NLM — Natural Learning Mechanisms

Intelligence
doesn't require
comprehension.

Navigation doesn't require maps.
Adaptation doesn't require language.
Explore the Simulation View Installation Proposal
12
Autonomous Rovers
$13
Per Unit Hardware
3
FSM Rules
0
Neural Networks
3.8B
Years of Precedent
"We built machines that talk. Now we've confused talking with thinking."
— NLM Manifesto, Blacksky Labs, 2026
The current paradigm

Language models predict the next word. Beautifully.

So beautifully that we've started to believe prediction is comprehension — that fluency is intelligence. The infrastructure required is concentrated in a handful of companies, in a handful of countries, consuming resources at a planetary scale.

  • GPT-4: ~50 petaflop-days of compute to train
  • Single training run ≈ emissions of 5 automobiles
  • Requires data centers, cloud accounts, API keys
  • Access is not the same as ownership
What nature already knows

A fire ant with a brain smaller than a pinhead navigates flood waters.

No training data. No parameters. No prompts. The ant doesn't understand what it's doing. It doesn't need to. The slime mold grew the Tokyo rail network — using chemical gradients alone — before human engineers solved the same problem.

  • Physarum polycephalum: no brain, solved Tokyo rail
  • Termite mounds: stable temperature, no thermostat
  • Murmurations: one organism, no leader, no plan
  • Mycorrhizal networks: 3.8 billion years of operation
The NLM proposition

What if we built AI that way?

Not AI that talks about navigation, but AI that navigates. Not AI that describes adaptation, but AI that adapts. Physical signals. Simple rules. Distributed authority. No cloud. No model. No language.

  • Local signal primacy — sense what is in front of you
  • Rule minimalism — emergent complexity from simple logic
  • Physical signal vocabulary — pheromones, not tokens
  • Distributed authority — no queen, no controller
  • Resource minimalism — $13 and a simple rule set
The Five Principles of NLM
01
Local Signal Primacy
Sense what is in front of you
All intelligence derives from signals originating in the immediate physical environment of the node processing them. A node does not query a remote server or a trained model. It reads what is in front of it — pheromone concentration, BLE signal strength, obstacle proximity — and acts on that reading alone.
02
Rule Minimalism
Complexity is emergent, not engineered
NLM systems are governed by the smallest possible set of rules capable of producing the required adaptive behavior. SlimeHive operates on three FSM rules: Alignment, Attraction, Repulsion. No rule requires more than a comparison operator and a threshold value. Complexity arises from the interaction of simple rules across many nodes and many time steps.
03
Physical Signal Vocabulary
The signal is the language
The signals that an NLM system processes must be physical, chemical, or environmental in origin — not linguistic, semantic, or synthetic. Text prompts and API responses are explicitly excluded. Valid NLM signals include: pheromone gradients, mechanical pressure, electromagnetic radiation, proximity, temperature differential. NLM systems speak the language of the physical world.
04
Distributed Authority
No queen. No controller. No master process.
No single node holds authority over the behavior of other nodes. Each node is sovereign within its local context — it reads its signals, applies its rules, and acts. Coordination emerges from the cumulative effect of many sovereign nodes acting on shared signal vocabularies in shared physical space. This is the architecture of every successful biological collective.
05
Resource Minimalism
A justice constraint, not a technical one
NLM systems must be deployable on commodity hardware available at consumer price points in any global market. A system that requires specialized hardware accessible only to wealthy institutions cannot serve as the foundation of a commons. The target platform is any microcontroller available for under $20 USD. The Raspberry Pi Pico 2W ($13) is the reference platform for SlimeHive.
Museum & Gallery Installation

The Physical Proof

"A terrarium full of scrappy robots made from electronic waste, learning to survive in a glass box. It's small. It's imperfect. It's exactly the point."

Twelve autonomous rovers navigate an enclosed glass habitat — finding energy, avoiding a predator, rebuilding paths as the terrain is reshaped around them. They carry TinyML neural networks so small they fit in 100 kilobytes. Models that don't predict words. They predict: Is this signal getting stronger? Should I turn left? Is this charging station worth the risk?

Day one, the rovers wander. They bump into walls. They miss charging stations. Week two, something happens. Routes get smoother. Stations get found faster. The predator gets avoided more often. Not because anyone programmed better behavior — but because the tiny networks learned from failure. From the environment itself.

Inquire about placement →
Installation Specifications
Rover Count 12 autonomous units
Hardware Raspberry Pi Pico 2W
Unit Cost ~$13 USD per rover
Sensor Array Ultrasonic + BLE
Intelligence Layer TinyML, <100KB models
Neural Networks 0
FSM Rules 3 — Align, Attract, Repel
Habitat Glass enclosure, variable terrain
Cloud Dependency None
Materials Primarily reclaimed electronics
Origin Blacksky Labs, Baltimore, 2026
SlimeHive installation — exhibition view 1
SlimeHive installation — exhibition view 2
SlimeHive installation — exhibition view 3

Curator & Gallery Inquiry

Seeking museum, gallery, and academic institution placement for 2026–2027. Please reach out to discuss the installation, loan, or exhibition opportunities.

All inquiries answered within 48 hours.

✓  Message received — we'll be in touch within 48 hours.
Live Browser Simulation

Watch Intelligence Emerge

The same principles that govern the physical installation — pheromone gradients, BOIDS flocking, emergent path-finding — run live in your browser. Configure a colony. Watch behavior appear that you did not program.

Live preview — scaled simulation

9 behavior modes. 2 pheromone grids.
Infinite emergence.

