- In a couple hours, at 1:15 pm New York time, there will be a press conference at the AAS meeting where LIGO and Virgo scientists will discuss “ongoing research” (webcast here). The general assumption is that there will be observations of new gravitational wave sources announced.
- At some other extreme in the space of science talks, the AAS meeting also featured a talk yesterday by Sean Carroll on “Normal Science in a Multiverse”. There’s some discussion of the talk here, and twitter has this. Some counterpoint from Joseph Silk here.
It seems that Carroll was arguing that the multiverse shows that we need to change our thinking about what science is, adopting his favored “abduction” and “Bayesian reasoning” framework, getting rid of falsifiability. Using this method he arrives at a probability of the multiverse as “about 50%” (funny, but that’s the same number I’d use, as for any binary option where you know nothing). So, from the Bayesians we now have the following for multiverse probability estimates:
- Carroll: “About 50%”
- Polchinski: “94%”
- Rees: “Kill my dog if it’s not true”
- Linde: “Kill me if it’s not true”
- Weinberg: “Kill Linde and Rees’s dog if it’s not true”
Not quite sure how one explains this when arguing with people convinced that science is just opinion.
- Among the many summer conferences one might want to take a look at, there’s last week’s Workshop on String Theory and Gender, and this week’s LHCP in Lund. Wilczek will be giving the “Theory Vision” talk at the end on Saturday.
- Today’s Wall Street Journal has an interesting article on the not so great job market for Ph.Ds (to avoid paywall, try Googling, e.g. “Job-Seeking Ph.D. Holders Look to Life Outside School”). I find the claim that the median income for Math/Physics mid-career Ph.Ds dropped 6% over the past 3 years highly remarkable (if true). Yes, US middle class incomes have been tanking, but I that number is pretty extreme, especially since this has been a period of modest economic expansion.
One other bit of news I learned from the article was what universities (including mine) are doing to help with the situation:
at Columbia University, Ph.D.s are taking classes in using Twitter to better communicate their work to nonacademic audiences.
Update: The LIGO news was a second black hole inspiral. I’m sure you can find good coverage of this elsewhere.
I hadn’t realized that the AAS is sponsoring a whole Multiverse Mania Fest, bringing in to promote a new definition of science not just Carroll, but Richard Dawid. Lenny Susskind this afternoon gave a talk (see here) that seems to argue that the Multiverse is a great idea, even though it won’t ever be testable. No news on what his Bayesian percentage is, or whether he’s willing to bet the lives of helpless pets. Sean Carroll made his usual straw man attack on the “Popperazi” (who, despite what he thinks, understand what indirect evidence is), see here.
Update: Another odd multiverse-related item. Laura Mersini-Houghton and collaborators have made well-publicized claims that they have testable predictions based on the string theory landscape. I’ve written about these several times here on the blog, see here, and this posting for one example that includes a response from Mersini-Houghton and Richard Holman.
Their claims are based on two 2006 papers, see here and here. Very recently Will Kinney posted this paper on the arXiv, which has in the abstract:
we compute limits on these entanglement effects from the Planck CMB data combined with the BICEP/Keck polarization measurement, and find no evidence for observable modulations to the power spectrum from landscape entanglement, and no sourcing of observable CMB anomalies. The originally proposed model with an exponential potential is ruled out to high significance.
See the conclusions section of the paper for the details.
This isn’t particularly surprising or odd, although one wonders if the Kinney paper will get a fraction of the attention that the original claims have gotten. What is odd is that I hear that Mersini-Houghton is asking to have the paper removed from the arXiv, on grounds that have something to do with the fact that she was originally collaborating on the project with Kinney, but is not listed as an author (although he offered to put her name on it). I can’t think of another example of this kind of thing ever happening before, perhaps others are aware of similar controversies.
Update: Much more detail at Backreaction, including comments from Will Kinney explaining the issue with the arXiv.
I do think that multiverse proponents have abducted the scientific method. But I don’t think that is what Carroll has in mind.
