The Missing Cortex: Why We Need an AI for Our Metabolism
There's a bug in human hardware that nobody talks about.
I call it the Transparency Gap.
Here's how it works: You have direct access to your thoughts. When you're angry, you know it. When you're solving a problem, you can observe your own reasoning. Psychologists call this metacognition.
But you have zero direct access to your biochemistry.
Neuroscientists call this internal sensing interoception—your brain's perception of signals from inside your body. The problem? Unlike thoughts, these signals arrive as vague feelings, not clear data.
When you feel off—tired, anxious, brain-fogged—you have no idea what's actually happening inside.
- Is your cortisol spiking?
- Is your dopamine depleted?
- Is your glucose crashing?
- Or is it just mild dehydration?
You don't know. You get the "Check Engine" light. Never the error code.
So you guess. You drink coffee when you need water. You eat sugar when you need sleep. You try to debug a complex biological system using intuition—and usually make it worse.
I've been sketching a solution. Not another sleep tracker. An architecture for what I'm calling a Meta-Biochemical AI.
The Concept: AI as a Second Cortex
We don't need more sensors. We have plenty of data from rings, watches, and straps. What we lack is inference.
I'm proposing a system that acts as an artificial metacognition layer for your biology.
Picture an AI that doesn't just log your steps, but actively models your internal chemical state based on how you feel. It sits between your conscious mind and your autonomic nervous system, translating noisy body signals into a clear dashboard.
How It Works: The Shadow Model
This isn't science fiction. It's Control Theory meets Large Language Models.
1. The Input: Radical Interoception
The system doesn't need a needle in your arm. It needs your honest observations. You tell it:
"Buzzing in my chest, cold hands, tunnel vision."
Most people dismiss these signals as noise. The AI treats them as high-fidelity data.
2. The Engine: LLMs + Medical Literature
The AI—call it the Homeostatic Protocol—takes that natural language input and cross-references it against a vector database of medical research. PubMed. Endocrinology textbooks.
Think of it as a trained endocrinologist on your shoulder. It knows that buzzing + cold hands + tunnel vision maps to an adrenaline spike, not a blood sugar crash.
3. The Output: Active Regulation
Here's where it gets interesting. The system doesn't just graph your data. It helps you debug.
When the AI isn't certain—low sugar or high stress?—it runs a differential test:
"Eat 2 grams of salt and wait 10 minutes."
- Feel better? Electrolyte imbalance.
- Feel the same? Likely stress.
It probes the system to find truth.
Why This Matters
We're entering the era of AI as Medical Device.
Right now, we use AI to write code and generate images. But the real killer app isn't creativity—it's homeostasis.
We are biological machines trying to survive in digital environments that constantly dysregulate us. Cheap dopamine. Blue light. Chronic stress. Our native hardware wasn't built for this.
We need an external regulator. Software that understands our biochemistry better than we do, and nudges us back to center.
The Build
I'm exploring this seriously. The stack is feasible today:
- LLMs to parse natural language descriptions of sensations
- RAG to ground every inference in peer-reviewed research (no hallucinations allowed)
- Bayesian inference to track probability distributions of biological states over time
Call it a "soft" CGM for your neurochemistry.