20 Years Later
math.columbia.edu
math.columbia.edu
The solution is never for the individuals to induce change. This doesn't work for consumers (i.e. you can't say "If everyone stops buying plastic bags, we do something for the environment", because well, nobody cares... You have to make selling plastic bags either illegal or add a significant surcharge that causes consumers to avoid them for reasons that actually affect them, like "too expensive")
And it doesn't work for anything else either. As long as there is funding, this will go on. The "solution" is to stop funding, but apparently since funding is still flowing, someone still values the research. So what is even the problem?
No progress has been made? Well, if funding is still there, whoever funds this seems to be beyond happy with the progress. This is a non-issue.
Not that that’s a good or bad thing. You wouldn’t want a “non-expert” to evaluate the proposal. And perhaps a tapering of the funding is better indicator of interest/progress of the field?
This is very close to "There must be a pony in there somewhere". [1] If the question is whether the field is worth continuing to fund, the answer can't be, "Well since people are still funding it, it must be worth funding." By the same logic, we should just keep putting billions into WeWork.
Research is hard to value, and fundamental research especially so. The logic of "if people buy it, it must be good" only patchily applies to normal commerce, where results are relatively easy to measure and feedback loops are short. It's wholly inadequate for feedback loops on the scale of decades.
[1] https://www.google.com/search?channel=fs&client=ubuntu&q=the...
I read you comment as another anecdote towards the general belief that higher education requires a refactor for many parties and the general growth of knowledge creation.
Lose your job as a foreman at a factory, and you can switch to being a manager at Mc D.
Automation offers more of this, and it’s essentially going to go feed the owners of that automation.
This is all optimization of resources and ownership of course so it’s essentially a question of how the share of the new efficiency derived surplus is split.
Part of which could go to things like keeping theoretical research programs alive.
I mean these are physics and mathematical specialists.
I’d rather have a country that can keep such people engaged and active than lose them.
I mean what other options do you see? They could join hedge funds. Open an ML based startup.
In the end someone else has to work on this material- the problem isn’t going to go away.
And it sounds like this requires tools to do measurements of astral phenomena. Considering that America lost its telescope recently, and further investment is unlikely, I don’t think the investment in ancillary fields is high either.
Moving about a bit is not losing people, and if they're not providing something useful then they've already been lost anyway, and the opportunity cost of what they might be doing usefully elsewhere is also spent. May as well employ them at McDonalds, except they'd probably be rubbish at it.
An economist visits Communist China and they proudly show him a railway being built. He notices the workers are using shovels to dig the holes, instead of construction machines.
- Why don't you use construction machines?
- We care a lot about the workers here, using construction machines would make some of them unemployed!
- Well, in that case, I suggest you make them use spoons instead.
Though how big institutes work i do not know.
Or you’ll dump the theoretical physics and spend on hair growth products and techniques.
The market doesn’t care for the long term- it’s an optimization system for competing needs, often with immediate needs overwhelming long term.
Markets also use regulatory forces to ensure that everyone isn’t mainlining addictive chemicals or dis-incentives are created for not wearing seatbelts.
Universities and higher learning isn’t and shouldn’t be a market. I’d argue that one of the reasons for the sad state of academia today is that too many MBA are running everything, including those things that aren’t businesses and shouldn’t be run like such. If researchers didn’t need to always chase after the next grant, a lot of the shortsightedness and lock in into specialities wouldn’t even be a problem in the first place.
It says something about how thorough neoliberalism won that few people react today when, again, it is taken for granted that everything is a market and what the market decides is right.
As a professor it is simply extremely unnatural to just drop your main area of research.
First, your academic network will almost entirely stem from the subfield you have been working on. This affects the expectations of your PhD students, the recruitment of postdocs (who are often swapped between colleagues in the subfield), and the conferences you visit -- where you will see all the familiar faces of colleagues you have known for years.
Second, your mind has been shaped significantly by years of working in your own little subfield. It is easy to come up with ten projects in your area, rank them by difficulty as well as the interest they would generate amongst your colleagues, and help your students to execute them. It is significantly harder to do anything like that outside your subfield; for one, you will not have the mentoring that a typical PhD student has.
Sounds like a textbook groupthink environment.
No one is going to lead a scientific revolution in that environment.
String theory doesn't actually do anything though, so there's no way to say it's better or worse than anything else. It's like someone saying they're working on solving the halting problem or something and eventually they'll get there and they've been working on it for 20 years without any code that actually works. At some point, you just have to give up and do something else. They keep getting grants to work on this stuff though for some reason. Someone needs to interview the people who are giving out string theory grants and ask them why the heck they are still giving these people money.
Imaginably, given the field's topics, they think in different dimensions than on a 20 year scale.
The one thing that kind of irks me is this Kaku book “The God Equation.” I have only seen one equation that is so universally applicable that it could deserve that title, and even then I would be hesitant about that because it might give people the wrong impression. (It is the transport equation—it keeps appearing and appearing, a bunch of other equations are special cases of it, it is involved in one of the million-dollar Clay Mathematics prizes so there is clearly something hard/intractable about it, and it has a term which refers to creation and destruction. It says, a box flows downstream, the time rate of change of stuff in the box is equal to the flow J of stuff through the walls of the box plus the rate Φ of stuff being created/destroyed in the fluid. Or, ∂ρ/∂t + (v · ∇) ρ = - ∇ · J + Φ.)
if your font support it, you could write any equation you want
Do you know any source(book or otherwise) that continues to build on top of topics taught in a Regression and Linear algebra course(such as Multiple Linear, Weighted Least square, Logistic, generalized least squares, principle component regression, singular value decomposition) and slowly move towards the current state deep learning?
No.. learning how to prove it will serve you well in the future. All deep learning is is figuring out different ways to combine all of this prior art inside of a computation graph.
Step N is linear regression, SVD etc... step N+1 is deep learning.
It would definitely be very hard to do any meaningful research from this position, but I know enough to be useful (er, or dangerous) and can read papers and code to keep up with recent advancements, and try things out like different convolution designs, layer designs, gates, functions, feedback, etc.
(On the topic of reading: Karpathy[1] and Colah[2]'s posts have a wealth of introductory conceptual information and images in them, and helped our team discussions a lot while we've learned.
The math and proofs. The algorithms. The software tools/frameworks Building NNs with those tools.
All of those have different levels of conceptualization and utility.
