Causal emergence in learning agents

Uses effective information (EI) and coarse-graining to link causal emergence with RL and biological learning.

23 members. Each node is clickable.

Loading graph…

Drawn from 4 sources

The papers/notes whose extracted claims & findings make up this cluster.

Emergence and Causality in Complex Systems: A Survey on Causal Emergence and Related Quantitative Studies15 members
The Causally Emergent Alignment Hypothesis: Causal Emergence Aligns with and Predicts Final Reward in Reinforcement Learning Agents7 members
2026-05-14_phil-trans-A-goodfire-aboutblank-impact.md1 member
cognitive-glue-and-alexander.md1 member

Bridges (5)

Other communities that share members with this one — cross-cutting threads or papers that sit at the seam between two themes.

Findings (12)

Biological networks exhibit the lowest EI among real networks and show the most significant causal emergence after coarse-graining.Finding from Klein & Hoel (2020) on real network analysis.
Causal emergence depends on the coarse-graining strategy: different partitions of the same boolean network yield EI values 1.55 (emergence) vs 0.18 (degradation).Example from Hoel et al. (2013) replicated in the survey.
Causal emergence measured by NIS+ increases with observational noise but decreases with dynamical noise.Insight that coarse-graining filters external noise but not intrinsic noise.
Causal emergence predictive of final reward early in RL training across multiple algorithms, architectures, and environments.Empirical result: CE measurements correlate with and predict learning performance in RL agents.
EI of ER random networks converges to -log2(p) with increasing size, with a phase transition at average degree ≈ log2(N).From Klein & Hoel (2020) analysis of artificial complex networks.
In AOMIC ID1000 movie-watching fMRI data, NIS+ finds a one-dimensional macro-state representing 100-dimensional micro-states.Real brain imaging result suggesting a compressed emergent representation.
In AOMIC PIOP2 resting-state fMRI data, NIS+ finds a seven-dimensional macro-state with widely distributed attributions.Contrast to movie-watching condition, showing context-dependent emergence.
NIS+ automatically discovers two-group macro-states in Boid model simulations matching the two boid groups.Yang et al. (2023) experiment on emergent herding behavior.
NIS+ captures emergent static/dynamic patterns such as 'gliders' in Conway's Game of Life within the latent space.Yang et al. (2023) demonstration of emergent pattern recognition.
NIS+ learns macro-dynamics matching ground-truth SIR dynamics from noisy micro-level data.Experimental result from Yang et al. (2023) reported in the survey.
NIS+ outperforms NIS, variational autoencoders, and feed-forward neural networks in out-of-distribution generalization experiments.Yang et al. (2023) result linking EI maximization to robust generalization.
Protein interaction networks across >1800 species exhibit macro-scale nodes with lower noise and higher resilience; eukaryotes show stronger CE than archaea.Klein et al. (2021) analysis of biological interactomes.

Claims (11)

Biological agents increase causal emergence after learning new memories.Prior empirical observation from biological systems; motivates investigation in artificial agents.
Biological and artificial agents share causal emergence as an axis of learning and reorganization.Interpretive assertion bridging Levin's biological cognition work with artificial RL; extends 'minds at all scales' thesis.
Causal emergence alignment with learning is a shared axis comparing biological and artificial creatures.Assertion that the correlation between causal emergence and learning constitutes another way biological and artificial intelligences converge.
Causal emergence can enable causal interventions to create better RL agents.Assertion that understanding causal emergence may lead to methods for manipulating agent representations to improve performance.
Causal emergence identification tasks can be understood as causal representation learning tasks.Authors propose a conceptual mapping between CE identification and CRL.
Causal emergence is widespread across measures of causation, not just EI.Claim by Comolatti & Hoel (2022) endorsed by this survey.
Causal emergence may be a previously undisclosed axis of reorganization of neural representations in RL agents.Authors' interpretive assertion that the observed alignment reveals a novel organizing principle of neural representation dynamics.
Causal emergence provides new perspectives for causal representation learning, interpreting latent variables as emergent causalities.Cross-fertilization claim made in discussion.
The NIS and NIS+ frameworks provide effective solutions for causal emergence identification from data.Central claim of the machine-learning section, summarizing the contribution.
Design choices propagate causally to subsequent centers through the unfolding mechanism itself.
Empowerment as intrinsic reward bridges causal learning and reinforcement learning in agent development.