Reproduce “Causality in the human niche: lessons for machine learning” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Richard D. Lange, Konrad P. Kording

The authors argue that causal reasoning in the human niche should account for goals, constraints, and object-property knowledge, not only fixed causal graphs. They describe causal induction as using higher-level causal theories to build situation-specific models, illustrated by needles and porcupine quills popping balloons. The review proposes a research direction and reports no tested system or quantitative transfer gain.

Key insights

  • Human causal cognition is operational and contextual, tied to an agent's ability to intervene and guide action, prioritizing 'good-enough' models over perfectly veridical representations.
  • Causal reasoning in humans is deeply adapted to the 'human niche,' leveraging principles like agency, active intervention, coarse-graining, causal induction, and social learning.
  • The assumptions of Structural Causal Models (SCMs), such as pre-defined variables, Directed Acyclic Graphs (DAGs), and independent exogenous noise, are often too restrictive for the complexities of the human niche and real-world, dynamic scenarios.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.