As some extreme examples: Anyone can build an iPhone image classifier drag and drop today. That’s one level. Developing an alternate to backprop is another level.
It isn't quite state of the art, and it is very light in terms of how to evaluate and deliver the solutions. But I thought it was good. My team at work has gone through it selecting one chapter at a time this year in our book club - it's provided a good catch up and share forum for us. Some of the folks in the team are Engineers and so haven't a strong ML background, a couple are Data Scientists and therefore were in the same place as you (+/- some use of deep learning in a few engagements) and a couple of the folks are diehard ML researchers, but focused on different things (time series, NLP, evaluation) so there was a lot of crossover for everyone.
Deep learning might be a good example. The underlying neural networks has been a “dormant science” for decades until its breakthrough.
I wouldn’t be so sure either that image feature processing won’t get another cite. More likely they will show up in preprocessing again as the boundaries of deep learning get pushed.
So no, I would not expect the currently known engineered image features to show up in preprocessing again - they simply do not add any value whatsoever (which can be experimentally verified) over the "learned preprocessing features". Perhaps someone will engineer a new substantially different type of feature that can't be currently learned from data, and that would get used and cited - but it won't bring citations to the old engineered features.
I don't work on OCR much (and when I did, it was for book digitization which doesn't have these specific challenges), so I don't have an opinion on what's the state of art for a task like you describe (I'm imagining analysis of receipts or something like that), however, across many domains of ML (incliding but not limited to computer vision) we are seeing the advantages of end-to-end optimization.
So, for example, the image preprocessing we would like to allow for input of raw phone camera images includes correction for lighting, angle, and crumpledness of paper. Obviously, I agree that these things should be done, but I do not necessarily agree that they must be done as separate engineered features.
I don't have an opinion on what's the best option "today in production" for OCR - it's plausible that the engineered feature way is still the best at the moment, but if we're looking at where the field is moving to, then I'd argue that there is a strong tendencytowards (a) using numerically optimizable methods for these corrections as opposed to hand-selected heuristics; (b) optimizing these corrections for a final OCR result as opposed to treating it as a separate task with separate metrics to optimize; and eventually (c) integrating them in an end-to-end system instead of a clearly separable stage of "correcting for X". I'm not certain where the state of art for noisy OCR is today on this, but it's a one-way direction; The key point of my comment above is that once (if!) we get these things to work I would not expect to go backwards to specific handcrafted features ever. It's plausible that some tasks are better treated as separate and can't be integrated well (perhaps the selection of text segments in your OCR example is like that), but for the features that we already have managed to successfully integrate (which was the topic discussed in the grandparent comment), IMHO there is no reason to ever go back.
Perhaps a relevant example of a field that has undergone this transformation is machine translation - where just a few years ago production pipelines included literally hundreds of separate subsystems for niche subtasks conceptually similar to those you mention regarding OCR, the shift to neural ML was accompanied by making these subsystems redundant as doing the same thing in an integrated end-to-end optimizable way gets better results and is simpler to implement, maintain and debug due to a more unified architecture.
Similar trends have also happened for face recognition and for speech recognition - I would presume that OCR is structurally similar enough to see the same fate if it hasn't already.
It’s not wrong to want to find out what the fruit tastes like. Both kids have good ideas about what it might taste like, and there are many such plausible ideas. But the tree is tall, and remains out of their reach, and all they can do is speculate. Meanwhile, there is no shortage of other interesting things they could be doing.
I think what you are pointing is really the most toxic part of tech: marketers and investors have found that tech is a good way to aggregate money, and so they have a thrown a lot of funding at tech that can aggregate for them. However, we haven’t actually proven that that tech is the best solution or a sustainable solution. We don’t understand most of what we do with computers very well, we just approximate until it works well enough (for the marketers and investors, of course).
ML and deep learning are very valuable, of course, but their recent market dominance doesn’t indicate that they are the final or most correct solution to the problems they are being used for. It indicates that people want to spend money on it right now.
A much better example is P=NP. This is a very rich area of research with many closely related problems that has been worked on for decades and will continue to be worked on for the foreseeable future.
2. The above might upset people but string theory is a good example - partly yes people are held hostage by funding (much like it's hard to avoid being typecast as an actor). But the solution to that is to fund lots more (experimental) movies, not to hope studio heads can pick better winners.
I am perfectly happy for a slice of my tax dollar to go on someone thinking about Pram theory (#) for a decade simply because we have no idea whatsoever how to tackle the big ones in physics, so spray and pray is going to be a lot better than most other options.
(#) Red Dwarf episode. worth watching episode just for this joke.
If you have a country rewarding intellectual effort, that’s a good country to be a part of.
I’m betting that there will be more smart people trying more smart things that will end up positive than not.
I mean we are on a hackers and startups forum - it’s the fundamental premise of a majority of people visiting this site.
Honestly this seems like an issues of not having testable hypotheses. I’d love to see how it gets to a testable state.
I can tell you that the vast majority of academics are not motivated by money. If that was the case they would move into industry where they could earn significantly more.
Good point, becoming an academic may already represent enough of a hoop, though that’s just the CS perspective. In other fields there may not be much of a lucrative industry, so I’m not sure whether this will hold generally.
No, they're motivated by safety. And there's nothing safer than the ivory tower. Academics would work on self-referential research until the end of time if they could get funding for it.
They create academia children in a perpetual loop trying their very hardest to avoid life and all its complexities (especially those of people, of which they are ludicrously blind) by living in their theoretical world of nothingness.
Depends on how you define “academic”. There are suddenly going to be a lot of “academics” showing up for their free handouts once there is no bar.
>I can tell you that the vast majority of academics are not motivated by money. If that was the case they would move into industry where they could earn significantly more.
Selection bias. The academics you are thinking of are the ones willing to put all of the work into getting grants, teaching, etc. There are an order of magnitude more PhD holders that nope out of academia after graduation because of the grim prospects.
If that's the case, you're inevitably going to need some way to decide which of the many worthy opportunities merit sufficient investment and which ones do not.
Perhaps rather than prescribing areas of research, we should just ban or severely limit areas that haven’t shown any promise.
https://en.m.wikipedia.org/wiki/Phlogiston_theory
Then, as now, there was a large body of entrenched individuals who would not budge until some ingenious experiments by the British and French (et al.) to entirely destroy the theory.
https://www.academia.edu/4084776/The_Vortex_Atom_A_Victorian...
