to the similar 1996 piece Gross and Witten published in the Wall Street Journal (a summary is here, an extract here). Gross and Witten were strongly disagreeing with Horgan, whereas it seems to me that Dijkgraaf implicitly agrees with Horgan that fundamental physics has hit a wall and theorists are moving on to do something else.
How ironic. Not two hours ago I decided on a Miltonish title for my next blog: Paradigms Lost.
Huh, I didn’t realize we understood the laws followed by dark energy or dark matter. My bad.
To add a bit more detail, Dijkgraaf starts by asking “Is physics finished?,” he answers that it is not, and he gives three reasons. His third reason is the one discussed in this blog post (condensed matter is a basically endless subject). His second reason though, is that we don’t understand dark energy and dark matter and physics can’t be finished without understanding them.
We are not reaching the end of physics, we are reaching the end of affordable physics. 100 years ago one could physics with one person, on a table top with extra time, and a few dollars. Now it costs a discernable amount of GDP from many countries and teams of 1000’s. We have found deeper layers of reality much faster than the growth of the economy by pooling money, resources and people. That can only go so far and we are reaching that limit. Progress in physics will inevitably slow down to the growth rate of the economy unless something revolutionary occurs like a miracle in AI. It is not the end of physics, it is just much slower.
Peter wrote:
David Appell wrote:
These sarcastic comments seem to overlook what Dijkgraaf wrote:
So, I don’t think he’s saying fundamental physics is done. Indeed his comments seem pretty well-balanced to me. The title, “Contemplating the End of Physics”, is overblown and misleading, as often for articles in pop-sci magazines.
Pardon me if I’ve just become too cynical, but I read Dijkgraaf’s Quanta article as just another PR piece, similar to those we have come to expect from senior management at CERN. (One clue: the bizarre depiction of phonons as somehow relevant, or even an accurate description of “sound” much less elementary quantum physics.) It’s another justification for why, despite diminishing returns and increasing irrelevance for much of fundamental physics’ “scientific” output, places like IAS (and, in the future, CERN) should continue to exist and be funded.
But it gets worse: If you just ignore the gaping holes in our current formulation of the 5% of the universe we’re supposed to now “understand”, that have been unable to be closed over that last half century, and concentrate instead on just re-arranging various pieces into ever more bizarre and intricate patterns, and call that a “search for knowledge”, then you essentially remove any chance of being objectively critiqued or judged as having been successful (or not). While I doubt that these attempts to put a “positive spin” on the situation [pun intended] are a pre-planned subterfuge by people like the directors of these agencies, that is the net result. While, as Peter has previously said, just combining a bunch of ingredients and essentially throwing the resulting mess against the wall to see what sticks and what doesn’t could hopefully someday result in something useful, no one wants to admit that’s what they/we are doing. Especially not to the general public, nor to funding agencies.
I don’t think physics is ending, but there is a serious possibility of a decline of human beings able to do it. We think about science as an endless progress, but history teaches us that there were dark ages where science was destroyed (e.g. the end of Hellenic science). Today, Big Science is the real threat for physics. I wrote some notes in a Section of my book Scienza e Linguaggio (Aracne, Roma, 2015; in Italian); the English translation can be found here: http://www.brera.inaf.it/utenti/foschini/Foschini_translation.pdf
We’re in the late stages of the latest AI hype cycle: this one is based on a fairly old idea which has only recently become practical, which is throwing deeply astonishing amounts of training data at machine learning systems. As with previous hype cycles, impressive early results lead to wild claims about artificial general intelligence coming in a few years usually followed by some kind of exponential runaway in intelligence. As with previous cycles it is not coincidence that many (not all) of the people making these claims are seeking funding for AI work. As with previous hype cycles the impressive early results will be followed by less impressive later ones as the tricks hit a wall – once you have trained your language model on all the natural language that there is and it still doesn’t produce very good natural language, where do you go? And what does it tell you about how intelligent systems learn language (hint: nothing)? And so the bubble of hype will burst, a new AI winter will follow, people will look back on the wild claims as being rather silly, until eventually they forget and the cycle repeats.
