Macheng Shen
Featured essay + research map

Macheng Shen

I think about intelligence as closed-loop, multiscale competence in finite physical systems: how useful structure is acquired, how control is maintained, and how learning signals move through real substrates.

Current framing. The main thesis lives in the featured essay and the ideas page. The wave-backprop work is now treated as a narrower research branch about credit transport and physical implementations of learning, not as a finished total theory.

Featured Essay (Full Text)

This PDF is the current overview of the broader agenda: intelligence as a finite physical system that builds, updates, and deploys internal structure sufficient for prediction, intervention, and control.

If the embedded viewer does not load in your browser, open the PDF directly: essay.pdf.

Research Map

Main thesis

Toward a theory of intelligence

The core question is what sort of object intelligence is. The working answer is not “static function approximation,” but multiscale closed-loop competence under physical, informational, and control constraints.

Ideas page

Line loss, viability, and distributed cognition

These essays develop the higher-level language: physical budgets, coordination costs, endogenous viability, and why multiscale organization keeps appearing in biological and artificial systems.

Research branch

Credit transport and wave-inspired learning

This branch asks a narrower question: how does task-relevant update information move inside a learning system? Some physical media can compute with fields and even measure gradients in situ; cortex may use its own cousin of this idea.

Original Raw Prompt

I am still sharing the original prompt for reproducibility. It was useful as a seed, but several claims on the site have since been revised into a more careful layered framing.

View raw prompt (.txt)

About

I am trying to find a language that can describe intelligent behavior across biology and AI without collapsing everything into current engineering jargon. That means connecting learning, control, multiscale feedback, physical implementation, and safety.