With string theory there is no evidence to work off. To me it feels like with nothing better to do everyone in the field is just trying to out math everyone else.
That's what makes this so strange. Everyone outside string theory seems to think string theory is pseudoscience. Why is it still being funded?
Should we stop giving taxpayer money to a group of people who can't convincingly make their case to the outside world, even if they are right?
Or do we trust their word without being able to check for ourselves?
I happen to favor the first approach.
It's it possible string theory will create an experiment which finds something else?
Three theory of Phlogiston appears to have lasted 100 years. Our understanding still has plenty of time to change
Research today is orders of magnitude both bigger and faster, and String theory has been around for at least half a century already.
String theory research isn't directly useful, but other fields of physics and mathematics reap the benefits.
In any resource allocation scheme where the grantors have no skin in the game, politics and cronyism will dominate in the long term.
That's probably what's happened here. String theorists get grants not because string theory produces practical results, but because there is now a network of string theorists and sympathizers in charge of the institutions that control professorships and grant making.
That's why resource allocation should be left to those with skin in the game. A market is one mechanism for forcing allocators to expose themselves to the downsides of bad investments ... but there are other mechanisms if people have an allergy to markets.
“Confirmations should count only if they are the result of risky predictions; that is to say, if, unenlightened by the theory in question, we should have expected an event which was incompatible with the theory--an event which would have refuted the theory.”
- Popper
You're welcome to set the bar of meaningful at any level you like; such a statement cannot in principle be disagreed with. However the point stands that string theory has made more accurate empirical predictions than any other theory of quantum gravity. That doesn't mean no other theory of quantum gravity will ever beat it, just that the focus on string theory among people and funders who are precomitted to studying quantum gravity is rational.
Put another way, so far string theory is an elaborate kind of curve-fitting — a mathematical construction that matched the empirical data available at its inception, but one that to this day has never predicted a novel measurement. Unencumbered by the theory, we would not have expected any recent experiment to produce different results.
And yes, that's what I meant by LQG being a less mature theoretical space. Either LQG will fail to find a solution for 3D space (so sad), or it will and its solution will make falsifiable predictions that are not part of existing theory (yay!), or it will propose a set of solutions that make no new falsifiable predictions (in which case it enters the same state as string theory). Both of those first two possibilities would be forward progress.
In book "Everything is Now" by Bill Spence writes that research in string theory led to developing new mathematics which led to helping to manage data in hadron collider which led to detecting Higgs boson. It is not direct cause and effect correlation but it helped with due process.
For example, astrology probably developed techniques of recording astronomical events and likely some mathematical techniques, but that does not mean that astrology is a correct theory.
I regard consciousness as fundamental. I regard matter as derivative from consciousness. We cannot get behind consciousness. Everything that we talk about, everything that we regard as existing, postulates consciousness ― Max Planck
How are you going to do experiments which test whether 'consciouness is fundamental' or not?
"The tribalistic sociology that has led to a large group of people calling themselves “string theorists” when what they do has nothing to do with string theory is also something I would have thought impossible."
It's my personal bias but I see every group this way.
As someone intrigued by 'cults' early in life, I've come to see that all groups exhibit quite a lot of in/out group behaviour, the weird things is when it happens among those most socially unaware people (often science) - combined with a highly objective discipline (again science) ... people are just unwilling to accept that social artifacts can dominate their sphere.
Now, the existential issue of the fact that their 'entire worldview' and 'life-long work' is tied up in something, delusions can become very strongly internalized.
String theory is at least as interesting for the 'sociology' part as it is the 'string' part.
String theory, and the 'reproducability problem' has kind of shaken my faith in science. I grew up really 'believing' in Science, I view it much more differently these days - including those who have this unassailable faith in it as likely being a little deluded themselves.
I have the same "bias".
What people seem to forget is that the scientific method was supposed to be a way to develop theories that made good predictions in spite of the fact that humans, both individually and in groups, are unreliable and do all sorts of crazy things for all sorts of crazy reasons. Calling oneself a "scientist" or having an advanced degree does not make one less susceptible to that. One has to set up institutions that will harness the craziness of humans and human groups in a direction that, on average and over the long term, is productive.
But what most people call "science" today is not such an institution. For example, the real message of the "replication crisis" is that most of what the average person is told is "science" is actually nothing of the sort. A sane institution of science would never have labelled that stuff as "science" in the first place; it would have said "sorry, we're still working on this, no results yet". And a lot of times it would say that and there never would be any results, and science would say "no results from this line of research; that's to be expected since many lines of research that initially look promising don't pan out".
Finally, I would argue that the large problem stems not from within science, but from how "science administratio/funding" has changed. Essentially, the idea of making science more (a)countable/measurable has lead to a state where scientists spend large fractions (>30%) of their time applying for funding. Also the funding is very much tied to some measurable outcomes, so there is just no time left to dedicade to reproducing results, because you can't get funding for it.
I believe we as society need to come back to a view of seeing science as a general good, without the need to produce some immediate measurable outcomes (in fact this is true for many aspects of society) . Unfortunately the MBAs are running the show almost everywhere so I don't see this happening anytime soon.
Yes, this is part of what I was referring to when I said that humans do all kinds of crazy things for all kinds of crazy reasons.
> In the end the best science always won
I'm not sure I agree. I would agree if you limited your claim to hard science--areas where we can nail down the best theory by lots of data and controlled experiments--but many areas that are called "science" are not amenable to that.
> by the very nature of us speaking about it, it can't be an existential crisis
I'm not sure I agree with that either.
> I believe we as society need to come back to a view of seeing science as a general good, without the need to produce some immediate measurable outcomes
I agree with this, but I would also add that science should be viewed as a tool for finding good predictive theories, no matter what those theories end up predicting. Unfortunately much funding of science now is given with the expectation that a particular kind of result will be what "science" predicts.
I think it's rather that we have had a bunch of great curious people who worked on interesting stuff and found out a lot about nature. But we can't equate that with institutional credentialed authoritative "Science".
I guess there is some underlying scientific ethos, that can be learned between the lines, but I don't think it's a clear method (as in the intro textbook definition).
The ethos is simple: all scientific models are judged by their predictive power, and nothing else. If your model does not make accurate predictions, it is wrong, period. Nothing else matters.
I agree that there is no single "method" that guarantees that models will satisfy the above criterion.
