Eric’s Memes

Eric Winsberg teaches philosophy at the University of South Florida. He specializes in philosophy of science, in particular computer simulations in science. He wrote a book that may interest AT’s readers, called Philosophy and Climate Science. We met over teh tweeter a while ago. What follows is an edited transcript (Eric being disgraphic, English  not being my first language) of our chat about Montréal, memes, and music.

[Willard, hereinafter W]
hello, good sir
an update – i have an article to write on the spark joy meme before our interview
if it’s still in the cards

[Eric Winsberg, thereafter E]
hey. what did you want to interview about? what format, etc?

[W] it is for a climate blog
readers may want to buy your book

[E] yeah yeah, sure. anything to pimp the book. 🙂

[W] since we both like memes, i think running through your meme presentation would work like a charm – you have it in pdf?

[E] i think so. were you there?

[W] no, i only saw some tweets, and compliments about your meme game

[E] ah, ok. it was mostly on a dare. i dont know how to send files on twitter. whats your email?


[E] do you do the interviews over skype or by text.

[W] we do it here, simpler
ok, watched your slides
saw you made a point about emergent ECS
but why, what did you want to say to a bunch of philosophers listening to the talk

[E] is this the interview??
I thought the idea of emergent constaint reasoning was interesting.
but that it is easy to misuse.
and I thought I could use it to draw an interesting connection between two points people had been making about the epistemology of climate modeling:
stufff about robustness, and stuff about process understanding.

[W] what is this stuff about robustness, and process understanding, is it in your book?

[E] yeah, chapters 11 and 12.

[W] i guess i have no excuse anymore
[reading aloud the introduction] ok, jessica williams and chora – spouse and dog?

[E] girlfriend and late dog.
(died between when the book went to press and when it came out)

[W] i like that you thank feminist philosophers in the introduction for…

[E] pushing philosophy of science to be more focused on being in the service of social good

[W] i like the book structure – data, models, simulations, chaos, probability, confidence, decision, values, skill, robustness, diversity, social epistemology
looks like a book i could use to teach a course on philosophy of climate science

[E] that was the idea!!!

[W] you are transparent, it’s a Good Thing
your presentation is about the last section: “the fact that some climate hypotheses are supported by a variety of lines of evidence, and of the fact that some hypotheses are jointly predicted by a whole ensemble of different models.”
so, your robustness is what climateball players call consilience


[E] yeah, consilience in philosophy of science was kinda taken by theory confirmation lit. robustness has been more common for talking about agreement of models.

[more reading aloud]

[E] by the way: do I get to find out who you are? 🙂

[W] it’d be easy to find me out [inaudible]
never thought i’d meet philosophers when i started in 2009
otherwise i’d have chosen an even more obscure nick, like wilfrid or wilfred

[E] hahaha. i didnt even realize it was quine. quine is way better than sellars or tupac

[W] more so that if i get the point of your book right, in the end, holism wins
“in the end, holism wins” is something my character says a lot

[E] haha, dunno. maybe
didn’t think of it that way. but holism is good

[W] it may improve diversity – the more ways you can attack a problem, the better

[E] def

[W] let’s connect with what you want to say with other slides
were you following your book closely?
there is “model robustness” and “instrument robustness” – what was your point?


[E]  the main point was: if you want model robustness, you need emergent constraint reasoning, and once you see that you see that what knutti calls process understanding, and contrasts with robustness, is actually what you need *for* robustness.
they in fact go hand in hand

[W] robustness and process understanding, alright

[E] later
always great to talk to a montreal, especially a montrealer.