Well, perhaps it will repeat. Previous AI hype cycles have been both started and stopped by various exponential processes: available computing power increased exponentially with time, but the computing requirements of AI increased either exponentially with a shorter time constant, or worse than exponentially. But computing power certainly can not increase exponentially for ever: its increase is bounded above by some constant power of time (which is probably effectively 2). We’re perhaps some way from the exponential increase hitting the physical limits, but probably not tens of years away.
The good part of all this is that, as with the previous hype cycles, good applications will be found. My favourite is weather forecasting: there is a vast torrent of data and tomorrow’s initial conditions are the data against which you can train today’s run.
But what isn’t going to happen is some artificial general intelligence deriving new and interesting laws of physics. That’s just what that the people hyping AI need us to believe so they keep getting funded.
A child discovered to be a mathematical prodigy may be directed from that point onward by establishment figures into doing research intended to advance ideas supported by the majority. In acquiescing the child will likely lose all chance of ever fulfilling his/her potential – of ever doing anything truly original.
Likewise, an AI might well be a computational prodigy, but it too will be nudged into areas supported by those feeding its algorithms data and methodologies. However, unlike the child prodigy, the AI will never have the potential to shatter paradigms, or to conceive something truly original. Indeed, it can not conceive at all.
Yes, we know that 95% of the universe consists of dark stuff, but this is “knowing” in the sense that renaissance astronomers knew that the universe is governed by 13 epicycles. Within the framework of epicycle theory, experiments implied that there were at least 13 of them, and within the framework of GR+QFT, experiments show that 95% of the universe is dark. Both GR and QFT have of course been tested to much higher accuracy than epicycle theory ever was, but only in domains where the other can be ignored, and we know that they appear to be mutually inconsistent. In particular, the naïve quantum calculation of the CC is off by 120 orders of magnitude, which may indicate that GR+QFT might fail in situations where the CC is important.
Dijkgraaf does say he has three arguments against physics being finished. Most of his article is about the third argument, which is what I discussed. For the first of his arguments, he himself gives the counterargument.
The second is the usual “we don’t understand dark energy/matter, and it’s most of the universe” argument. Everyone who makes this argument likes to cite the relative abundance, but I don’t see how that matters for the question of how large a hole in our understanding these things are.
On dark energy, it appears there’s a single number that describes it, one of 30 or so numbers in our best theory which we don’t understand. The issue of understanding this particular number has been dead in the water for a long time: I know of no viable idea for calculating it. The string theory landscape people will tell you it is inherently uncalculable, it’s an environmental artifact of the baby universe we happen to be in.
Dark matter is a much more complicated phenomenon, and it has received a huge amount of attention from theorists over several decades. All the relevant data is astrophysical, and I’m willing to believe astronomers in coming years will get us more. But, it seems to be very tough to move beyond “there’s something out there gravitating, but we don’t know what it is”. Maybe it’s right-handed neutrinos, or some other sort of unknown fundamental particle whose interactions are purely gravitational. But maybe it’s instead something we don’t understand about astrophysicsics, e.g. primordial black holes. Sure, this is an open problem, fundamental physics is not finished. I don’t think Horgan’s claim was that no open problems would remain, it was that progress towards solving them would stop.
I mean, you could criticize Dijkgraaf’s piece for being anodyne—yes, *of course* fundamental science has been gradually shifting its emphasis for decades from “what is” to “what could be,” that’s why so many of us now work on subjects like quantum information or topological matter or synthetic biology, welcome to the club! But like John Baez, I find it hard to argue that Dijkgraaf’s thesis is *wrong.*
Indeed, let me propose a corollary: if (contrary to John Horgan) fundamental science does continue indefinitely, then it will be for basically the same reason why *pure math* continues indefinitely—namely, because the relevant limits turn out to be only those of possibility-space, not of physical space.