Then String Theory is rubbish because it predicts nothing?
Every single theory in science started out with some 'predicate knowledge' and based on 'intuition', scientists reached out a few leaps with some ideas.
Some proved right, many proved wrong.
Some took a while to come together.
"String Theory" is different - it was not based on 'intuition' or any predicate foundation.
'Strings' are a fantasy - just completely made up.
Some people took a fantasy 'new origin' and put tons of math around it.
Since then, they've never been able to map it to reality, or even conceive of an experiment let alone do one.
What it looks like is religion - not only in it's foundation, but in the behaviour of it's acolyltes.
How does a Priest react if you tell him 'God does not exist'?
How does a String Theorist react if you tell him 'There is no String Theory it's just your imagination'.
Perhaps in the greatest of ironies, I actually believe God is more likely to exist than String Theory (!) or at least, there might be some kind of actual spiritual reality to our existence.
Strings? It's just an idea that snowballed into something.
How you come up with ideas and directions is a mystery. It's luck but not just. Kinda like investing.
Feynman also just described that step as "Guess it".
Its not clear at all that our institutional organized science process (grants, papers, academic positions and evaluations) correlate to better success at finding out stuff.
No, we don’t just say that. Recognizing that “knowledge” about the universe is always tentative and contingent and that an understanding of the universe is to be replaced when one with strictly better at predicting observations is identified is the scientific process. And it's a pretty big change from the approaches taken prior to it's adoption.
> In this way science can never be really wrong, because it is always open to being proved wrong
“Science” is almost entirely a combination of openness to being proven wrong and a particular understanding of what it means for something to be proven wrong. Since it is an epistemology, it cannot be “really wrong” if it is consistent, except in terms of a different epistemology.
> I think it's rather that we have had a bunch of great curious people who worked on interesting stuff and found out a lot about nature.
We had a lot of that without the scientific method, too. What the scientific method did is to provide a framework in which the work of great (and not so great) curious people could be effectively aggregated so that they built-in each other, whereas pre-scientific approaches did that notably less well in terms of building an ever-better understanding of the physict universe, because work was judged by standards other than predictive power.
> But we can't equate that with institutional credentialed authoritative "Science".
Science is exactly about authority having little to do with institutional credentials.
Much of the 3 was created using approx 1, before 2 really. 3 is the original source of respect for the masses, the landmark results, the household names, Newton, Galileo, Copernicus, Darwin, Einstein etc. But then come along 2 with their little games and citations and positions and visa requirements and recommendations, CVs, conferences, journals, publish or perish, the push for sexy results for PR announcements and the next grant, p hacking, tyrant supervisors who only care about having a paper at any cost, massaging the data, not speaking up because if funding dries up for the institute, the newly birthed PhD student may lose her job and be deported, the big name prof must be worshipped etc. They co-opt the reputation accrued by the giants, and then if anyone has any criticism, they are scolded for being against the ideal 1, the noble endeavor and ethos and upright honest methodology. In many cases people good at playing game 2 need quite different skillsets than people who'd be good at 1.
But who will produce the next big item for category 3?
Is 2 a good mechanism to encourage practicing 1 to get to 3? I don't think so.
It depends on what the disdain is based on. If the disdain is just based on "your field isn't physics (or chemistry, etc.)", I agree that's not justified. But if the disdain is based on "your models don't make good predictions", I think it's justified. Scientific models are supposed to be judged based on whether or not they make good predictions; if they don't, they're wrong. Models that make good predictions are much, much rarer in the "soft" sciences--many such fields basically have no models with any predictive power at all, yet their proponents claim they are "sciences" and want their models to be used to, for example, make public policy.
Humanity can not afford to spend all their energy and resources for scientific research. We have many things we need for survival, and some for entertainment, etc. Obviously we do want to invest some amount of resources and human energy into research - but the question is how much is enough and which fields should we invest more or less ?
I claim that these questions are fundamentally unknowable / unsolvable. So we guess blindly (and very occasionally with a tiny bit of very myopic foresight), and mostly end up unwittingly relying on various historical accidents.
This is a very, very difficult question for humanity overall, and we're never going to get "right" amount of funding or "right" structure as there's no right answer. "MBAs" do and can hinder progress unnecessarily, but not always - they do exist for a reason and they represent society's collective attempt to avoid waste or runaway spending.
Governments spending bazillions just want some measure of accountability and that's it.
Aside from the more existential issues you've noted ... there probably a host of things that could be improved.
One thing may be the recognition that a lot of science just isn't worth repeating at all. To simply 'dial back the faith' we have any in any single experiment, and assume that every first pass is fuzzy.
I don't think anyone is accusing anyone of fraud here, this is not a problem of 'a few bad apples' - it's more of a broad systematic issue of 'looking for positive results'.
Perhaps some opportunities lie in getting more behind null hypothesis stuff as being 'respectable'.
I agree it is a cliche, but it is a very observable one over the last decades. Regarding resource constrains. I argue it's just the other way around. If we were resource constrained we would not build huge administrative bodies to check that the scientists are actually doing what they are supposed to. Take an example people here are likely more familiar with: in startups there are generally much less administrative processes, those come only in once companies get to a certain scale. Startups can simply not afford to pay lots of people to administer their "progress".
> Humanity can not afford to spend all their energy and resources for scientific research. We have many things we need for survival, and some for entertainment, etc. Obviously we do want to invest some amount of resources and human energy into research - but the question is how much is enough and which fields should we invest more or less ?
Considering the amount of energy we spend on things that are not needed and are actively hurting the planet and our survival, I would say we can't afford not to spend more on science.
> I claim that these questions are fundamentally unknowable / unsolvable. So we guess blindly (and very occasionally with a tiny bit of very myopic foresight), and mostly end up unwittingly relying on various historical accidents.
> This is a very, very difficult question for humanity overall, and we're never going to get "right" amount of funding or "right" structure as there's no right answer. "MBAs" do and can hinder progress unnecessarily, but not always - they do exist for a reason and they represent society's collective attempt to avoid waste or runaway spending.
MBAs are quite ill equipped for this tasks though. That's the my main gripe they try to apply a process that has problems even on the scale of a business (the problem of focusing on short term gains instead of long term sustainability) and try to apply it to society at large, even though there is no evidence that it would work.