[W] hello again, if you have time, i have a clarification to ask
Andrew ranted about emergence a few days ago
i sense an equivocation needs to be clarified

[E]  Emergent Constraint [EC] reasoning is great, but its also had to do well.
I actually wrote a piece in the Frankfurter Allgemeine Zeitung criticizing an EC paper in nature. I’ll email you the real article and also the english version.
dessler is absolutely right that you can’t use EC reasoning to get right at climate sensitivity

[W] i like that

If the degree to which they cancel out changes as the climate does, the apparent emergent constraint could be spurious, and provide false confidence. What does this mean for the new uncertainty estimate in the Nature letter? The IPCC gives about a 20% probability of ECS falling above 4.5C. The Nature letter says 1%. To really be confident in this, we need to be at least 99% sure that the alleged emergent constraint is not founded on compensating errors. Given the vast number of processes and feedbacks that contribute to ECS, it is hard to see why we should think the likelihood of this is anywhere near that low.

[E] so the idea that from consilient lines of evidence emerges robustness does not imply we go all in in emergent constraint analyses
yeah, so this last point is complicated and might be hard to make in a pithy line or two.
but the idea is: consilience, properly understood, isn’t cheap.
you need a lot of background knowledge to understand when you are accumulating more and more evidence and when you are looking at multiple copies of the same newspaper. and EC reasoning is similar.

[W] that meme is my favorite
if we could rule out alternative explanations, that’d be great
rule out alternative explanations with RA, i.e. Robustness Analysis ?


[E] right. that’s the schupbach account of RA, that each new addition to the set, in order to add to robustness, should rule out a possible explanation of why the old set might have erroneously indicated that the hypothesis was true.

[W] you believe in that?

[E] ben ouais

[W, flipping] i see that you do on p. 205
oh, you rule out ECS under 1.5C because of volcanos, solar cycles, and paleo
(and i know at least one reader who will like this)
but we can’t rule out >6C
and you have no confidence that we can constrain these results more than we already do

[E]  not right now, no
i think >6 is unlikely, but can’t be ruled out

[W] it’d be foolish not to allow the possibility as a swan-like event
the other end requires we revise physics, or we get really strange cloud effects

[E, in a more professoral tone] The main difference between the left tail and the right tail is that we know that low ECS correlates with long equilibration time. So if it were below 1.5, we would have seen it equilibrate at a low value. But if it’s over 6, there are reasons we might not have seen it good. What the Schupbach account shows us, is why we can’t rule out >6 we don’t have a set of detection methods that each collective rule out something the other don’t have, because the main way to rule out >6 is with really long data, and really long data is also full of errors. So nothing will rule out the possible explanation that it takes a really long time to see the >6, and that’s just being hidden in the messy really old data. The good news is ecs is probably not >6 and if it is, it will almost certainly take centuries to get there.

[W] ok, i think we got everything, but we can’t end like this
we got Eric, the montrealer and the memes, we need the music
why tupac, like here for instance

[E] I don’t like tupac a) because he got biggie killed and b) he has no flow

[W] flow, who has flow

[E] best flow: biggie. other contenders: method man (he’s lazy though and not many great verses–try his verse on “shame on a n..”), Big L, and Snoop Dogg, Ghostface Killah on “Ice cream” is also great flow.

[W] wow, so much to listen to now
alright, that’s a rap
thanks for your time
nice to see that philosophers are stepping in

[E] ok, ttyl


About Willard
This entry was posted in Interview, The philosophy of science, We Are Science and tagged , , , , . Bookmark the permalink.

37 Responses to Eric’s Memes

  1. Steven Mosher says:

    somebody dissed rhapsody in blue?

    where is the clown I havent done a beat down in ages

  2. BBD says:

    oh, you rule out ECS under 1.5C because of volcanos, solar cycles, and paleo
    (and i know at least one reader who will like this)


  3. The emergent constraint comment is interesting. I’m not entirely sure I follow, but I think the argument is that even though we don’t tune for something like the ECS (it emerges) we should be careful of suggesting that because it emerges provides some kind of validity to these estimates. I think I agree with this, but do think it’s still important to make clear that climate models are not specifically tuned to produce a particular ECS.

