@irl_danB Like this? https://t.co/3ztUr0GiVb It’s also useful to point out, next token prediction is - a kind of Solomonoff Induction (universal brute-force probability distributions), and - the purpose of the Bellman equation (ideal reinforcement learn
@matspike This is not *not* that https://t.co/As5NJKqbQD
@gwplaw @kareem_carr @SamiKohvakka_ I'm just gonna leave this here https://t.co/As5NJKqJGb
First reference Gigerenzer & Selten (2002). Second reference describing advantages of ecological rationality here. https://t.co/YMCMG0Lsch And @delong lists are best next-up algo: which is better decisions under Knightian uncertainty—the map or territ
https://t.co/WNAq6jCzoD “…the causal structure of some systems cannot be fully captured by even the most detailed microscopic model”
@logopunk @marielgoddu I keep coming back to this, and some day I may finally understand it https://t.co/rg2EleMhX1
RT @TheAnnaGat: LOVED LOVED LOVED being on @erikphoel's podcast. (Yes, he's starting one. No, it's not like the others.) The 2016 paper w…
RT @TheAnnaGat: LOVED LOVED LOVED being on @erikphoel's podcast. (Yes, he's starting one. No, it's not like the others.) The 2016 paper w…
RT @TheAnnaGat: LOVED LOVED LOVED being on @erikphoel's podcast. (Yes, he's starting one. No, it's not like the others.) The 2016 paper w…
LOVED LOVED LOVED being on @erikphoel's podcast. (Yes, he's starting one. No, it's not like the others.) The 2016 paper we're referencing, which back then made me so interested in Erik's work, is here: 'When the map is better than the territory' https:/
@anoncept @aucksAdventures @matspike @marielgoddu I keep forgetting: Erik Hoel's paper here which I will some day have the math to fully grok AIUI starts to lay some of the mathematical groundwork for what we mean when we talk about emergence https://t.co
@RIngthorss @PessoaBrain There is infinite regress, but some levels have higher predictive power than others (drawing on Hoel, https://t.co/As5NJKqJGb)
@PessoaBrain I tend to lean heavily on Hoel when I think about this (who I, like, only half-understand) https://t.co/As5NJKqJGb
@AlanzosBlog @bayesianboy Eventually i will be good enough at math to know if this is nuts or not https://t.co/As5NJKqbQD
@nosilverv A lot. Shoulda linked the paper, read the abstract https://t.co/QLdQGfuEmH
@defnotbeka @againstutopia Somewhere between work like this and all the AI nonsense right now, we are going to start to characterize this—but to your point, it's really only something we could justify within the last few years https://t.co/As5NJKqJGb
RT @architectonyx: @eigenrobot good read! i agree with most points, tho i am a hard platonist so there is naturally a little disagreement…
@eigenrobot good read! i agree with most points, tho i am a hard platonist so there is naturally a little disagreement i thought you might be interested in this paper by @erikphoel : https://t.co/lbBjL6SaBJ
@doriantaylor Just gonna leave this here: https://t.co/As5NJK8As3
RT @AdamCElwood: @idearadar @algekalipso Also, scale is important for causal power. For example, see this really interesting work from @e…
RT @AdamCElwood: @idearadar @algekalipso Also, scale is important for causal power. For example, see this really interesting work from @e…
@idearadar @algekalipso Also, scale is important for causal power. For example, see this really interesting work from @erikphoel https://t.co/KKaRrI2M50
@norootcause More causal capacity! https://t.co/As5NJKqJGb
@myrrlyn And sort of riffing on this paper I keep coming back to https://t.co/D1U4A2KhwO
@lorenlugosch I’ll list two of my favorites (though the latter one qualifies on an honorary basis because it does have multiple formats): https://t.co/a1AFTiYv2b https://t.co/asNPnr90ph
@ClaireJHartnell @HiredThought The flip side is, at least if I understand this paper correctly and it bears out, that that level of understanding isn't always necessary or useful to predict the future behavior of the system https://t.co/As5NJKqbQD
This notion of using channel capacity from Claude Shannon's original information theory as causal emergence for treating macroscale as a "map is better than the territory" is compelling. I feel as if I should try to poke holes in it and yet, I like it...