Nowadays we have a meritocratic expectations and want to give equal opportunity to all, our cultural ideal of governance is democracy etc. It this framework a big thing to avoid and a big source of angst is that people will take these tax paid jobs and do jack shit and just sit around having fun. We want to avoid that, and replace any lazy non working people with people who will do it better.
For this we need some accountability. To see that papers get written and not only that, but they are also having an impact on other research (citations). What else can you do? You can have time sheets of working hours or computer software that tracks usage to see if someone just watches YouTube or reads the news at work... I mean, sure you could base it on common sense and talking with the researchers, but that's almost like asking them to write a report, but then why not publish the underlying work as well if it was good work?
Earlier however, science was more like a hobby and the background system was various forms of feudalism. Many scientists were rich noblemen who had the wealth to pursue their interests. If nothing came of it, they were still well fed and had nobody to report to. Or it was enough to have a patron who liked work enough that he fed the scientist. Or the it happened within the church with some theological justification/disguise, but that wasn't democratic either.
Since we now conceptualize power bottom up, from the masses to the intellectuals and not top down from the monarch/God to noblemen/intellectuals to the masses, we need a system to justify funding decisions to the people. And the easiest way is to point to papers and awards and experiments and machines etc.
In a way the old aristocratic system perhaps gave more freedom to a small number of scientists, today we have tons of scientists who all need to follow quite a rigid path.
I guess there was a time in-between, till the seventies or eighties maybe, when things were pseudo aristocratic with relatively few scientists with lots of freedom to do whatever the hell they please, including just reading newspapers or books all day long if they so decided.
I think our collective conscience is catching up and demands objective accountability to things are working in a just manner.
But papers and citations are the wrong metric for accountability. What we want from government funded science is models that make good predictions. That's what we need if the knowledge gained from government funded science is going to be useful to the government and the taxpayers. So that's the metric we should use to judge it. Number of papers written and number of citations are, at best, irrelevant to that metric, and at worst, actively opposed to it; it's much easier to generate papers and citations if they don't have to be based on models that actually make good predictions.
This is very frequently repeated, but I have never encountered any reason to believe it is true. Certainly, better science has been seen to win, in cases, but the amount of shit hurled first has always been enough to drive any sensible person to do something else less likely to attract hurled shit.
We can get a sense of how often correct theories are suppressed by looking at cases where a now-accepted theory was anticipated decades or even centuries earlier, but failed to gain traction at the time, not for lack of evidence, but simply because the evidence was unwelcome.
Theory of causation was understood a full century ago, but suppressed by the statistics mandarinate, and is only in the past 20 years starting to resurface. Evidence conclusively demonstrating human presence in the Americas before 15kya is still routinely dismissed. Requirement for surgeons to wash their hands took decades to catch on; to this day, more hospital ICUs have the traditional 4% incidence of fatal infection than the very gradually spreading 0%. The entire lack of any negative health impact from eating saturated fat was well demonstrated decades ago, but has only been accepted in this decade. These are just cases I know about; most have been better interred.
I am forced to conclude that there remain many more successfully-suppressed correct theories than accepted correct theories, and that things are getting worse, not better. I.e., correct theories still win sometimes, but the fraction of correct theories that win is decreasing in the face of tenured grant committees.
People still believe the Earth is flat.
Similarly in Alzheimer's research, still enforcing the failed amyloid "consensus".
I do not know what is your experience with science, but it is not how it works. In most cases research leads to unexpected results in grand scale. F.e. Research in string theory field led to developing new mathematics which eventually helped to calculate particle paths in LHS, which helped to make gathering data manageable and which lead to discovering Higgs boson.
>most socially unaware people (often science)
99% successful scientists are very socially aware. Usually socially unaware people do not become successful even in science.
>social artifacts can dominate their sphere
Because thats how world works. F.e. phd student does his own research, then he consults with his professor, who suggests how to add marketable layer to his research and with that layer his research becomes one for string theorists, though it was not his initial goal. This marketable layer helps him get the grant.
>Now, the existential issue of the fact that their 'entire worldview' and 'life-long work' is tied up in something, delusions can become very strongly internalized.
This is made by popular media, movies. Most scientists live normal life, create families, try separate work from life, therefore research does not come from worldview.
I think this is koolaid.
String Theory is Probably Wrong.
Thousands of PhDs have been minted, a zillion talks and publications, conferences, think tanks, books, 'Science PR' - all for wish-wash.
That 'String Theory helped some thing over in the corner' does not validate all the investment.
With the clearly lacking 'Self Awareness', it's possible that Science could have gone in a variety of directions, producing or necessitating all sorts of other, likely more useful things, out of which some 'useful math' may have also been developed.
But instead we put a lot of effort into vaporware.
Why? Because it's human nature - people don't want to accept they are riding a meme, and lack objectivity.
...
"This is made by popular media, movies. Most scientists live normal life, create families, try separate work from life, therefore research does not come from worldview. "
Again, I don't think you're maybe grasping what I am saying.
It's normative for people, especially serious people, to be wrapped up in a 'world view'.
Businessmen talk about 'free markets' all day long, often completely ignoring real externalities.
Socialist politicians, clouded by their goodwill, are often dangerously unaware of basic economics, the realities of Supply and Demand.
Military Generals may sometimes see a possible 'military solution to every geopolitical problem'.
It's even 'more normal' for someone working on something arcane and abstract to be tied up in their passion and ambition.
Case and Point: String Theory.
It seems as though String Theory is probably mostly hogwash.
How many String Theorists would admit that? Or even the very real possibility of it?
And how many will take their 'belief' to the grave?
It's obviously not 'entirely a waste' - but if Science wants to really maintain credibility, it needs a giant dose of self awareness there - and also with the 'reproducibility' problem.
Edit: for some nice examples of cultish, self re-affirming behaviour [1].
Usually those kinds of overviews are limited to things like media genres, not fields of science. (eg best video games of the last 10 years, best movies and how they changed the industry etc). Something a bit more academic would be nice...
But what is really clear is that there is no real institutional effort to do what you suggest above - folks are not really tracking how well the funding system is working or what has worked and what hasn't. All of the reviews that I have seen are forward looking and driven by science insight, administrative change seems to be done by MBA fashion - or inertia.
1. 20 years ago the author wrote a paper asserting that string theory was a fruitless field of study.
2. The paper is written in a confrontational and snarky tone, for example: "The theory has been spectacularly successful on one front, that of public relations." Not a way to build bridges.