Scott,
I’m not at all arguing that Dijkgraaf’s thesis (or that of John Horgan from 1996) is “wrong”. Sticking to the case of fundamental physics, I think Horgan was correct in pointing out that research was hitting limits, and that fruitless speculation like string theory was evidence of this. Lots of physicists were outraged by Horgan’s argument, for an example of the reaction at the time, see a Wall Street Journal editorial by Gross and Witten
https://www.wsj.com/articles/SB837118256541339000
They argued that new revolutions were on the way, with string theory and supersymmetry promising ideas about to be tested at the Tevatron and LEP.
What I think is remarkable is the way Dijkgraaf now seems to be agreeing with Horgan. That much of the focus of theoretical physics at Princeton for many decades has been a huge failure is not something Dijkgraaf is going to say, but it’s implicit in his not mentioning it. His job is to generate enthusiasm for theoretical physics and to get funding agencies and wealthy people to pay theoretical physicists, so he’s putting the best face possible on what is happening.
My own agenda is somewhat different, so I’d rather see a clear-eyed examination of what went wrong and of what possible ways forward remain for those who don’t want to give up and do something else. Maybe one day we’ll see that in Quanta also.
Peter, thanks for kicking off this discussion of the themes my old book. I found Dijkgraaf’s essay to be, as you and others suggest, mere marketing, and not very original at that. Michio Kaku and others were saying in the 90s that learning the laws of nature is like learning the laws of chess. The game is just beginning! And so on. I addressed that wan hope in 1996, as well as the hope that super intelligent AI will jumpstart a new era of physics, which Hawking proposed in the early 80s. I too, would love to see a “clear-eyed examination” of the status of physics, but that’s asking a lot from people who have so much invested in positive outcomes.
The point of view of the Dijkgraaf article is not that unusual: essentially it is what the condensed matter physicists have been saying for decades. The only unusual thing is that a string theorist is saying it.
There is no very mild way to put it, Dijkgraaf’s view is pathetic. I’m happy to have my tax money going into basic science and avant-garde engineering, but not at all to engineering that poses as basic science.
tulpoeid: So condensed matter theory is just engineering and not basic science?
There is a ridiculous amount of real science related to condensed matter theory that we don’t understand, and one approach to trying to improve our understanding of it is to build specially designed materials.
Is making new chemicals just “engineering that poses as basic science” if you’re doing it to better understand the laws of chemistry, and not for a specific application? (Maybe some research directions that involve making new chemicals are not worth funding, but I would hope that the funding agencies would decide not to fund them; the same should be true of new materials in condensed matter physics.)
A minor comment, maybe not even worth making, is that you can do fundamental physics with condensed matter. For example, it is possible to imagine an alternate version of history where the leap from classical to quantum physics was made entirely via condensed matter experiments, basically because hbar has units of (length) * (momentum), and so you don’t need to go to short distances to find it– you can see quantum effects on large scales by going to low momenta (or low temperatures). And so if there’s another layer of physics out there beyond quantum mechanics, and/or another basic constant like hbar, it may well show up via condensed matter experiments. (Of course, the timescale for such a basic discovery could well be centuries, which is why this comment is a bit hollow.)
I am just seconding things already said here, maybe with a different emphasis. The heart of the controversial “third argument” of the piece is this:
“The aim of physics is to understand in a precise, mathematical way all manifestation of matter and energy in the universe — and we have barely started to explore this infinitude of possibilities. Claiming that physics is finished is akin to arguing that mathematics ended after the introduction of natural numbers and basic arithmetic, or that chemistry was over with the advent of the periodic table. Learning the rules of chess doesn’t make you a grandmaster.
The truth is, the realm of the smallest particles is not the only place you can find the fundamental laws of physics. They can also “emerge” out of the collective behavior of many constituents.”
There is a name for the discipline devoted to to studying how systems can be constructed using known or given fundamental laws so that it behaves in a certain way. That name is “engineering”. That can be as hard, and insightful, and interesting as any scientific discipline. Quantum information theory and quantum computation, in this sense, are obviously engineering. They ask the question: *Given* that the constituents of the system are governed by the laws of quantum mechanics, how can systems that transmit or process information be built?”. That is an extremely interesting and hard question.
