Reproduce “Acting as Inverse Inverse Planning” 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

Kartik Chandra, Tzu-Mao Li, Josh Tenenbaum, Jonathan Ragan-Kelley

Researchers at MIT CSAIL developed "inverse inverse planning," a computational framework for generating animations by optimizing character actions to evoke desired audience perceptions. This approach, which models human social cognition, effectively produces narratives depicting elements like plot twists and character intent, achieving significantly higher rates of correct audience interpretation compared to traditional planning methods.

Key insights

  • Audience perception of narrative elements, such as character intentions or plot developments, can be computationally modeled as an "inverse planning" problem, allowing for strategic design of actions to influence these inferences.
  • "Inverse inverse planning" provides a unified, principled computational framework based on optimization over Bayesian inference, capable of naturally expressing diverse storytelling elements like character depiction, plot twists, irony, flashbacks, and narrative arcs.
  • Optimizing animations directly for desired audience experiences, rather than solely for in-world rationality, yields significantly more effective and unambiguous communication of narrative intentions to human viewers.

Summary by alphaXiv.

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