3. The author seems bitter that the paper wasn't accepted into Physics Today, and that the suggested rewriting into a letter to the editor did not result in publication either.
4. The author is sticking to their guns.
No matter how well-founded the idea that string theory is a dead end, this isn't likely to improve things.
If you're offended by the audacity of scientists criticizing pseudoscience, you may want to read about the concept of "polemics". It's an acceptable, longstanding form of rhetoric for when a respected figure publicly disagrees with a direction of another group of people.
https://en.wikipedia.org/wiki/Polemic
Incidentally, the author even noted his article was a polemic in the original abstract https://arxiv.org/abs/physics/0102051
I don't think it is possible to make such a determination yet.
If string theory is in fact a flub - and the field can't take a basic article of this 'tone' without flinching - the science has a problem.
Preferably a book that doesn’t go off into the deep end with alternate realities and dead/alive cats in boxes.
I read Higgs by Jim Baggot, it did a good job of providing an overview - however I would like something that goes into a little more detail while still being a popular science book.
I say this as someone who is quite familiar with QM, so it's very possible I'm just neglecting to consider all the implicit context I'm working with.
This is because we teach students the "Shut up and calculate" formulation of Quantum Mechanics by default. Interpreting your mathematics doesn't build transistors or fund particle accelerators, for example.
It's still an important thing to consider but it doesn't have the same logical structure as day-to-day Quantum Mechanics
Plus, I think it's possible to have a smart discussion about QM interpretations without reference to cats in boxes.
When I first understood what entanglement was, mathematically, I was profoundly disappointed. Years of pop-sci had made it seem like a crazy thing, but mathematically, it's quite.. tame. (Some tensor products don't factorize).
For quantum mechanics, I strongly suggest going the quantum computing route if you have a background in computing. Pick up Quantum Computing since democritus (https://www.scottaaronson.com/democritus/), Nielsen and Chuang(http://mmrc.amss.cas.cn/tlb/201702/W020170224608149940643.pd...), and learn how to program in Q# (Microsoft's quantum programming language) and solve their quantum katas: https://docs.microsoft.com/en-us/quantum/tutorials/intro-to-....
I've done the same thing as part of a course at university: https://github.com/bollu/quantum-course-exercises
and it made a lot of things click for me.
I'm sure you know this, but it's that some elements of the tensor product space don't factorize (i.e., are not tensor products).
Is the _math_ quite tame, or is the phenomenon itself?
That's the beauty of quantum computation; it harnesses all the intuitions we already have about computation, skips the messy parts of solving stuff in quantum mechanics while still keeping all the fiddly interesting bits that makes quantum quantum.
I really do believe that learning quantum computation is the "correct way" (for some folks) to gain an intuition for quantum mechanics
It is only a few months old & doesn't shy away from math.
Reminds me that my brain literally turns to mush whenever topology and differential geometry comes into play - I can power through it, so far at least, but it doesn't seem to slot into my head as well as almost anything else I've come across.
Next, I want to be a FAANG developer without knowing how computers or programming languages work. Or how to type.
There's only so much you can say about a topic, which is why you end up with the cats in boxes.
It's really hard going for a non-physics person. I tried to read it end to end once and spent 9 months on it, I subsequently tried to read various interesting chapters out of order and didn't come away feeling that I had mastered the material or thinking. But it's all there, if you are able to crack it open.
I've spent a lot of time thinking about this subject (in fact, I studied string theory at Columbia which is where Peter Woit is). The main issue that people have with string theory is the lack of testability / falsifiability. I think there's definitely some issues there, but I think blaming string theory is not quite right.
There are two reasons why string theory is tough to test. One of them, and the most obvious one, is that the energy scale where the strings exist is super high (likely at or near the Planck scale). This means that if we wanted to detect the direct signatures of strings, we'd have to create an experiment that is sensitive to these energy scales. This seems to be completely infeasible (people talk about particle colliders having to be the size of our solar system etc.) Now, interestingly this is not actually only a problem with string theory but rather with any theory of quantum gravity. This is because the fundamental scale of a theory is usually determined by combining its fundamental constants in some way as to produce an energy scale, and any theory of quantum gravity must somehow contain Planck's constant, Newton's constant of gravitation, and the speed of light. Taking these together gives you the Planck scale.
So, in other words if the reason you don't want to study string theory is that it only is testable at the Planck scale, then what you're really saying is that you don't think we should study any theory of quantum gravity. This is, I think, way to extreme of a position.
Now, interestingly, string theory is actually more than a theory of gravity, so unlike something like e.g. loop quantum gravity which is only a theory of gravity, it's conceivable that string theory somehow within it contains information that "trickles down" to lower energies and thus could potentially be testable at something like the LHC. This leads me to the second reason why string theory is very difficult to test.
It turns out that the equations of string theory are more-or-less unique at high energies but that as you start lowering the energy at which you probe the theory, multiple distinct solutions emerge. These solutions turn out to have a very nice physical interpretation: they are the different ways in which we can compactify the extra dimensions of string theory. Regardless of this physical interpretation, the fact remains that there are many many distinct solutions of string theory at low energies, and in order to make predictions that are falsifiable, we need to know which of these solutions we're living in. This is where the crux of the problem lies. It turns out that there are so many solutions of string theory that we cannot even in principle go through them one at a time to see if they're feasible (people throw around the number 10^500).
Now, it turns out that the real problem is not actually in the number of distinct solutions to string theory (~10^500), but rather in the way their structure is poorly understood. In fact, any theory of physics contains an infinite number of theories within it. For example, consider the mass of the electron as a free parameter. Until I tell you what the mass is, you can't make a complete prediction for what the energy levels in Hydrogen are. In fact, you could argue that since the mass of the electron is a real number, there are in fact an infinite number of predictions to the energy levels. A bit sarcastically you could then say that at least string theory has "only" 10^500 different theories within it, unlike traditional physics that has this continuous infinite set of theories.
The distinction between these two cases is then that for traditional theories, we can go the other way. If we measure the energy levels of Hydrogen, we can infer the mass of the electron. Then, knowing the mass of the electron, we can make other predictions. It's this last step that's currently missing in string theory. Currently we only know how to move in one direction: give me the solution you're talking about and I might be able to make predictions, but give me observational data and I can't work backwards to determine which solution I'm in. It's almost like a one-way hash.
