Reproduce “The Philosophical Foundations of Growing AI Like A Child” 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

Dezhi Luo, Yijiang Li, Hokin Deng

This paper theorizes that current large language models struggle with robustness and intuitive tasks due to a lack of "core knowledge," advocating a shift from mere "scaling up" to a "growing up" developmental paradigm. It proposes integrating foundational cognitive abilities into multimodal models through systematic synthetic data generation and "cognitive prototyping" to foster more grounded and reliable AI.

Key insights

  • The robustness challenge and Moravec's Paradox in LLMs are unified by a fundamental discrepancy between human "growing up" (structured skill acquisition) and machine "scaling up" (parameter and data increase).
  • The absence of "core knowledge"—foundational cognitive structures regarding objects, actions, number, space, and social relations—is the crucial missing link preventing robust, generalizable AI abilities.
  • LLM core knowledge deficits are not necessarily inherent architectural limitations but potentially an issue of knowledge being "buried too deeply" or ungrounded due to unstructured learning trajectories.

Summary by alphaXiv.

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