  4. Willard says:

    > it’s still important to make clear that climate models are not specifically tuned to produce a particular ECS.

    Indeed. I think that Eric’s overall point is simply that robustness, diversity, and process understanding are all connected. The properties that emerge from simulations adds to the robustness only if it improves our understanding. The more modulz we build in various ways, the merrier. If we only seek emergent properties for the fun of getting published, as Andrew suggested, it can lead us astray.

  5. Willard says:

    Here’s where Andrew echoes Eric’s overall point:

    As I see it, diversity hinders mining for specific properties. Varying quasi-experimental conditions is the main (perhaps the only) way we have to distinguish artifacts from our apparatus and results about the system under study. Simulations are supposed to improve our understanding, not replace it.

    To that effect, I should add a follow-up question that I could not fit in the chat. We could be tempted to think that robustness analysis would replace experts’ judgment (like the one we see in the summary of the IPCC’s reports), as it is one way to estimate various lines of evidence. Eric argues against the idea, as the tool isn’t meant to replace our understanding, but to make sense (and eventually formalize) our intuition of consilience.

  6. Joshua says:

    Willard *

    that’s a rap…

    Can I assume pun intended?

  7. dikranmarsupial says:

    Building models to help us understand things? Seems somewhat familiar!

  8. Willard says:

    > Can I assume pun intended?

    Yes, you can. That chat ain’t rap, but we tried to make it flow.

  9. This is a reply to the comment from ATTP:

    The worry is not that the models are tuned to ECS. I know that’s not the case. The worry is really the opposite: that ECS is a highly emergent parameter, and that therefore so will be any seasonal analog of it. (Emergent constraint reasoning is all about seasonal analogs). But because of that, the degree to which a cluster of models is getting right that seasonal analog *might be due to compensating errors***. And if they are, then the fact that they are clustering around some true value might not project well to, say, end of the century projections. TBH, it was a very complicated topic to try to get through in an interview, but I think Willard wanted to ask about my PSA presentation because of the memes, and that’s what it was on.

  10. @ATTP:

    “I think I agree with this, but do think it’s still important to make clear that climate models are not specifically tuned to produce a particular ECS.”

    Absolutely. In fact the point is almost the opposite. The problem is that ECS is so emergent in the climate models. Moreover, it emerges as the result of a large number of highly interacting processes, many of which are parameterized. When you do emergent constraint reasoning (ECR), you find a seasonal analog of the climactic variable you are interested, and you use the amount of bias between your models and the observable seasonal variable to estimate the uncertainty in the climactic variable. This should only be done, I am arguing, if you know that the climactic variable and the seasonal variable come from one physical process. because if not, two or more responsible process might be badly represented in ways that cancel out. If that happens, they might cancel out in a good way wrt the seasonal variable, and not with respect to the climactic one. If that happens, your estimate breaks down. In the talk, I used the example of snow/ice albedo feedback. This is also an emergent variable. but it emerges from (mostly) one physical process: melted snow reduces albedo. So if your models are getting seasonal snow albedo to within some error bar x, and the seasonal and climactic values have a mathematical relationship given by the function f, then you know that seasonal snow abledo is accurate to within f(x). But this only works because you know (reasonably surely) that there are no canceling errors. you simply cant do this with ECS. its too complex. you need to break down ECS into all the separate physical processes that give rise to it, and try to ECR on those. and then sum them alll up. A very difficult task but one worth working through even its never going to be fully achieved.

  11. Steven Mosher says:

    for you white guys flow is basically using your voice(words) in a drum beat pattern.

  12. For me, flow isn’t just doing that, but doing it in an interesting and non-repetitive way. A good rap verse should be such that if you just read the lyrics, it should be hard for you to figure out how the rapper can make each line fit into the same space until you hear it. Tupac and most Nas lyrics have obvious meter when you see them written on a page, because they both tend to repeat the same rhythm over and over.

