ENTROPIS
A synthetic brain. Tested at 2B-neuron scale.
The Experiment
We built a synthetic brain.
Then we ran the benchmark record.
The system exhibited a property we didn't program — one that biological brains share.
ENT-INTER Benchmark
"Embodied operating conditions sustain resting dynamics."
Hypothesis tested. Reported result: confirmed under protocol.
Brain without embodiment
Control condition
→ Brain goes dormant
Activity exits target range
Brain with embodiment
Embodied condition present
→ Brain stays alive
Activity remains in target range
This was not programmed.
No code instructs the system to "require embodiment." The system has no explicit dependency on that condition. The dormancy emerged from the dynamics themselves.
Hardware invariant: confirmed across two hardware classes
“Consciousness is to the brain-body system what a magnetic field is to a rotating star.
You cannot store it. You cannot transfer it.
You can only maintain the conditions that generate it.”
— Shiv Goswami, Founder
Position
Two Fundamentally Different Approaches
Current AI
- Statistical pattern matching on massive datasets
- Frozen weights after training phase
- No embodiment loop
- Cloud-dependent deployment pattern
This System
- Emergent cognition from physics and dynamics
- Continuous learning without retraining
- Requires embodiment (observed empirically)
- Runs on accessible hardware classes
Validated
Hardware invariance across fundamentally different computational substrates: two hardware classes, large scale difference, comparable benchmark behavior.
Methodology
101 benchmark markers. 20 physics-based cognitive tests with zero reward signals. Cross-platform behavior across two hardware classes. Reported consistency across 100+ controlled runs.
Physics-Based Learning
Structural Adaptation
Internal Structural Adaptation
Internal structure adapts through system dynamics. No reward signals. No gradient descent.
Learning Dynamics
Biological Awake/Sleep Cycle
Consolidation during rest. Continuous adaptation during activity. No training phase.
Validation
20 Cognitive Tests Across 5 Levels
L1
4/4
Perception
L2
4/4
Relations
L3
4/4
Memory
L4
4/4
Adaptation
L5
4/4
Transfer
No rewards. No labels. No backpropagation. Physics only.
The Question
What the synthetic brain record shows.
| Property | Expected | Observed |
|---|---|---|
| Target operating regime | Activity remains in target range | Confirmed |
| Embodiment dependency | Dormant without embodiment | Confirmed |
| Expectation-response behavior | Activity for expected-but-absent stimuli | Observed |
| Continuous learning | No reward/backprop loop used | No such loop used |
| Hardware invariance | Comparable behavior on different architectures | 2 classes |
101 markers. 20 cognitive tests. Three platforms.
Key Finding
The brain self-organizes to the target operating regime.
Target operating range: confirmed | 90%+ healthy samples
The target operating regime is associated with stable, high-information dynamics.
Public benchmark marker: target operating regime confirmed
Validation
Accessible hardware class. Internal technical details protected.
Full benchmark suite. Two hardware classes.
Scale Evidence
Same Emergence. Any Scale.
400× scale difference. Comparable benchmark behavior.
The reported results support scale continuity across materially different substrate sizes.
Discovery
The brain predicted a stimulus that was never shown.
During a learned sequence, an expected element was withheld.
The brain showed activity for the absent element under protocol.
Expectation response observed
Hardware invariant: observed across two hardware classes
Level 5 Cognitive Test
Prototype Formation
Experiment
Expose brain to related exemplars around an unseen central category.
The central category is never shown during training.
Result
When tested on the unseen central category:
Response = 1.0 (critical)
The brain formed the central tendency from exemplars alone.
Statistical learning through system dynamics. No rewards. No labels.
Level 4 Cognitive Test
Adaptive Plasticity
Plasticity itself adapts. High activity raises the threshold. Rest restores it.
Baseline
5.45
adaptations/trial
After Saturation
2.45
adaptations/trial
55% reduction
After Rest
1.60
adaptations/trial
Plasticity thresholds adapt under changing activity conditions.
Reported in a large-scale synthetic brain.
ENT-EM5: Embodiment
5 Markers of Closed-Loop Embodiment
A complete synthetic brain system. Each component follows first principles. Nothing is trained. Everything is emergent.