Carroll: “About 50%”
Polchinski: “94%”
It’s just an absolute nonsense these probability guessing games, aren’t they? I have no idea why highly-regarded physicists are doing this.
When a weather forecaster tells me the probability of rain tomorrow is 50%, I translate it as “I don’t know.” With a greater than 50%, I hear “There is more reason to think it will rain than it won’t” and vice versa with less than 50%. Given Sean Carroll’s past championing of the multiverse hypothesis as an explanation for fine tuning, I’d have thought he’d assign a somewhat higher probability.
I’m surprised there’s so much hostility to these percentage based confidence levels.
Because I’m 99% sure these guys are full of sh*t.
I do not like to see the very important LIGO results juxtaposed with nonsense (a wee bit more polite Tim, no?).
I agree with paddy that more politeness would be a good idea. Not good to emulate the tone of a certain other blog.
On the other hand, I’m somewhat fascinated by the juxtaposition of real scientific progress (LIGO) and pseudo-science (the Multiverse) at the AAS meeting. Given the difficult situation of HEP theory, I can understand why string theory meetings sometimes feature Multiverse stuff, but if your field is healthy with serious new advances like this going on, why would you schedule all sorts of pseudo-scientific talks of this kind? The contrast must be quite jarring.
I like to believe that this interest in the multiverse is not permanent. It’s likely that a lot of people got now-defunct theoretical training, so this is what they can do. They may have trouble acquiring and training new generations of elite graduate students, who will be more interested in pursuing more dynamic research programs, such as gravitational wave modelling and LHC phenomenology.
Do you have a link/reference for point 4. and 5. regarding multiverse (Weinberg/Linde)? I hadn’t heard of these before.
Arguably, Bayesian decision theory (or a subjective, qualitative version of its logic) plays a legitimate role in the scientific method when there are empirical data on which to base prior and posterior judgements. As far as I can tell, the multiverse is really nothing more than an idea or a hunch – I wouldn’t even accord it the status of a hypothesis (see http://www.jimbaggott.com/articles/status-anxiety-all-theories-are-not-the-same/). In this regard, Carroll’s assertion is surely no more scientific than Pascal’s wager.
Jim,
The problem with using Bayesian reasoning for “theoretical” evidence (as Sean suggests) isn’t the priors. The problem is that the supposed evidence isn’t about the theory itself, but rather about scientists’ beliefs about the theory. Hence, if you use Bayesian inference to take into account non-empirical facts, you are changing the question from “How likely is X to be a correct description of Nature?” to “How likely do theorists think X is a correct description of Nature?” The latter is a quantifiable probability (and not even an entirely uninteresting one) but not actually the one you wanted to know.
This situation can get rather murky in theoretical physics because it’s an empirical fact that mathematics works very well. So if you use logical reasoning based on mathematics, that could count as an empirical argument. (Philosophers would disagree on calling it that way, but let’s not fight about words.) The problem is though that no proof is ever better than its assumptions. And if you look at what it is that enters these probability estimates, it’s primarily scientists’ opinion on how likely certain assumptions hold far beyond the regime where they’ve been tested.
The crux of what you are saying, Bee, is that “How likely is X to be a correct description of Nature? is not a scientific question. “Likely” has no place in science.
Bee,
See the last paragraph of
https://arxiv.org/abs/hep-th/0511037
I really mean “opinions” have no scientific value.
Peter,
Thanks!
Another Anon,
I don’t know if you did this on purpose or whether it’s a copy-and paste mistake, but this isn’t what I wrote. “How likely is X to be true?” can be a scientific statement (depending on what X is about), and in fact I would argue that all scientific statements are essentially statements of likelihood (you can never prove any statement about reality to be true). What I said was that statements about likelihood of beliefs of some people aren’t normally what scientists are interested in.
(You might actually want to read Sean Carroll’s recent book, which I reviewed here. Whether or not you like his opinion about the multiverse, he does a pretty good job explaining Bayesian reasoning and its relevance for science.)