Map and Territory
@adereth @palvaro About that... https://t.co/As5NJK8As3
@mathiasverraes @madplatt And I’ll reference the paper I always reference here, which I confess I don’t fully understand, but what I do get provides tantalizing glimpses at https://t.co/As5NJK8As3
@mcbcmcbcm If you're interested in the technical details of the accounts of causal emergence I am going to discuss, then these three: https://t.co/BoyEi3lw6B, https://t.co/FFv8Bq5nh3, https://t.co/KJH8behLuH. For the general approach to causation I want to
@rmcomplexity @IanColdwater Erik Hoel has written a *fascinating* paper on this which I'm slowly trying to learn enough information theory to understand https://t.co/As5NJK8As3
@lhochstein Might be related to that Hoel paper I'm constantly referencing; in the limit the system has an incredibly large number of states, but at the right abstraction level the system is tractable https://t.co/98MsFsqPFi
What an awesome title for a paper https://t.co/DwEhQyoJv5
@Aelkus Whoops, forgot the link https://t.co/As5NJK8As3
@yashaddad_ Three examples I am working on right now: https://t.co/bMvqAYodkq https://t.co/BoyEi3lw6B https://t.co/KJH8behLuH
RT @_onionesque: "A macro model of a system (a map) can be more informative than a fully detailed model of the system (the territory). This…
RT @_onionesque: "A macro model of a system (a map) can be more informative than a fully detailed model of the system (the territory). This…
"A macro model of a system (a map) can be more informative than a fully detailed model of the system (the territory). This has been called causal emergence. [...] this paper grounds the phenomenon in a classic concept from information theory" https://t.co/
@dakami @SwiftOnSecurity Yuuuuuuuup. All abstractions necessarily lose information. In fact that's what makes them so powerful! Erik Hoel has done some *extremely* provocative work trying to justify this information theoretically which I wish I understood
@irrideo Tô lendo esse aqui agora ó: https://t.co/zAqlNLIWej
There will *always* be information in the physical object that no virtual representation can hope to capture, only to approach. And if Erik Hoel's work is right, the most important information is *harder* to work with below certain levels of abstraction ht
@allafarce (Which I'm basing a lot on this paper: https://t.co/As5NJK8As3)
@Aelkus Hoel starts to explore why some models might be more useful (meaning, predictive) than other though: https://t.co/As5NJK8As3
@ThosVarley @smellosopher There is something interesting mathematical work being done on emergence by Erik Hoel and colleagues (see e.g. https://t.co/m1BvzdRZ8p, https://t.co/FFv8Bq5nh3, https://t.co/J2E0hBgXZV), which I think could have useful philosophi
@context_ing Have I linked you this yet? https://t.co/As5NJK8As3 It attempts to provide an information theoretic answer
@mutual_ayyde @againstutopia I linked this on your older thread, but I'll link it here too https://t.co/As5NJK8As3
@mutual_ayyde You might find https://t.co/As5NJK8As3 interesting
@litgenstein @bayesianboy @lastpositivist @erikphoel Yes! That's also how I discovered him and I've been stanning and weirding him out ever since. for others, this is what litgenstein is referring to: https://t.co/Gi70reFAEB Later published at entropy: h
@kochseb @JacksonKernion I assumed it was about something like this - https://t.co/fWNQuOXMON - how informative/effective intervention is at different levels of detail, but I wasn't sure!
@context_ing @againstutopia And on that topic, I don't have quite enough background to fully understand this paper, but I find what I *do* understand of it extremely interesting https://t.co/As5NJK8As3
@janellekz @simonbrown here's a fun thing to think about related to this https://t.co/PhNGhCIoRC
@marick @casio_juarez While I look, can I interest you in this also very interesting paper on a related topic? https://t.co/HCFgv5IwuP
se isso n é profundamente relevante n sei mais o que é https://t.co/ti24vMuzNG estudos nessa linha podem transformar um debate filosofico ("holismo" vs individualismo") numa questao empirica => qual unidade/nivel de analise é melhor para qual questão
@ReinH I’m not sure that’s true. There are intriguing hints that emergence is actually an information theoretic phenomenon https://t.co/As5NJK8As3
@freerecall @EPoe187 @erikphoel haha I suspect I would! My personal views on multi-level causality is from Erik Hoel (also author on the actual causation paper) https://t.co/nQsjSK3yTF Importantly, I think the a priori account isn't relevant to the nature
RT @jkbren: Almost 2 years ago to the day, @erikphoel and I met in NY to chat about his terrific paper on causal emergence https://t.co/ZNB…
RT @jkbren: Almost 2 years ago to the day, @erikphoel and I met in NY to chat about his terrific paper on causal emergence https://t.co/ZNB…
RT @jkbren: Almost 2 years ago to the day, @erikphoel and I met in NY to chat about his terrific paper on causal emergence https://t.co/ZNB…
RT @jkbren: Almost 2 years ago to the day, @erikphoel and I met in NY to chat about his terrific paper on causal emergence https://t.co/ZNB…
RT @jkbren: Almost 2 years ago to the day, @erikphoel and I met in NY to chat about his terrific paper on causal emergence https://t.co/ZNB…
RT @jkbren: Almost 2 years ago to the day, @erikphoel and I met in NY to chat about his terrific paper on causal emergence https://t.co/ZNB…
