nucleation

LIT-1038: Quantifying causal emergence shows that macro can beat micro

3 min read

Properties7
StatusActive — read in full 2026-10-13 (NOTE-726) from the PubMed Central PDF of the published article (PMC3856819, pmc.ncbi.nlm.nih.gov/articles/PMC3856819/pdf/pnas.201314922.pdf), fetched through the Internet Archive capture of 2022-11-13 because PMC answered scripted requests with a reCAPTCHA page and pnas.org returned 403; the online Supporting Information was not read. Worth reading as the source of effective information as a measure of causal emergence: it shows exactly when a coarse-grained description of a known mechanism can carry more EI than the micro, and its macro advantage rests on imposing a uniform intervention distribution at each level. For QUESTION-093 it is an interventional criterion met with no closure and no self-individuation; it complements CLAIM-154 and THEORY-133 and is neither a case nor a rival of them; it measures a different object from the information-decomposition account of LIT-025.
Tagscausality, complex-systems, information-theory
Cited as “extends” byLIT-1129 — Dynamical Reversibility and A New Theory of Causal Emergence based on SVD, LIT-1136 — Causal Emergence 2.0: Quantifying emergent complexity, LIT-1843 — Causal emergence is widespread across measures of causation
Cited as “grounds” byARG-041 — From the effective-information bound and its ties to proportionality: a gain that depends on the intervention distribution is not a property of the system, the coarse-grainings that lose nothing are, and the coarsest of them is where the causes are
Cited as “paper” byNOTE-726 — Hoel, Albantakis & Tononi — Quantifying causal emergence shows that macro can beat micro
Cited as “rivals” byLIT-1833 — Multi-Level Cause-Effect Systems
Cited as “source” byTHEORY-1167 — When one intervention distribution over micro states is used at both levels, the effective information of a coarse-graining of a discrete mechanism, with each macro intervention realized by that distribution within its block, cannot exceed the micro effective information; so a positive causal emergence of effective information, scored with a uniform distribution at each level, equals the micro gain from changing the distribution minus a non-negative loss, and is a gain from the change of distribution, THEORY-222 — With uniform interventions imposed separately at each level, a coarse-grained macro description of a known discrete mechanism can have more effective information than the micro description it supervenes on, so the level at which effective information is highest can be a macro level, THEORY-787 — A system's causal structure can be most effective at a macro scale, so that causation emerges at a coarse-grained level and the macro description is the one at which the system's causes are found, as Hoel, Albantakis and Tononi propose from effective information

Graph View

Hoel, E. P., Albantakis, L., & Tononi, G. (2013). Quantifying causal emergence shows that macro can beat micro. PNAS 110(49), 19790–19795. DOI 10.1073/pnas.1314922110

Key takeaways

Read in full (NOTE-726), main text only; the Supporting Information was not read.

  • EI is effectiveness times size. Effective information is the mutual information between a uniform distribution of interventions on a system’s states and the resulting next states. Divided by log₂ n it is effectiveness, which is average determinism minus average degeneracy.
  • Causal emergence is a difference in EI. CE = EI(macro) − EI(micro), and it splits exactly into a gain in effectiveness and a loss in size. The size term is always negative, so the macro wins only when coarse-graining makes the mechanism much more deterministic or much less degenerate.
  • Four constructed examples (noise, degeneracy, time, space and time) give 0.40, 0.57, 0.62 and 2.92 bits. Systems already near maximal effectiveness at the micro level cannot show emergence.
  • The macro is scored under its own uniform interventions, each macro state set by averaging over its micro realizers. The authors say this is what lets the macro win. It is a choice, and the paper does not try another.
  • Interpretation. The EI-optimal level is called intrinsic, and the macro, when it wins, is said to exclude the micro, turning Kim’s exclusion argument around. The claims about brains and the sciences are conjecture.

Why this is here

A7 and A11 cited it for when a coalition deserves a higher-level operator description. The filing’s summary was right about the result. It left out the condition that carries it, the uniform intervention distribution imposed separately at each level, and it said “causally privileged” where the paper shows only that EI is higher.

What the reading found, for the codes it was filed for:

  • QUESTION-093. EI is defined by interventions, so CE measures interventional autonomy, with predictive autonomy only for predictions from imposed states. The paper’s systems are fixed logic gates with no self-maintenance, and the coarse-graining is an analyst’s search, so the first criteria can be met with neither closure nor self-individuation. The paper’s “self-definition” is a figure of speech for the EI maximum.
  • CLAIM-154 and QUESTION-040. The same shape of trade-off, a gain against a size cost always paid, but for a known mechanism, with no data, prior or posterior. It complements the reification rule and is not a case of it. With an estimated mechanism, the maximum over many coarse-grainings would be biased upward, which is QUESTION-040’s worry, and the paper does not discuss it.
  • THEORY-133. A complement. Both reach downward exclusion system by system; degeneracy reduction is realization insensitivity in information terms. Neither paper cites the other.
  • LIT-025. A different formal object under the same name: interventional and comparing two models here, observational and asking what a macro feature predicts beyond the parts there.

THEORY-222 records the result as Proposed.

Backlinks