Totally unsurprising. The Great Recession has taken its toll and a lot of these people have been hopping from temp job to temp job over the last 8 years. People wonder why the economy won’t pick up, and yet we’re talking about prime spending age people who don’t have the careers to support anything of the kind. Welcome to Japanification.
Bee,
I’ve actually never been a fan of Bayesianism, because of the difficulty of establishing priors even when there is an abundance of empirical evidence and because I believe it is virtually impossible to be objective. My point is really that all the proponents of Bayesianism that I’ve read argue for its utility in theory assessment in the light of evidence. I think (I confess I haven’t researched it properly) that Dawid and his colleagues are the only philosophers of science advocating Bayesianism in non-empirical theory assessment (see ‘String Theory and the Scientific Method’, pp. 68-72). Of course, this is a gift to theorists who want to dress opinion in the language of probability. Hence my remark about Pascal’s wager.
Whilst I’m mostly in agreement with your view, I do think it’s important to avoid what I think is a critical category error. In ‘Farewell to Reality’ I tried to distinguish between ‘correspondence truth’ and ‘coherence truth’. In science we take something to be true if it corresponds to the empirical facts – for this reason I’d regard general relativity to be ‘true’, on the understanding that it may one day be superseded by an even more general theory which establishes correspondence with even more facts.
But I can also establish the ‘truth’ of a network of mathematical relationships which may (or may not) have any relation with empirical facts. Now, establishing the coherence of a set of mathematical structures or a set of logical assertions might give me some confidence that these structures or assertions have something useful to say, but this DOES NOT COUNT as empirical evidence (sorry, this IS about words and I think it’s worth fighting for). I would go so far as to suggest establishing coherence truth is no real guide at all to correspondence truth – for this we need predictions (I’ll even settle for predictions of existing facts) and so far we have none. The history of science is littered with coherent structures that have nothing whatsoever to say about physical reality.
Another point that these (few) theorists seem to overlook is that new theories are adopted by the community only when they are shown to work better than the established theories they’re meant to replace. This isn’t even about making testable predictions – it’s about making testable predictions which demonstrate that the new theory (string theory or the multiverse) goes further or has greater explanatory power than the standard model of particle physics or big bang cosmology. My opinion (I can put a probability on it, if you want) is that any prediction, for example of some subtle effect in the cosmic background radiation, is likely to be accommodated by tweaking one or more of the many approximations theorists have to make in order to apply the established theories, and Occam’s razor will tend to favour the conservatives. In this view the proponents of new theories will have a real uphill struggle to make a convincing case, even if they’re ever able to make testable predictions. I’m afraid there’s just too much we don’t know and so too much flexibility, irrespective of how well the mathematics work.
“..and at Columbia University, Ph.D.s are taking classes in using Twitter to better communicate their work to nonacademic audiences..”
I think that courses teaching specific tools used in the prospective industries would be far more useful. What employers seek is practical knowledge and ability to solve problems using the tools they know. They usually don’t look for a person who would make for them goundbreaking discoveries.
Jim/Bee,
I appreciate the serious discussion of the issues of assigning probabilities, but maybe more relevant in this case is that the particular question here seems to me to not pass the laugh test (average expert on the issue thinks the idea is a joke). I take Weinberg’s comment about Rees/Linde not as a serious attempt to characterize the “probability there is a multiverse”, but as a joke making fun of the whole concept of doing this.
The serious issue is more that of, once you have Carroll, Dawid, Susskind giving prominent talks at this kind of conference, who is the joke on?
Sean Carroll believes the Multiverse idea is a middling obscure phenomena that indirect evidence could reveal.
Peter Woit and others (myself included) believe the Multiverse is a completely obscure hypothesized phenomena that no evidence (indirect or otherwise) can illuminate at all.
Peter, have you gone through Sean’s indirect evidence point-by-point to see why he believes this to illuminate in some way the Multiverse idea?
adamt,
I don’t know what specific point-by-point list of indirect evidence there is, but the arguments for the multiverse have been addressed I think exhaustively and ad nauseam here for more than a decade. Some of this is in the site FAQ, lots more if you poke around in the 125 postings here
http://www.math.columbia.edu/~woit/wordpress/?cat=10
I suppose I should put together a single location of serious arguments about this, but it’s hard to get motivated to do this, since my experience has been that the policy of Multiverse proponents is to steadfastly ignore the serious problems with what they are claiming, and instead attack straw men.