RT @jkbren: Almost 2 years ago to the day, @erikphoel and I met in NY to chat about his terrific paper on causal emergence https://t.co/ZNB…
RT @jkbren: Almost 2 years ago to the day, @erikphoel and I met in NY to chat about his terrific paper on causal emergence https://t.co/ZNB…
RT @jkbren: Almost 2 years ago to the day, @erikphoel and I met in NY to chat about his terrific paper on causal emergence https://t.co/ZNB…
RT @jkbren: Almost 2 years ago to the day, @erikphoel and I met in NY to chat about his terrific paper on causal emergence https://t.co/ZNB…
RT @jkbren: Almost 2 years ago to the day, @erikphoel and I met in NY to chat about his terrific paper on causal emergence https://t.co/ZNB…
RT @jkbren: Almost 2 years ago to the day, @erikphoel and I met in NY to chat about his terrific paper on causal emergence https://t.co/ZNB…
RT @jkbren: Almost 2 years ago to the day, @erikphoel and I met in NY to chat about his terrific paper on causal emergence https://t.co/ZNB…
RT @jkbren: Almost 2 years ago to the day, @erikphoel and I met in NY to chat about his terrific paper on causal emergence https://t.co/ZNB…
RT @jkbren: Almost 2 years ago to the day, @erikphoel and I met in NY to chat about his terrific paper on causal emergence https://t.co/ZNB…
RT @jkbren: Almost 2 years ago to the day, @erikphoel and I met in NY to chat about his terrific paper on causal emergence https://t.co/ZNB…
RT @jkbren: Almost 2 years ago to the day, @erikphoel and I met in NY to chat about his terrific paper on causal emergence https://t.co/ZNB…
RT @jkbren: Almost 2 years ago to the day, @erikphoel and I met in NY to chat about his terrific paper on causal emergence https://t.co/ZNB…
Almost 2 years ago to the day, @erikphoel and I met in NY to chat about his terrific paper on causal emergence https://t.co/ZNBVvq0dhy Since then, we’ve been building a formalism to study causal emergence in networks. Today we posted our first paper on it
RT @neuropoetic: "The map is not the territory" - Alfred Korzybski https://t.co/zV894E8Amk "When the map is better than the territory" - @…
RT @joe_dewhurst: Tyler Millhouse describes an approach to computational mappings based on Kolmogrov complexity (https://t.co/aPZ6K4Cp55),…
Tyler Millhouse describes an approach to computational mappings based on Kolmogrov complexity (https://t.co/aPZ6K4Cp55), and my current feeling is that the approach to causal emergence described by Hoel might also be closely connected to real patterns (htt
RT @neuropoetic: "The map is not the territory" - Alfred Korzybski https://t.co/zV894E8Amk "When the map is better than the territory" - @…
"The map is not the territory" - Alfred Korzybski https://t.co/zV894E8Amk "When the map is better than the territory" - @erikphoel https://t.co/oKx67tsfRd
@davidsarac @trishankkarthik @HundredthIdiot @normonics There are quite a few people doing some interesting work in this area. I mention the map-territory issue (where a 'map' isn't the thing you hold in your hand as Disneyland) because this paper thinks a
@richardhuskey Maybe you have *too much* information! Coarse-graining seems promising for many applications, though tricky to get right. Something I've thought about but never applied. Random things I've read that inspired: https://t.co/tWEgjISSxs https://
@richardhuskey On scale-crossing: More conceptual: https://t.co/Bh4fMXyeAy More math-y: https://t.co/EGL8DEDQWa Somewhere between: https://t.co/J9ViyedMkb My attempt to translate to comm: https://t.co/dYeKXDKwq0 2/2
@gershbrain @samuelmehr @erikphoel 's work would be relevant - https://t.co/8766u2ghR4 But also check out counter-arguments - https://t.co/c8EJBAvBsd
A very interesting paper suggesting "that causal #emergence is related to a classic concept in information theory, Shannon’s channel capacity, thus grounding emergence rigorously in another well-known mathematical phenomenon for the first time." https://t
RT @mattsiegel: reposting because it's cool "a macroscale description of a system (a map) can be more informative than a fully detailed mi…
i gotta read this paper, bc skimming is not helping me see why Crutchfield’s focus on information channels in his computational mechanics doesn’t figure more largely than the brief mention in the paper indicates. eg https://t.co/meCi21wCys
RT @mattsiegel: reposting because it's cool "a macroscale description of a system (a map) can be more informative than a fully detailed mi…
oh this is rad
RT @mattsiegel: reposting because it's cool "a macroscale description of a system (a map) can be more informative than a fully detailed mi…
RT @mattsiegel: reposting because it's cool "a macroscale description of a system (a map) can be more informative than a fully detailed mi…
RT @mattsiegel: reposting because it's cool "a macroscale description of a system (a map) can be more informative than a fully detailed mi…
reposting because it's cool "a macroscale description of a system (a map) can be more informative than a fully detailed microscale description of the system (the territory). This has been called “causal emergence.” paper: https://t.co/7Ki8EUttIi
@qirennia i'm gonna say yes but i don't like that definition (and here's some work i like much better :) https://t.co/iZTTWDphC9
@hillelogram meaningful effect. Cf., as always, https://t.co/rg2Elf3kZ1