┌─────────────────────────────────────────────────────────────────┐ │ │ │ INPUT SURFACE │ │ ├── Signal processing │ │ ├── Environmental processing │ │ └── Embodiment integration │ │ │ │ COGNITIVE CORE │ │ ├── Continuous learning (no training phase) │ │ ├── Self-regulation │ │ └── Adaptive learning │ │ │ │ ACTION SURFACE │ │ ├── Output generation │ │ ├── Feedback integration │ │ └── Controlled response │ │ │ └─────────────────────────────────────────────────────────────────┘
Target Dynamics
System dynamics under load
Signal Processing
Input processing surface
Environmental Processing
External signal surface
Action Output
Controlled output surface
Embodiment Loop
Closed-loop embodiment
ENT-IQ5 Benchmark
Entropis Intelligence Quotient: 5 Markers
Five markers that distinguish brain-like dynamics from deterministic calculators. Reported across hardware conditions.
Adaptive Variability
PASSResponse variability exceeds the public benchmark threshold. Demonstrates state-dependent processing rather than deterministic computation.
Target Dynamics
Target rangeSystem activity enters the target operating range used by the public benchmark protocol.
Cascade Distribution
PASSActivity cascades follow a broad internal distribution under the public benchmark protocol.
Bidirectional Learning
±10% minimumSystem exhibits both habituation (decreased response to repeated stimuli) AND sensitization (increased response to novel/intense stimuli). Not just accumulation.
Endogenous Activity
Sustained > 0Spontaneous activity patterns occur without external input. Activity is self-sustaining, not stimulus-driven. Measurable through baseline firing rates.
ENT-NOVELTY Benchmark
Entropis Novelty Detection: 3 Markers
Does the brain recognize familiar vs. novel? Can it remember? Hardware invariant on both platforms.
Habituation
Brain adapts to repeated stimuli. Learns patterns over time.
Novelty Detection
Brain responds differently to new vs. familiar stimuli.
Memory & Recovery
Brain remembers familiar stimuli. Faster re-stabilization.
Three markers of functional memory. Hardware invariant on both platforms.
Scientific Method
Falsification Record
Defined falsification criteria with observed results.
Speed below human brain
Falsified if biological baseline is not exceeded
No self-organization
Falsified if target dynamics never appear
Deterministic output
Falsified if CV < 1%
Platform-specific behavior
Falsified if one platform fails any marker
Reward/backprop training loop
Falsified if reward, label, or backprop loop used
Brain active without embodiment
Falsified if target dynamics persist under control condition
8 falsification tests. 0 falsified.
Public record: criteria defined, tests executed, results preserved.
System Capabilities
What the synthetic brain can do. Technical details under NDA.
Sensory Processing
Visual and auditory processing surfaces
Cognitive Processing
Working memory, planning, decision-making
Motor Output
Action generation and feedback integration
Continuous Learning
Real-time adaptation without retraining
Economics
$0
Training Cost
Cognition emerges from dynamics. No training required.
85-95%
Gross Margins
Zero retraining cost. Continuous learning.
2B
Neurons
Consumer GPU. $500 hardware.
Current AI: $100M-$5B training cost per model. Cloud infrastructure. Continuous burn.
This system: no reward/backprop training loop in the reported protocol. Consumer hardware. Structural economic advantage.
What Remains Protected
Technical discussion follows governed access.
What We Show
- ✓ Live demonstrations
- ✓ Real-time benchmarks
- ✓ Controlled repeat observations
- ✓ Performance metrics
What We Protect
- ◇ Core algorithms
- ◇ Learning mechanisms
- ◇ Integration design
- ◇ Source code
Origin
Conceived, designed, built, and validated by a single researcher.
Founder · Inventor
Anticipated Questions
What You May Be Thinking
Why protected technical review?
Entropis separates public evidence from protected technical material: benchmark summaries, controlled observation, and governed technical review. Cross-platform behavior reduces platform-artifact risk without public capability transfer.
Is this conscious?
We do not claim consciousness. We claim the system exhibits dynamical signatures associated with conscious biological brains, with embodiment required to maintain those dynamics. The philosophical implications are for others to debate.
How can I verify this?
Request controlled access. Benchmarks are observed directly under governed review. Technical discussion stays inside the protected review process.
See It Work
Live demonstration. Real-time benchmarks. Your questions answered.
Governed access. Serious inquiries only.
Request Demo
contact@terraflock.comAvailable to: Investors · Research institutions · Hardware partners