I would say that this last objection is a roadblock that we're currently facing, and it's not perfectly clear that it's not solvable, nor is it clear that it is. I think that until we solve this problem, string theory will be stuck and people will be pointing fingers at the theory calling it a fool's errand. I personally think this criticism is misguided.
The upshot of this is that most people who work on string theory work in areas that are not plagued by this bifurcation to low energies. For example, you can use string theory to study the structure and behavior of black holes and holography, something called AdS/CFT, an area that has been incredibly successful.
I do not think it is correct to say that special relativity for example gives 10^500 possible classical theories. We know special relativity needs to produce same outcome as Newtonian mechanics or Maxwell’s equations in the limit as velocity tends to 0. In my simple example, we have 2 well established theories which bound the possible outcomes. This is all possible even without knowing the speed of light.
I like the analogy of a hash function. Imagine that someone gave you the exact specification of a hash function (e.g. sha256) as well as the hash value of a list of inputs. The only thing missing from the story is the salt that was used in hashing the inputs. You're asked to make a prediction of what ought to happen when you hash the string "hello", but unless you know the salt you can't figure it out. So, you study all the examples provided and try to find collisions so that you can figure out what the salt is. The problem is that while it's easy to hash values, it's very hard to find collisions. It's really frustrating because in some sense you have all the information you need, but unless you're able to find vulnerabilities in sha256, you can't move forward. So, you spend a lot of time trying to understand what this hash function is really doing. Maybe some day you'll crack it at which point you'll be able to figure out the salt and ultimately make your prediction. However, until that day people around you keep telling you that you're being silly because your theory lacks predictive power. They say things like "your theory can predict anything you want it to, just pick your favorite salt and it'll output whatever you want!". It's not that the theory is necessarily wrong, it's that you don't fully understand it yet.
If the theory is neither testable nor verifiable, then how do you (or anyone else) know you're even on the right track?
Keep working on it? You won't know until you take it to the end, or at at least until other theories show promise.
You have created a mathematical structure that is so flexible that it can fit any data. It is useless. You have abandoned the hypothetico-deductive model.
> For example, you can use string theory to study the structure and behavior of black holes and holography, something called AdS/CFT, an area that has been incredibly successful.
Here, "successful" only means that other (famous) string theorists have found those ideas worth pursuing. Then, other people, in turn, picked up the idea because the former people had praised it and then they, too, would receive praise. (It's almost like what people on Reddit call, pardon my French, a circle jerk.) However, AdS/CFT hasn't produced a single bit of verifiable experimental evidence in the realm of black holes. It's not even clear what AdS/CFT has to do with our (clearly non-AdS!) universe.
> However, AdS/CFT hasn't produced a single bit of verifiable experimental evidence in the realm of black holes
No, I don't think it is. I agree of course that it's hard to test predictions regarding black holes. But I was addressing OP's statement that
> For example, you can use string theory to study the structure and behavior of black holes and holography, something called AdS/CFT, an area that has been incredibly successful.
and I merely intended to express my disagreement with the claim that AdS/CFT has been "incredibly successful". Successful in what way? Measured by what metric? Clearly not by the usual metric that says that experimental evidence is what counts. And on top of all of that AdS isn't even close to the spacetime describing the known universe.
> Given the observational knowledge of today, even a true theory of quantum gravity beamed back from the future would fail to produce verifiable experimental evidence in the realm of black holes.
I don't think this is necessarily true. See, for instance, https://backreaction.blogspot.com/2020/01/how-to-test-quantu...
For many years it was conjectured that gravitation is actually most clearly understood as a two-dimensional theory (people call this the principle of holography). The rationale is that the entropy of a black hole is proportional to its area rather than its volume. Since a black hole is in a very specific sense the object of maximum entropy, that means that the amount of information that we can store in space is not proportional to its volume, but rather the area of the encompassing "sphere". This is super surprising since it suggests that what we perceive as 3d is really just an illusion and that the correct formulation of physics ought to be a 2d theory.
Now, for a long time we didn't have a way to actually write this sort of duality down. We sort of knew one side of the story (gravity with general relativity), but it fails in precisely the domains where we want to investigate it (black holes). String theory, whether you believe it's the true theory of everything or not, is nonetheless a mathematically consistent theory of quantum gravity (in a sense it's at least an existence proof that gravity can consistently be quantized). As such, it's at the very minimum a great arena to analyze these problems carefully. The first explicit construction of the duality between a 3d theory of quantum gravity and a 2d theory without gravity is precisely AdS/CFT. It says that a quantum theory of gravity is AdS space is mathematically and physically equivalent to a 2d theory of a conformal field theory that lives in 2d.
I don't think it's fair to say that AdS/CFT is being studied because famous people like it. It really does have a lot of value if nothing else than as a playground to understand how one could in principle formulate these dualities consistently.
> String theory […] is nonetheless a mathematically consistent theory
I hear this claim being perpetuated a lot but I have yet to see a string theorist give a mathematically rigorous introductory lecture on string theory. Don't get me wrong, I'm not saying there are no mathematically precise results in the realm of string theory but I know enough about functional analysis and the issues surrounding the mathematical underpinnings of quantum field theory (or even quantum mechanics) that I'm not buying your claim and my current view is that some parts of string theory are very rigorous, whereas (most) others are not. (AdS/CFT is one such part which belongs to the latter category.) I'd love to be proven wrong, though, so please feel free to send me papers etc.
When I say mathematically consistent, I mean it in a looser sense. If we take a step back to before string theory, there was no way to get consistent results from quantum gravitational calculations. The usual tools that we use to renormalize quantum field theories do not work for gravity. This suggests that there's some type of "ultraviolet completion" of general relativity. In other words, the theory of GR ought to come with some implicit energy cutoff beyond which the theory somehow changes. String theory is such a change in that the stringy corrections to GR would only come into effect around the Planck scale. It's by no means necessarily the unique such completion, but as of now it's the only one we know of.
As to your other point, I think it's a good idea to reframe the work done in string theory as (what I used to joke) "theoretical theoretical physics". In other words, it may be the case that string theory is a true theory of nature, but even if it isn't, the theory lets us explore what such consistent theories could look like and how various paradoxes (like the information paradox in black holes) get resolved. These types of insights may point us in a direction of further investigation that may very well fall outside string theory.
In other words, at the very least (and I personally think this is underselling string theory by a lot) string theory is a proof of concept and a very powerful, sophisticated, and rich arena in which we can begin to understand the salient features of quantum gravity. One such example is AdS/CFT.