Modest economic expansion can happen while many industries (and perhaps the majority of the middle class) are tanking. The mean can become very different from the median. As conversations most of my long-suffering colleagues, would-be colleagues, and especially students (current and ex-) will attest, the reports of the end of the Great Recession have been greatly exaggerated.
Anon/RandomPaddy,
I’m not especially surprised by a negative median income trend, it’s just that the 6% change seems high over such a short period of time. I did a quick check and couldn’t find the real source of those numbers, may take a better look when I find a moment.
I also think Multiverse is a good theory… Just a philosophical one, not a scientific one. Trying to push it as if it were science is a disservice for humanity.
RE: decline in median income for Math/Physics mid-career Ph.Ds
That would be consistent with what’s happening to many other white-color occupations. This economy is being driven by the values and priorities of professional investors and Wall Street. They don’t want to sacrifice a portion of their returns to compensate well-educated, highly skilled workers any more than anyone else, if they can find ways to avoid it. That attitude is spilling over into areas beyond business, including the administration of colleges and universities.
Wouldn’t the puzzlement at this kind of discussion in conferences go away if we assume that such conferences have two parts: one in a serious scientific vein and one in a lighter, slightly jocular vein? I am not sure Sean Carroll for instance treats the multiverse with the same kind of seriousness he treats LIGO. I think it’s ok to put numbers on probabilities like this as long as you make it clear that you are engaging in philosophy and even humor and not science.
Chris W.,
I’d describe the trends I see differently, with some highly-skilled people doing well, but also fewer mid-range jobs, so more people ending up in poorly paid jobs. At universities, faculty at places like Columbia are doing well, and if you go into university administration your pay has soared dramatically. On the other hand, nationwide there are more teaching jobs being filled by poorly-paid adjuncts. From what I hear of the Silicon Valley world, Google is offering head-spinning sums to some people, and that’s probably true of other large successful technology companies. But a few such companies have gobbled up a large part of the economy, with not so much left over for those not working in these few places.
All in all, the pattern in industry after industry seems to be one of “winner take all”, with those doing well doing very well (and this is not just Wall street people), but fewer middle-class jobs. More and more cases are visible of middle-aged people who used to have a reasonably well-paid job which they lost as their industry was “disrupted”, and now are unemployed or working for much less than they used to.
This trend though isn’t new, it has been going on for years, and I’m still suspicious of the size of the claimed effect over 3 years.
Curious Wavefunction,
I’m sure Carroll knows the difference between serious experimental results like LIGO and speculative stuff like the multiverse, but I’ve never seen evidence that he isn’t quite serious about the multiverse, or has any particular sense of humor about it. When I first heard the Polchinski argument about “94%”, I was sure it was a joke, but it seems that he and others take this seriously. What I’m afraid is that conference organizers conceptualize this kind of speculation as, yes, different than solid science, but think of it as “inspirational”, as opposed to possibly silly.
I think the NSF survey on PhDs that you’re looking for is here:
https://ncsesdata.nsf.gov/doctoratework/2013/
Median salaries are here (table 53):
https://ncsesdata.nsf.gov/doctoratework/2013/html/SDR2013_DST53.html
To take a pair of data points (2013)
Mathematics/ statistics (11-15 years since PhD) median income: 109,000 standard error 6,000
Physics (11-15 years since PhD) median income 110,000 standard error 4,000
To look for the decline one would have to go to a previous version of this survey.
Here’s the 2010 survey:
https://ncsesdata.nsf.gov/doctoratework/2010/html/SDR2010_DST53.html
The 2010 pair of data points reads:
Mathematics/ statistics (11-15 years since PhD) median income: 100,000 standard error 4,500
Physics (11-15 years since PhD) median income 114,000 standard error 7,000
I’m sure other commenters will better be able to slice and dice this data. But on the face of it, it is physics PhDs, not mathematics/statistics PhDs that have had a decline in median income, for this range of experience.