Spin up a flock of 50 worker drones and 3 hopper scouts. Set exploration to High. Watch the swarm spread outward, mark food sources with ghost pheromones, then converge back to the queen carrying what it found. Switch to Ghost viz mode to see the communication network underneath.

Behavior Modes
FEED_QUEEN
BOIDS
FORAGE
SCATTER
SWARM
ALIGN
FLOCK
AVOID
RANDOM
Run Simulation Download Code
Runs entirely client-side — no server, no data collection
Blacksky Apprenticeship

NLM 101 — Intelligence Without Words

A 2-week apprenticeship for students who want to understand intelligence before it learned to speak. No prerequisites. No coding required. Just curiosity and the willingness to watch a simulation fail.

2 Weeks
4 Sessions
1 Capstone Demo
Week 1 · Session 1

The Intelligence Beneath the Words

What intelligence looks like when there is no language — and why that matters. Students spin up their first flock in SlimeHive and ask the fundamental question: where did the intelligence come from?

  • Emergent behavior
  • BOIDS algorithm
  • IoT as nervous system
  • Centralized vs. distributed
Week 1 · Session 2

The Nudge — Steering the Swarm

Thaler & Sunstein's choice architecture applied to swarm behavior. Students add attractors and obstacles and measure how the smallest environmental change steers collective behavior toward different outcomes.

  • Nudge theory
  • Platform algorithms
  • Attractor design
  • Ethics of steering
Week 2 · Session 3

The Eye That Doesn't Think

Computer vision basics — how machines see without understanding. Students sketch what a CV system would see in their simulation. They compare machine perception to human perception and locate where they diverge.

  • Pixels to features
  • Edge detection
  • IoT camera limits
  • The NLM paradox
Week 2 · Session 4

Demo Day — Present Your Intelligence

Each student presents their SlimeHive capstone: explain the configuration, show the emergence, break it deliberately, and connect it to a real-world system. Understanding failure is the beginning of mastery.

  • Live simulation demo
  • Emergent behavior analysis
  • Designed nudge effects
  • Real-world parallel
"Build a simulation that produces behavior you didn't directly program — then design a nudge that steers it."
Capstone Project Brief
Inquire About the Course
Research Framework
We are the small axe. We intend to be sharp.
Mario Moorhead — NLM Framework Paper, 2026
Natural Language Modeling (NLM): Emergent Intelligence Through Physical Signal Processing Without Large Language Models
Mario Moorhead — Blacksky LLC
2026 · Two Working Proofs

This paper proposes NLM — Natural Language Modeling — as a framework for building adaptive, responsive, context-aware systems using only local physical signal processing. No transformers. No training data. No cloud dependency. Intelligence, in the NLM framework, emerges from what the Caribbean has always known: that who feels it knows it. We present two working proofs: SlimeHive and Ceiba. Neither contains a neural network. Neither requires internet connectivity. Both are alive.

Read the Paper
Five NLM Principles
01
Local Signal Primacy
02
Rule Minimalism
03
Physical Signal Vocabulary
04
Distributed Authority
05
Resource Minimalism
Two Working Proofs
SlimeHive
Swarm robotics on Raspberry Pi Pico 2W. Pheromone gradient navigation.
Ceiba
Modular self-powered infrastructure node. Named for the sacred tree of the Caribbean and West Africa.
Artist & Researcher
Mario Moorhead
Mario Moorhead
Founder, BlackskyLabs · Baltimore, MD

The author of this paper learned to code on a $35 Radio Shack computer in Charlotte, North Carolina. He was born in St. Croix, United States Virgin Islands. He has stood in Havana garages where mechanics kept 1930s automobiles running by fabricating parts from scratch — adapting, improvising, resolving. He has walked the souks of Marrakech where artisans transform discarded components into objects of precision and beauty. He did not wait for the technology to trickle down. He built with what was in front of him.

NLM is that instinct formalized into a methodology. SlimeHive is the proof of concept. Blacksky Labs is the practice — a design and research studio at the intersection of technology, culture, and the communities that built the modern world and were then told to wait for its benefits.

St. Croix, USVI Charlotte, NC Baltimore, MD BlackskyLabs · 2026
Manifesto

Natural Language Modeling

Sixty-five million years ago, the most complex organisms on Earth were destroyed — not by something smarter, but by something that needed less.

The creatures that inherited the planet were small, efficient, and built for a changed world. We are in a similar moment with artificial intelligence. The dominant paradigm — large language models trained on billions of dollars of compute, running in data centers that consume the electricity of small nations — is optimized for conditions of abundance. Those conditions are ending.

The question is not whether the correction is coming. The question is whether the intelligence we are building is designed for the world after it.

Natural Language Modeling is that design. It begins with a redefinition: language is not human speech. Language is any signal that carries meaning — and the oldest languages on Earth are chemical, physical, and environmental. A pheromone trail. A pressure drop. A shift in light. These signals have been governing adaptive behavior for 3.8 billion years, long before the first word was spoken and long before the first transformer was trained.

NLM encodes this ancient intelligence into commodity silicon. Two working proof systems demonstrate what becomes possible.

SlimeHive

A swarm of robots, each under $40, that navigates and coordinates using the same mechanism a single-celled organism used to independently replicate Tokyo's rail network.

Ceiba

A solar-powered infrastructure node, under $250, that produces electricity, clean water, light, and mesh connectivity for communities that cannot wait for the grid.

Neither contains a model. Neither requires the cloud. Both work now, in the field, built by one person in Maryland from parts available in any electronics market on any continent.

This is not a critique of AI. It is a correction to its definition. Intelligence is not computation. It is emergence — and emergence belongs to everyone.