IIT Expert

Precomputed result

The bottleneck network of the IIT 4.0 paper condenses into the complex Ab and four monads

Substrate Six binary units A–F with states in {−1, 1}, each a logistic of its weighted inputs (Eq 60) with k = 4. Unit A projects to each of B–F with weight 0.9. B projects back to A with weight 0.22, and each of C–F with weight 0.195. Each of B–F has a self-connection of 0.1. PyPhi ships the substrate, with the weights supplied by the authors, as pyphi.examples.iit4_2023_fig6a_substrate.
State (1, −1, −1, −1, −1, −1) — Abcdef, unit A ON
φs 0.05532977871393924
Minimum partition For the first complex Ab, the directed bipartition cutting B→A (weight 0.22). Each monad's minimum partition cuts its self-connection. The same under both formulations.
Formalism iit-4.0-2026
PyPhi 2.0.0rc4
Script albantakis-2023-fig6a-bottleneck.py

Albantakis et al. (2023) use this network to show how degeneracy and indeterminism keep complexes small. Unit A forms a bottleneck: all causal interactions among B–F pass through A. The current state of a large candidate system therefore has many possible causes and effects, which lowers its selectivity and its intrinsic information.

complex φs, Fig 6A 2023 2026 distinctions Φ, Fig 6A Φ, computed
Ab 0.05 0.0553 0.0553 2 0.48 0.4881
c, d, e, f (each) 0.008 0.0079 0.0079 1 0.02 0.0158

The distinctions of Ab match the figure: A → B | B with φd = 0.1865 (figure 0.1864) and b → a | a with φd = 0.0576 (figure 0.0575). Each monad specifies one distinction, c → c | c with φd = 0.0079. Ab has two relations with Σφr = 0.2440, and each monad one relation with Σφr = 0.0079, as in the figure.

The values are the same under both formulations: the intrinsic information requirement of Mayner et al. (2026) does not bind for any of these complexes.

The script uses the default configuration of each formulation. To reproduce, run the script with uv, which installs PyPhi 2.0.0rc4 for the run:

uv run --with "pyphi==2.0.0rc4" python canon/albantakis-2023-fig6a-bottleneck.py

Sources