The only issue seems to be about terminology. No one would doubt that quantum information and computation are disciplines that can properly be called “physics”. No one should doubt that they also can be properly called “engineering”: the terms are not mutually exclusive even if the academic world creates different “schools” for each. If some physicists had not spent a lot of time denigrating engineering (cf. The Big Bang Theory) that observation would go down easier. The last ditch of the fight is then over the term “fundamental”. In the usual sense in which that term is used, engineering and, say, condensed matter and solid state physics are not “fundamental”, and the principles they employ are not “fundamental”: they are both “emergent” and often just reliable generalizations that can, physically, be violated (like the “Second Law” of Thermodynamics which in a clear sense in not a physical “law” at all since it can be violated by a physical system).
There are some people (like me) who are largely interested in fundamental physics, fundamental ontology and physical law (properly speaking). That is just a sort of intellectual preference or matter of taste. Some of us think that the whole project of trying to figure out these fundamentals went bad, and the badness goes back to Bohr and Heisenberg. We wonder whether progress has slowed not merely because the issues are harder in various ways but because the whole intellectual approach has been corrupted. The flourishing of the non-fundamental disciplines really does not help us in our particular arena of interests, but it is certainly a thing to be celebrated.
To add to anon’s remarks above: you can measure $\hbar$ with photocells or light-emitting diodes.
Hi Peter,
Why is it you think potential future physics discoveries by AI could be not “interesting enough to be worth understanding”?
Is it because one can envision the AI being programmed to think based on the same ways theoretical physicists do and thus will further engage in the kinds of stagnation we’ve seen with things like string theory etc?
I find it plausible that the AI could eventually think very differently to the way humans do and thus circumvent our limitations.
My take on Dijkgraaf’s article is that no, physics is not finished, but for much better reasons he has been able to expose. He is even borrowing the success of other fields, like cosmology, to hide the lack of success of his own field (HEP and theoretical physics) in the first place.
-Not because the first two decades of this century have been pretty successful for physics, which they have not, even if the Higgs, gravitational waves and the first image of a black hole’s event horizon have been discovered or taken.
-Not because 95% of the universe is missing. This is just a statistical fact or a teaser, not a reason. Maybe it is missing because we do not have the correct theory of quantum gravity. Or even the correct theory of quantum electrodynamics.
-Not because any authority claimed that natural numbers or the periodic table had put an end to maths or chemistry, or at least not as explicitly as those who claim now that progress in HEP and theoretical physics has come to an end because we cannot afford more expensive particle accelerators. They might help, but nothing else.
-Not because the behavior of phonons is similar to the behavior of photons, because it is only very vaguely similar.
-And certainly not because the fundamental laws of physics, which are quantum laws, can emerge out of the collective behavior of many constituents. We already know that the laws of the macroscopic world are quite different from the laws of the subatomic world.
The rest of the arguments are even weaker imho.
No, it is because, despite the lack of support or further experimental evidence, some researcher or group of researchers will be able to make their way through it. And because this effort is already underway.
David,
All evidence I’ve seen is that the only ideas theorists have about using “AI” in theory are ideas for how to automate dumb things people are doing so they can be done more quickly and on a larger scale. Large scale dumb is still dumb.
I see how AI can have a role when there’s a lot of relevant data that a human mind is going to have trouble making sense of. The basic problem of current fundamental theory though is that there is virtually no relevant data (i.e. data that disagrees with our best model).
As John Horgan points out, already back in the 1980s people were talking about this. Nothing ever came of it, I don’t see any reason to believe that situation will change anytime soon.
On the other hand, invoking computers has historically been a good way for theorists to try and get money. That has often worked, is working now, and will work in the future.
To me the question of whether condensed matter physics is “fundamental” or “engineering” is infinitely less interesting than… condensed matter physics!
You can make a supersolid where holes in a crystal – the absence of electrons – form a Bose-Einstein condensate and flow like ghosts through the crystal lattice. You can make “liquid light” using exciton-polaritons, which are a blend of photons and electron-hole pairs. You can make a Wigner crystal: a 2-dimensional periodic structure of repelling electrons. And it goes on and on: we’re in the golden age of new forms of matter! A physicist would need to have a really sour disposition to remain unexcited by these things.