  13. I dare anyone who doesn’t know this song to figure out how its supposed to flow:

    Yo RZA, yo razor, hit me with the major
    The damage, my Clan understand it be flavor
    Gunning, humming coming at ya
    First I’m gonna get ya, once I got ya, I gat ya
    You could never capture the Method Man’s stature
    For rhyme and for rapture, got niggas resigning, now master
    My style? Never!
    I put the fucking buck in the wild kid, I’m terror
    Razor sharp, I sever
    The head from the shoulders, I’m better, than my competta
    You mean competitor, whatever, let’s get together

  14. Steven Mosher says:

    talk to the triplet.

  15. Eric,
    Thanks for reply. That is more complicated than I had realised. I think I see what you’re saying, but I’ll have to give it some more thought.

  16. Eric,
    Has anyone done an analysis that tries to work out how likely it is that two, or more, physical processes might operate in such a way that their errors cancel to make the seasonal variable behave as expected, while resulting in the climatic variable not behaving in a way that properly represents that aspect of the system? I can see how this is possible, but it might also be reasonably unlikely.

  17. Dave_Geologist says:

    I can think of a gas-field analogy: water coning vs. aquifer influx.

    Water coning is local to the near-well region, and as the name implies is a cone of water drawn up from the aquifer, to and perhaps into a producing well (assume for simplicity that it’s a vertical well). It’s in response to a very high pressure differential (many thousands of psi) imposed over a small area in a short time-frame. It’s reversible: when the well is shut in, the water slumps back down. If you have a uniform reservoir with multiple similar wells, you may see many wells behaving in much the same way. That’s your seasonal analogue: the well will often be shut in during the low-demand season while non-coning wells or fields carry the load, or intermittently when water-loading kills the well or water production exceeds processing capacity.

    Aquifer influx is field-wide, and is driven by the average pressure differential between the depleting gas field and the underlying aquifer. It’s slow, unidirectional and continues even when all the wells are shut in. That’s your decadal analogue.

    If you didn’t understand the geometry and processes, and assumed that the observed rise and fall of water in the wells was field-wide not local, you’d make totally wrong predictions about the speed, extent and reversibility of aquifer advance. That’s an extreme thought-experiment, but I know of real-world examples where we did understand the geometry and process, and it still took years to untangle local from field-wide.

  18. Eric Winsberg,
    Since you have written extensively on the topic, what’s your impression on the vast division between the complexity of computer simulations and the relative simplicity of analytical mathematical formulations for describing certain climate behaviors?

    I can give you several examples of this but want to hear your untainted take first. BTW, this also has implications for the canceling of errors that you are talking about.

    Incidentally, there is a software development analog for this, as debugging software with compensating errors is one of the most time-consuming activities to be encountered. The traditional way of debugging — making slight mods to the code — always move the operating state to a worse condition, and only be making changes in synchronization is the bug revealed.

  19. Chubbs says:

    I’d like to see an emerging constraint study of all the emerging constraint studies i.e. what happens when model performance is improved: 1) random effects depending on the performance factor or 2) does improved performance usually point in one direction when performance improves. I have a feeling it is the latter but would like to see all the studies analyzed together.

  20. @ATTP,

    I think Dave’s comment is right. But also: really all you have to think of is, lets say, a feedback that depends strongly on at least two parameterized processes. Now imagine these are both parameter values that are hard to estimate from physical intuition, so they are tuned to real observations. Imagine also that its well recognized that this tuning is partly cancelling out errors that dont come from the two processes being parameterized—like maybe top of the atmosphere radiation loss. It wouldn’t be very surprising to find out that they were better tuned for observed regimes (present day or recent past) than they were for future unobserved regimes. Maybe someone who knows the modeling better than me can chime in with real examples, but I think there’s lots of anecdotal evidence that the models can and sometimes do behave this way.

