Can Konuk

Interactive models / Causal selection / Interactive demo

Causal selection: interactive demo

Can Konuk. Runs the model of chapter 4 of the dissertation in your browser; the notebook explains each step.

Write a causal rule, set how likely each cause is and what actually happened, then step the model through it: compile the rule into a network, sample counterfactual worlds, learn from them, and score every candidate cause.

The network and its samples

The rule compiles into one hidden unit per clause. Each sample proposes a counterfactual world near the actual one; filled units are true in the current sample.

  • True in this sample
  • False
  • Candidate cause
  • Carries signal, width = weight
Network

Chain

Last 40 samples, oldest on the left. Grey outlines mark a rejected proposal; the ringed column is the current sample.

Scoring causes

Relevance flows back from the outcome to the causes in the actual world; a cause set's importance \kappa is its relevance divided by the path complexity of its routes.

Candidate cause set

The full model on Experiment 2

The same steps, run 500 times per scenario with the fitted settings, give one predicted rating per cause set. Move the parameters and compare the predictions with what participants said.

Scenario
1.15
20
0.58
  • Participants, mean and 95% interval
  • Model
FitMean responsibility ratings in Experiment 2 (311 participants, 1–9 scale) against the model's prediction 1 + 8\kappa^{\gamma}, one row per cause set. The fitted values are \eta = 1.15, n = 20, \gamma = 0.58.
© 2026 Can Konuk Model code: causal-model.js, a port of the R notebook