John Baez: There is a sense in which I agree with your comment, but it also illustrates the sense in which the fundamental is, well, *fundamental* in a way that creates an asymmetrical dependence of understanding. It certainly sounds quite amazing that the *absence* of an electron, or a *hole*, can behave like the *presence* of an electron or a *particle*. And one would like to understand that better. But what claim is even being made cannot really be grasped without first understanding what an electron *is* and what are the physical conditions that make it accurate to say there is one *present* and to ascribe to it a location. When reading your comment, one reflexively falls back on a Democritian picture of electrons as sharply located particles, maybe even point particles. And maybe that is even correct! But most physicists would flatly deny that that is the right way to understand what an electron is. So this is a nice illustration of how unclarity about the foundations inevitably flows upward to a sort of unclarity about what is built on, or constructed out of, the foundations, while there is no obvious necessity of an analogous retro-transmission of lack of understanding downward. That’s part of what draws some of us to the fundamental questions.
” why *pure math* continues indefinitely”
Scott, I’m not sure this is really the case. A long time ago, people (Wigner and probably others) used to say that “God is a mathematician”. But of course the dual formulation of that statement is that “Physics is divine mathematics”. Not necessarily because divine math is better than mundane math, but because it is the kind of math that God the mathematician does and publishes in Nature (the bitch, not the journal).
But then again, some of us believe that divine math really is better than mundane math. So if fundamental physics runs out of steam, so does divine math.
The conversations are good indicators of where these elder statesmen of the field feel we are going. But they are not the only voices we should listen to.
Physics will end when the brightest young people no longer want to pursue it. I have not seen this. In fact, even during the pandemic there are many undergraduate and graduate students wanting to work in particle physics. Generally speaking these are top tier students who could probably join any group in the department.
Peter Orland and anon., historically in the way quantum mechanics was developed statistical mechanics played a huge role, say in the story of the black body radiation or in most of Einstein’s contribution to the subject. It could be an illustration of why often what is “fundamental”, to the extent that this is even an interesting question at all, can be only decided after the fact.
Every so often, human beings like to believe that their century is special, and its impressive new tools and advances will take us to some sort of final clarity. Sounds a lot like Constantin’s world-view. It is good to remember that today’s queries which are firmly in the realm of science were viewed not so long ago as pertaining to the sacred.
The most striking aspect of universe-view today is the lack of paradoxes and contradictions with experiment! As noted by Mark Twain, history does not repeat itself, but it sure rhymes; science is not so different. I wonder how people felt after Newton’s death where the explanations of all phenomena led to his world-view. The challenge came not from the mind of great people but from the study of electricity. Will the study of dark matter provide the clues to go beyond?
We live in the post-Einstein era, a time of great synthesis that left us with the General Theory of Relativity and Quantum Mechanics, neither challenged by thought nor experiment. Yet both seem incomplete in different way. What price compatibility?
The spectacular Standard Model explains most forces in the early universe; but it still is a fragmented view of matter with many loose parts.
As Brownian motion nailed the molecular theory of bulk matter, will the Cosmological constant be a pointer to the underlying theory of space-time?
So many questions, so little time …
Read Einstein’s paper on the photoelectric effect. There is a wonderful discussion of in volume II (1963) of “The Natural Philosopher” by the late Martin J. Klein. That is what sublime physics is like!
Voltaire’s “Lettres Philosophiques” (letters 14,15,16) discuss the world views of Descartes and Newton.
Peter Shor:
If something is application of science then, yes, it is engineering.
Examples of condensed matter applications specifically with the goal of advancing “basic” knowledge would be useful, though.
In any case, I can read this passage only as trying to sell engineering as basic research:
“All of this is part of a much larger shift in the very scope of science, from studying what is to what could be. In the 20th century, scientists sought out the building blocks of reality… This century, instead, we will begin to explore all there is to be made with these building blocks.”
What Dijkgraaf seems to be saying is that engineering will replace and supersede physics. As an engineer, this is music to my ears, which is precisely why I do not believe Dijkgraaf. If it sounds too good to be true, then it probably is. Without physics, there will be no engineering. Period.