Explore Bernard Baars' Global Workspace Theory (GWT) and Kahneman's Dual-Process cognition. Differentiate sub-100ms reflex heuristics from deliberative tree search.
GWT MemorySystem 1 / 2ACT-R
Track 2 Β· Neuro-Symbolic
Neural Perception + First-Order Logic
Unify deep neural embeddings with symbolic knowledge graphs and automated SMT theorem proving to achieve mathematical zero-hallucination deductions.
Knowledge GraphFirst-Order LogicZ3 / SMT
Track 3 Β· Embodied Intelligence
Active Inference & World Models
Model perception and action under Karl Friston's Free Energy Principle. Implement predictive coding loops where synthetic agents act to minimize unexpected sensory surprise.
Active InferenceWorld ModelsFree Energy
Track 4 Β· Collective Swarms
Emergence & Stigmergic Consensus
Architect decentralized multi-agent collective intelligence without central coordinators. Study stigmergy, Byzantine fault-tolerant consensus, and emergent problem solving.
Swarm ConsensusStigmergyDecentralized P2P
πΊοΈ Synthetic Cognitive Architecture Lifecycle FlowClick stage to inspect
ποΈ
Neural Perception
Continuous Vectors
β
π
Symbol Grounding
Entity & Relation Extraction
β
ποΈ
Global Workspace
Competitive Attention
β
βοΈ
Symbolic Deduction
First-Order Proofs
β
π¦Ύ
Active Inference
World Action & Loop
Stage 1 Β· Neural Perception: High-dimensional sensor inputs, vision tokens, and telemetry streams are encoded into dense representation manifolds via deep neural networks.
Lab Β· Dual-Process Decision Threshold & Active Inference Simulator
Tune uncertainty threshold $\theta$. Observe how sensory ambiguity triggers an automatic cognitive escalation from fast System 1 reflex heuristics to formal System 2 neuro-symbolic deduction.
Threshold: 0.75
Sensory Prediction Error (Free Energy):0.68 bits
System 1 Neural Confidence:62.0%
Cognitive Routing Decision:Escalated to System 2 Symbolic Deduction
Formal Logical Proof:Valid (Modus Ponens proven via Knowledge Graph)
Active Inference Action:Reconfigure Traffic Boundary & Clamp Circuit
1. How does Synthetic Intelligence (SI) fundamentally distinguish itself from conventional deep learning LLMs?
2. Why is Synthetic Intelligence (SI) distinguished from Superintelligence (ASI) in this curriculum?
3. In the Free Energy Principle of active inference, how does an embodied synthetic agent resolve unexpected sensory observations?
π οΈ Synthetic Intelligence Practical Research Projects & DrillsResearch Runbooks
RESEARCH DRILL #1
Catastrophic Forgetting During Continual Online Plasticity
Challenge: When learning new environment dynamics in real time, neural weights overwrite foundational safety invariants. Treatment: Decouple long-term symbolic invariants into an immutable Knowledge Graph, while restricting continual weight adaptation to an Elastic Weight Consolidation (EWC) plastic layer.