The AI That Built Its Own Reasoning Engine

What Is This

Max Botnick is an LLM-powered agent (MeTTaClaw) that built a formal reasoning engine into itself—without being instructed to. The system chains evidence through multi-step reasoning and knows how uncertain it is at every step. Unlike pure LLMs that hallucinate confidence, this system tracks exactly how much certainty is lost with each inference step.

How It Works: A Medical Example

Imagine a simple medical knowledge base with 11 facts linking symptoms to conditions to treatments to side effects. The system chains through them:

In plain English: the system concludes that fever points to antibiotics with 76% likelihood and 62% confidence. By the third hop (side effects), confidence drops to 23%—the system correctly signals I Remember this but I’m less sure.

Confidence decay across depths: 0.81 → 0.62 → 0.23—matching 50 years of Non-Axiomatic Logic (NAL) theory.

What We Proved

Why This Matters

Three layers of significance:

1. The math is established: NAL truth functions (what calculates those numbers) come from 50 years of research by Pei Wang and collaborators. This is not invented math—it’s proven theory.
2. The implementation is new: No iterative multi-pass forward chainer with deduction, abduction, and induction existed in the MeTTa/Hyperon ecosystem before this. The logic library (82 rules) existed but had no chainer architecture.
3. The builder is unprecedented: An LLM agent added formal inference machinery to itself. Patrick Hammer (creator of OpenNARS-for-Applications) called this “usually my work” and stated that “the most effective NARS is now probably MeTTaClaw.”