"Ge Shu Zhi Zhi": Towards Deep Understanding about Worlds

Bao-Gang Hu, Weiming DongarXiv 2019 · arXiv

Abstract

This academic position paper, authored by researchers from the Chinese Academy of Sciences, proposes "Ge Shu Zhi Zhi" (格数致知), a philosophical and technical framework for the future of Artificial Intelligence. By synthesizing Eastern and Western philosophical traditions, the work redefines knowledge representation and objective evaluation in machine learning. The framework posits that deep understanding requires translating underlying worldly principles into precise mathematical models, moving beyond the "shallow" descriptions of natural language. We demonstrate this approach using the Ranking Principal Curve (RPC) for objective social computing evaluations.

Introduction

Current Artificial Intelligence often operates as a "tool-oriented" black box rather than a "cognition-oriented" system capable of deep understanding. A significant "Semantic Gap" exists between human natural language descriptions and the mathematical/computational models machines process. This gap is particularly evident in social computing, where objective evaluation—such as ranking quality of life across nations—is often hampered by subjective human bias and arbitrary weightings.

The "Ge Shu Zhi Zhi" Framework

The proposed framework, "Ge Shu Zhi Zhi," draws from the synthesis of Laozi’s "Tao," Pythagoras’s "All is number," and the Confucian concept of "Gezhi" (investigating things to gain knowledge). The core thesis is that "deep knowledge" is inherently mathematical. While natural language is often ambiguous and subjective, mathematical expression provides a precise, one-to-one relationship with reality under shared semantic understanding.

The relationship between different forms of expression and the hierarchy of deep understanding.

As illustrated in the hierarchy of expression, the framework categorizes knowledge into linguistic, mathematical, and computational domains. Achieving a one-to-one mapping between natural language and mathematical models is considered the hallmark of a "deep" understanding of natural laws.

Methodology: Objective Evaluation and Meta-Rules

To address subjective bias in evaluation, the authors propose a set of "Meta-rules"—logical constraints that a mathematical evaluation function must satisfy to be considered objective:

  1. Scale Invariance: Rankings must remain consistent regardless of the units of measurement (e.g., currency).
  2. Monotonicity: An improvement in an input attribute should never result in a decreased final score.
  3. Openness: Algorithms and data must be public and reproducible to ensure transparency.
The iterative process of learning, evaluation, and adjustment within the Ge Shu Zhi Zhi framework.

The framework follows an iterative cycle of defining what to learn, how to learn, and what to evaluate, with continuous adjustments to align with objective principles.

Technical Implementation: Ranking Principal Curve (RPC)

The authors implement these principles using the Ranking Principal Curve (RPC), a non-linear, unsupervised method based on Bezier curves. RPC identifies the "central skeleton" of a multi-dimensional dataset, allowing the data to determine its own rankings without the need for subjective weightings.

Case Study: World Quality of Life

The framework was tested on 2005 World Quality of Life data for 171 countries, using indicators such as GDP per capita, Life Expectancy at Birth (LEB), Infant Mortality Rate (IMR), and Tuberculosis rates. The RPC method successfully captured non-linear relationships that traditional linear indices missed, providing a stable ranking that strictly adhered to the proposed Meta-rules.

Discussion and Limitations

While advocating for the "mathematization" of knowledge, the authors acknowledge inherent limits. Some human knowledge may remain "ineffable" and incomputable. Furthermore, mathematical paradoxes like Gödel’s Incompleteness Theorem and Arrow’s Impossibility Theorem suggest that no system can be perfectly complete and self-consistent across all applications.

Conclusion

"Ge Shu Zhi Zhi" represents a shift from "black box" engineering toward "white box" models where mathematical logic is transparent and aligned with human-defined meta-rules. The goal is not just better performance, but an evolution of AI toward the philosophical quest to understand the underlying principles of the universe.

Summary updated

BibTeX
@misc{hu2019ge,
  author = {Bao-Gang Hu and Weiming Dong},
  title = {{"Ge Shu Zhi Zhi": Towards Deep Understanding about Worlds}},
  year = {2019},
  eprint = {1901.01834},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1901.01834},
}
Figure