Zero-Shot Generalization of Prospective Planning in Macaques
- 1New York University
Abstract
Achieving a goal in natural environments often requires a sequence of actions whose consequences extend beyond their immediate outcomes. Because actions within a sequence can hinder or facilitate subsequent actions, successful behavior often demands prospective planning: selecting current actions based on their consequences for future actions. Although prospective planning is well documented in humans, its generalization to unfamiliar problems in nonhuman animals remains poorly understood. Here, we developed maze-navigation tasks in which macaque monkeys moved a token toward a goal while navigating around obstacles. Monkeys selected routes based on the positions of downstream obstacles, and their choices could not be explained by simple whole-maze visual summaries. In previously unseen, complex branching mazes, monkeys avoided deceptive dead ends--routes that initially approached the goal but ultimately failed--even when those dead ends extended several maze segments downstream. Prospective route selection was evident from the first encounters with multistep mazes and generalized to previously unencountered arrangements of familiar obstacles. These findings demonstrate zero-shot generalization of prospective planning: macaques applied a prospective strategy to unfamiliar multistep problems without learning the appropriate action sequences through repeated experience.
Related
- Zero-shot evaluations expose structured generalization limits in predictive models of the brain
- ASTRA enables zero-shot super-resolution of spatial transcriptomics across capture geometries
- Inherent limitations of zero-shot mutation prediction in protein engineering
- Physics-Grounded Simulation Enables Zero-Shot Molecular and Reaction Analysis in Scanning Tunneling Microscopy
- Towards Sparse Causal Features for Zero-shot Mutation Effect Prediction in a Protein Language Model
BibTeX
@misc{bak2026zeroshot,
title = {{Zero-Shot Generalization of Prospective Planning in Macaques}},
author = {M. Seong Bak and B. Zhu and Y. Minai and J. Sakon and G. Okazawa and R. Kiani},
howpublished = {bioRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.19.760148},
url = {https://www.biorxiv.org/content/10.64898/2026.10.19.760148v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Zero-Shot Generalization of Prospective Planning in Macaques
AU - M. Seong Bak
AU - B. Zhu
AU - Y. Minai
AU - J. Sakon
AU - G. Okazawa
AU - R. Kiani
PY - 2026
DA - 2026/10/19
PB - bioRxiv
DO - 10.64898/2026.10.19.760148
UR - https://www.biorxiv.org/content/10.64898/2026.10.19.760148v1
AB - Achieving a goal in natural environments often requires a sequence of actions whose consequences extend beyond their immediate outcomes. Because actions within a sequence can hinder or facilitate subsequent actions, successful behavior often demands prospective planning: selecting current actions based on their consequences for future actions. Although prospective planning is well documented in humans, its generalization to unfamiliar problems in nonhuman animals remains poorly understood. Here, we developed maze-navigation tasks in which macaque monkeys moved a token toward a goal while navigating around obstacles. Monkeys selected routes based on the positions of downstream obstacles, and their choices could not be explained by simple whole-maze visual summaries. In previously unseen, complex branching mazes, monkeys avoided deceptive dead ends--routes that initially approached the goal but ultimately failed--even when those dead ends extended several maze segments downstream. Prospective route selection was evident from the first encounters with multistep mazes and generalized to previously unencountered arrangements of familiar obstacles. These findings demonstrate zero-shot generalization of prospective planning: macaques applied a prospective strategy to unfamiliar multistep problems without learning the appropriate action sequences through repeated experience.
ER -