  21. Willard says:

  22. Willard says:

    > Maybe someone who knows the modeling better than me can chime in with real examples

    Gavin, via DM, offers:

    It’s hard to pinpoint precise interactions, but it’s evident that models with equally reasonable skill in seasonal cycles, have very different responses (and skill scores) to interannual variability or trends. Try Santer et al 2009: figure 3 for instance.

    Here’s figure 3:

    Caption reads:

    Relationship between model skill in simulating the mean state and skill in simulating the annual cycle pattern (A), amplitude of monthly variability (B), and monthly variability pattern (C). Results plotted are the “average errors” shown and described in Fig. 2. The black lines are the fitted least-squares regression lines. Models to the left of the vertical dashed gray line are ranked in the top 10 based on values of the mean state metric α̂. Models below the horizontal dashed gray line are ranked in the top 10 based on values of the annual cycle pattern metric β̂ (A), the variability amplitude metric φ̂ (B), and the variability pattern metric ϕ̂ (C). The gray shaded region indicates the intersection of the two sets of top 10 models plotted in each graph.

  23. Thank you Gavin! (And Willard!)

  24. Willard says:

    To return us to more flowy matters, here’s Chilly Gonzales the musical genius, who like Gavin was trained at McGill:

    There’s a cat playing keyboard at 2:15, so you got no excuse.

  25. Eric,
    Thanks, that makes sense.

    [Edit: I should have added that from Gavin’s example (that Willard posted) it seems clear that this has been tested and that it does show quite nicely that skill in one context doesn’t imply skill in another context.]

  26. Joshua says:

    Since I can’t understand the topic of the post, I’m trying to see if I can understand the topic of flow. If I got it right, this might be an example of white flow:

  27. Willard says:

    It’s as if scientists know what they’re doing:

    The whole thread is good.

  28. Steven Mosher says:

    Err Joshua.

    easiest way to understand it is that the words replace the drum kit. The human voice as drum.

    then, you can start to do boring things with that ( a boring drum) or you can do interesting
    things with that. So you have the rhythm or prosody of the lyrics as we would speak them
    ( take imabic pentameter) now that accent pattern can align with the musical beat naturally
    ( an you get something predictable and boring after a while–singsong like) OR you
    can change the verbal stress pattern to align with the musical beat pattern OR you can
    create tension between the natural verbal stress pattern and the metrical beat.

    this is accessible

  29. Steven Mosher says:

    “It’s as if scientists know what they’re doing:

    twelve maps of sea-surface temperature anomalies. Six are real and six are samples from a multivariate normal distribution. But which are which?”

    Ya, when I saw Rohde was playing I just quit.

  30. Listen to snoop deeohdoublegee. he knows what he’s talking about.

  31. Steven Mosher says:

    As someone who transitioned from physics and math to Philosophy and English
    I found this quite fascinating

    also funny was the difficulty Eric had reading his own handwriting and spelling.. DOH!

    “The first philosophy class that I took at college was part of the common-core requirement in the humanities. It centered around Plato’s dialogues, some Hume, and a few other things. I did pretty well in it. I was quite surprised by this. I think it was due in part to fact that we studied the Socratic dialogues. It was easier for me to sit still and read them than most of what we had read in high school. ”

    same experience up the road from Eric at Northwestern. core requirement, Plato and Hume in one class, ah Descartes too.

    Gavins example has helped me understand the argument. seems like I should get Eric’s book

  32. Steven Mosher says:

    that’s a good video eric.

    you might like this guy

  33. Willard says:

    Readers may like NPR’s Tiny Desk Concerts. I liked Mac Miller, Anderson .Paak & The Free Nationals, August Greene, Big Daddy Kane, Rakim, Big Boi, Wyclef Jean, and the one and only Wu-Tang Clan. With violins and all.

    Pursuing the Montréal vibe:

  34. Pingback: The Hawkmoth Effect | …and Then There's Physics

  35. Nathan Tetlaw says:

    Or… they may like “The ‘Gurge”

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.