This article presents some caveats in interpreting the NSF survey of PhDs for newly minted PhDs:
http://www.sciencemag.org/careers/2016/05/employment-crisis-new-phds-illusion
Anonyrat,
Thanks! Those numbers by age cohort have large “standard errors”, not clear whether they’re useful. The numbers for all ages are
Math
2003 80,000
2006 81,800
2008 95,000
2010 97,000
2013 98,000
Physics
2003 94,000
2006 99,900
2008 108,ooo
2010 110,000
2013 111,000
These aren’t inflation adjusted. Using a CPI inflation calculator I get, in 2013 dollars
Math
2003 101,290
2006 97,570
2008 102,790
2010 103,630
2013 98,000
Physics
2003 119,000
2006 115,440
2008 116,860
2010 117,520
2013 111,000
These do both both show a huge drop 2010-2013, when US median household income was going back up a bit after the recession, see here
https://en.wikipedia.org/wiki/File:US_GDP_per_capita_vs_median_household_income.png
An odd thing about these numbers though is that they show incomes going up significantly from 2006-8 to 2010, but this was the worst of the recession, with median US household income going down quite a bit. So, I’m still suspicious about these numbers, the 2010 ones are unexpectedly high, the 2013 ones unexpectedly low.
Not to rain on anyone’s parade, but those salaries (like mine) are quite a bit over median US incomes. And I enjoy what I do. I would rather complain about the ratio of the salaries of my plumber to that of my banker.
Is it possible that the multiverse is indeed string theory playing its end-game? The failure to make useful predictions about our own Universe has been turned into a strong indicator for a Multiverse. Perhaps even going so far, through this odd use of Bayesian statistics, as some sort of proof of a Multiverse.
But where to from here? String theory is no more likely to be able to say anything useful about these alternative realities than it is about our own, and as these alternates will almost certainly remain beyond any hope of our perceiving them then any conjectures will forever remain as conjectures.
Okay, a few more years tinkering with the probabilities of a Multiverse, but that’s about it. Pats on the back all around. Multiverse proved. But now what?
The only game in town is playing its final hand. And guess what, it’s all aces! Indeed 10^500 of them!
Peter,
Only indirectly related to today’s post, but would you have any article (either yours or others’) to recommend on the similarities between strings/multiverse and religion? And I don’t mean it as a parody.
Pingback: Probability of multiverse calculated! | Uncommon Descent
An outsized influx of young PhDs, or retirement of elderly PhDs would have the effect of driving the median salary of (all ages) down.
But yes, it looks like we have 1-sigma results only for 2010-2013.
tulpoeid,
I can’t recall ever seeing any such article and I haven’t tried to write such a thing. Religion is a very complex social phenomenon, and so is string theory/multiverse research. It might be possible to say some interesting things about the similarities and differences, but I haven’t seen that done.
Sean Carroll is an interesting figure here, he’s rather explicit about wanting to replace a religious world-view with one based on theoretical physics, with the multiverse playing a role perhaps closer to theological speculation than to conventional science.
Re God, Carroll and the multiverse. Carroll addresses some of these questions in The Big Picture, in particular Chapter 36 (Are We the Point?), where he posits his 50-50 (“perhaps”) credence figure, and to a lesser extent in Chapter 18 (Abducting God). In Chapter 36 he states that the multiverse is not a theory but a prediction of other scientific theories (string theory and inflation, which he admits are “speculative”), whereas God is not a prediction of any scientific theory.
has there been any discussion that instead of a multiverse, God did it, in terms of Bayesian statistics?
Curious
In Chapter 18, Carroll discusses whether there is any evidence to update our priors about divine creation (such as, why is there evil then?) and concludes that the concept of God is too imprecise and variable to apply Bayesian updating, leaving us just with our priors. I thought that was obvious all along. Peter is right in thinking that this sort of debate is a monumental waste of time outside of its entertainment value.