So, take as a silly example a regular pendulum of mass m and length l swinging on earth where the acceleration due to gravity is g.
Suppose we want to figure out what the period of the pendulum is (i.e. how long it takes for it to make one full swing from left to right back to left).
We could go ahead and solve this by using Newton's laws, but at the end of the day whatever formula we find should have within it the parameters mentioned above (as well as regular numbers like 2s, pis, etc.). If we take a look at the units of the parameters above, we see:
mass of pendulum (m): kg
length of pendulum (l): m
acceleration due to gravity (g): m/s^2
Out of these we want to produce something with units of time, and it turns out that there's a unique way of combining the above to get that: 1/g has units of s^2/m, so l/g will have units of s^2. Just take the square root and you find sqrt(l/g).
Now, this doesn't necessarily mean that the period of the pendulum is exactly sqrt(l/g), it just means that it needs to be proportional to this. In fact the formula for the period has an extra factor of 2pi in it: 2pi sqrt(l/g). The point is that the general magnitude you'd expect for any timescale associated with the pendulum would be roughly equal to sqrt(l/g).
For quantum gravity the story is the same. We have some parameters that we expect should be deeply part of any prediction: Planck's constant, the speed of light, Newton's constant of gravitation. The units are as follows:
Planck's constant (hbar): m^2 kg / s
Speed of light (c): m/s
Gravity (G): m^3 / kg s^2
From these we want to figure out what the relevant energy scale is (i.e. when will this theory start differing from a non quantum mechanical version of gravity). Well, there's only one way to combine these into an expression that has the dimensions of energy. This particular way of combining the constants is huge and is traditionally called the Planck scale:
E_Pl ~ sqrt(hbar c^5 / G) ~ 10^19 GeV
In contrast, the LHC is currently probing physics on the scale of ~ 10^4 GeV so we're not even close to probing things in the realm of quantum gravity.
Note that this analysis has nothing to do with string theory at all and generalizes to any theory of quantum gravity.
I guess to parametrize the theory it takes something other than masses of elementary particles and a few constants?
If I had that big of a solution set I'd definitely try to search through it. If one string theory solution can give me a predict() function that I can verify, I'll find a way to parametrize it (deep neural networks, graph data structures) and then check if predict() matches the predictions we know.
ditto, of course, goes to holding accountable the economic beliefs of politicians and talking heads of all persuasion. the people conscious of how much we don't know get drowned out by the people who don't care.
I gather that Alzheimer's research is still almost exclusively about amyloid plaques and tau tangles, despite absolutely negative results and disastrous drug trials for decades.
And, of course, magnetic confinement fusion is still 20 - 30 years off, and always will be.
I like to know the subject before I wander off HN following a nondescript title...
Plus, a little bit of randomness is the spice of life.
The author is a physicist/computer scientist writing on his personal blog. He is writing about some of his own work, 20 years later, and his likely audience probably has context about that work, or will be linked from somewhere that provides that context. The fact that you and I stumbled upon it from a decontextualized aggregator is a bug.
> Please don't do things to make titles stand out, like using uppercase or exclamation points, or saying how great an article is. It's implicit in submitting something that you think it's important.
> [...] Please use the original title, unless it is misleading or linkbait; don't editorialize.
My personal observation, there are three types of bad titles.
1. Clickbait titles editorialized by a submitter to make everyone know how great/important the article is.
2. Clickbait titles from the source.
3. Non-descriptive title from the source.
60% of the bad titles are Type-1 title from the submitters (as everyone who ever browsed Reddit would know). To prevent such behaviors, the "use original title" rule is implemented on Hacker News.
30% of the titles are Type-2 title. On Hacker News, if many readers loudly criticize that the original title is misleading, eventually a moderator will correct it manually.
10% of the titles are Type-3 title, non-descriptive title that contain little information, but it's neither misleading or clickbait. On Hacker News, usually, nothing happens. This article is one of them. Also, any attempt to change the title could be seen as Type-1 title editorialization, so the original title remains. So sometimes, even after the title has been changed by the submitter to be more descriptive, it could be reverted back to the original, non-descriptive one.
> This site doesn't allow comments without substance, but for some reason link titles don't have the same standards.
Whenever people compliant that Hacker News allows non-descriptive titles, please consider the fact that "use the original title" does more good than harm in general with a false negative rate of only 10%, which is a small price to pay.
Lemme paraphrase the great Arthur C. Clarke about this. We are fish in water. And some time fishes do jump out of the water. And sometime fishes, while jumping out of the water are seeing fires on land.
That's String Theory for me - it's a fire on land that we get a glimpse of it and can't quite make what it is. And until we will not walk on land and make fire ourselves we'll never fully understand its potential. 20 years is nothing in this regard.
My 2 cents.
I think there's a good candidate for this shape. Many books, rather old, but completely inadmissible in the court of modern science, describe the smallest particle in a way that curiously resembles a string. The string forms a coil, very similar to a wire coil in electromagnets, that revolves almost 3 times around the central axis and then returns back to the entrance. The outer shape is an oblate spheroid and the entire thing is dynamically stable. The coil itself carries electricity, and because of its shape, it produces a magnetic field that streams thru the center. The particle performs rapid gyrations around the central axis - that's probably the same effect as the Lamour precession. There are two types of this particle: one with clockwise direction of the coil and the other with counterclockwise. The string enters and exits the area in a direction perpendicular to all 3 dimensions, so it kind of appears from nowhere. The string itself is made of a finer coil, and that of even finer coil and so on a few times. If the 10 dimensions guess is correct, then there are probably 6 or 7 such finer coils. If the entire thing is unrolled, it would form a circle of very large radius. While such a particle, that probably corresponds to the electron, has enormously complex internal structure, it's still a fundamental particle of 3d world, because cutting the string in any place makes it recombine into numerous tiny coiled strings, but all of them freely float along the 4th dimension and barely interact with the 3d particles, so they become kind of background noise of unknown origin. The magnetic stream that flows thru the central axis of each 3d particle is made of these tiny 4d particles, and those probably replicate the entire arrangement at a smaller scale, so there probably are even smaller 5d coiled strings and so on.
Like I've said, this kind of description is inadmissible in science, but it's ok to "guess" it and invent a plausible explanation that this shape minimizes some weird functional in a hilbert space of the strings theory.
Oh really.