Not to try to drag things off topic, but my understanding is that such statements denote that rain is likely for 50% of the covered forecast area.
Narad, that works out to the same thing.
https://en.wikipedia.org/wiki/Probability_of_precipitation
I understand it’s a human compulsion, but why do we persist in thinking that our theories, which are essentially calculation methods to produce numbers that (one hopes) agree with experiments, have ANYTHING to do with reality? Go through history: every time we have thought, “well this theory(n-1) worked well enough at the time, but it was later seen to have nothing to do with the reality we *now* know due to theory(n).” Yet here we are with GR and the Standard Model, and whatever framework is supposed to succeed them, seriously talking about the probability that their implications exist. Well, a Frequentist actually has data, and that probability is Zero.
(I’m resisting the urge to claim proof by induction, but Raoul Bott defined induction as “You work out n=1, you work out n=2, you keep going until you’re bored. Then you have a proof.” I think the historical record satisfies his criterion..)
Perhaps I misunderstand their claim. Multiverse proponents are NOT, as far as I can tell, claiming probabilities for results of experiments. Or even probabilities that theory(N), for some large N, will make concrete predictions and involve multiverses, are they? But even then, theory(N+1) will show how silly they were to believe such a thing.
“Argument from ignorance.” Have none of these physicists heard of this fallacy? If you don’t know something, then all you can conclude from this is that you don’t know. So why the assenine guessing games about probabilities? Why the philosophical debates about accepting the reality of something without falsifiability? I don’t know, you don’t know, point.
Hi Peter,
Coindidentally I was about to write a blogpost about the Kinney paper, which is now online here.
@Jim Baggot
I think I agree with what you say. I should check out your book though to get my vocabulary right 🙂
The American Mathematical Society has some statistics.
The mathematics PhDs awarded by year 2003-2013 in the US are here:
http://www.ams.org/profession/data/annual-survey/2013Survey-NewDoctorates-Supp-TableD3-4.pdf
From there:
2008-09 : 1605
2009-10 : 1632
2010-11 : 1654
2011-12 : 1798
2012-13 : 1843
Median starting salaries for new PhDs 2000-13 in Teaching & Research, Academic Research Only, Government, and Business & Industry categories are available here:
http://www.ams.org/profession/data/annual-survey/2013Survey-StartSal.pdf
The median numbers look like this
Academic teaching & research (11-12 month)
2008: 56.0K
2009: 60.0
2010: 57.0
2011: 55.0
2012: 60.0
2013: 60.0
Research (11-12 months)
2008: 50.0K
2009: 50.0
2010: 51.5
2011: 55.0
2012: 55.0
2013: 54.0
Government:
2008: 81.5
2009: 82.5
2010: 80.0
2011: 70.0
2012: 82.0
2013: 87.0
Business:
2008: 90.0
2009: 90.0
2010: 90.0
2011: 94.3
2012: 95.0
2013: 100.0
I suppose 2011 was the year of government shutdown?
Anonyrat,
Wow, I haven’t looked at those numbers in a while. I can’t believe how many PhD’s there are now! I think the year I got mine (’92) there were around 800. And the job market sucked then.
@Another Anon:
Thanks for the reference. Perhaps I misinterpreted Neil’s original comment, but it’s neither here nor there in this thread, so I’ll just pipe down.
Regarding the probabilities of the multiverse, you are right that the ‘average expert on the issue thinks the idea is a joke’.
.
I’ve never seen even a sketch of a calculation for e.g. a Bayes-factor for the multiverse vs. universe (or even enough detail to begin a calculation), and suspect that such a calculation would prove deeply problematic.
.
Caroll et al simply seek refuge from accusations that their work isn’t scientific in Bayesianism, but their probabilities are just guesses.
.
However, whilst I’m glad you’re ridiculing their numbers, I hope it’s clear to your audience that there are many important legitimate applications of Bayesian statistics in science.
Jerome Ravetz in the Guardian, June 8
How should we treat science’s growing pains?
https://www.theguardian.com/science/political-science/2016/jun/08/how-should-we-treat-sciences-growing-pains