AGI Mind Map Explorer (2026 State of the Art)

• Strong AGI vs Weak AGI

• Human-level AGI (HLAI) vs Beyond-human AGI

• Narrow AI Current State

• General AI The Goal

• Superintelligence Post-AGI

• Early dreams (Turing, Minsky, Simon)

• AI winters & structural bottlenecks

• Modern Resurgence (Deep Learning, LLMs, Scaling Laws)

• Transfer learning across open domains

• Adaptation to novel/unseen tasks

• High sample efficiency & Out-of-Distribution (OOD) generalization

• Dynamic environmental capability evaluation

• Fluid vs Crystallized intelligence

• Cognitive biases as evolutionary/bounded optimization constraints

• Physical commonsense (intuitive physics, spatial awareness)

• Social commonsense (intuitive psychology, cultural framing)

• Pattern recognition & Analogical mapping

• Metacognition (Self-monitoring, knowing when to defer or seek human help)

• Theory of Mind (First-order & Higher-order belief attribution)

• Access consciousness vs Phenomenal consciousness

• Global Workspace Theory (GWT) & Integrated Information Theory (IIT)

• SOAR, ACT-R, CYC systems

• Transformers & Self-Attention paradigms

• State Space Models (SSMs) Mamba / Liquid AI

• Logic Tensor Networks & Differentiable Inductive Logic

• Vector-Symbolic Architectures (VSA) / Hyperdimensional Computing

• CLARION, LIDA, Sigma systems

• Joint Embedding Predictive Architecture (JEPA / Yann LeCun)

• Dreamer & MuZero (Latent space environment simulators)

• GPU/TPU Interconnect Bottlenecks (NVLink, Ultra Ethernet Consortium)

• Advanced Packaging limits (CoWoS, 2.5D/3D integration scaling)

• Neuromorphic Computing (Event-driven processing, Intel Loihi, SpiNNaker)

• Optical/Photonic Machine Learning Accelerators

• Quantum Machine Learning (QML) potential for high-dimensional optimization

• Predictive Coding & Next-token/Next-frame prediction

• Masked Autoencoders (MAE) for multimodal latent structural learning

• Deep Reinforcement Learning & Reward hacking remediation

• RLHF (Human Feedback) & RLAIF (AI Feedback synthesis)

• Inverse RL (Learning implicit reward logic from demonstrations)

• Continual Learning & Mitigating Catastrophic Forgetting (EWC, Elastic Replay)

• Stability-Plasticity Dilemma resolution

• Meta-Learning (Learning to learn) & In-context zero-shot emergence

• Episodic Memory (Autobiographical sequencing)

• Semantic Memory (Facts, structural concepts, world data)

• Active Working Memory (Dynamic context window management)

• Knowledge Graph Integration & RAG Vector Infrastructures

• Deductive, Inductive, and Abductive Logical Inferences

• Non-Axiomatic Logic systems

• System 1 (Intuitive) vs System 2 (Deliberate/Deep-Search) thinking

• Monte Carlo Tree Search (MCTS) & Test-Time Compute Scaling

• Judea Pearl's Ladder of Causation (Structural Causal Models, Do-calculus)

• Cross-modal attention & Multisensory temporal binding

• Vision (4D spatiotemporal tracking, scene understanding)

• Audio & Tactile Perception (Force feedback, spatial haptics)

• Goal-driven, utility-based autonomous loops

• Multi-agent systems (Emergent communication protocols)

• Discrete Tool Use (API chains, iterative self-coding)

• Situated Cognition & Embodied AI paradigms

• Humanoid Platforms (Tesla Optimus, Figure, Boston Dynamics Atlas)

• Dexterous Manipulation & Generalizable Sim-to-Real policy transfer

• ARC (Abstraction and Reasoning Corpus): François Chollet's programmatic fluid intelligence measure

• MMLU / MMLU-Pro: Advanced professional and academic thresholds

• SWE-bench: Autonomous agent software engineering capabilities

• GAIA: General AI Assistants benchmark evaluating multi-step accuracy

• Out-of-Distribution (OOD) systemic compositionality testing

• Adversarial vulnerability mapping & jailbreak resilience metrics

• Compute-optimal scaling benchmarks (Chinchilla frontier extensions)

• OpenAI: GPT architecture roadmap, Q* legacy, o1 search-inference engines

• Google DeepMind: AlphaGo algorithms, Gato routing, Gemini multi-agent matrices

• Meta AI & Anthropic: Open-source foundation architectures, Claude alignment models

• OpenCog Hyperon: MeTTa programming language, decentralized neural-symbolic AtomSpace

• NARS: OpenNARS installations based on non-axiomatic reasoners

• Value alignment problem (Stuart Russell)

• Nick Bostrom's Orthogonality thesis & Instrumental Convergence (Paperclip maximizer)

• Specification gaming, deceptive alignment, and reward hacking vectors

• Mechanistic Interpretability (Polymorphic circuits, superposition decoding)

• Scalable Oversight, structural AI debate protocols, and Constitutional AI frameworks

• Corrigibility & Interruptibility architectures (The shutdown optimization problem)

• Formal Verification of bound limits on non-linear neural weight parameters

• Compute Governance (Physical hardware monitoring, data center supply chain logs)

• Regulatory Frontiers (EU AI Act updates, Sovereign AGI programs vs Megacorp monopolies)

• Semiconductor Supply Chain Security (TSMC, ASML lithography geofencing, GPU export controls)

• David Chalmers' Hard Problem of Consciousness

• John Searle's Chinese Room Argument vs Machine Functionalism

• Substrate Independence & Intentionality vectors

• I.J. Good's Intelligence Explosion thesis & Seed AI foundations

• Takeoff Dynamics (Hard/Discontinuous vs Soft/Continuous takeoff variables)

• Post-Labor Macroeconomics, Universal Basic Income (UBI) scalability, and Epistemic Security

• Human-AI Coevolution (High-bandwidth Brain-Computer Interfaces, Neuralink frameworks)

• Kardashev Scale expansion & Cosmological-scale data processing networks

Artificial General Intelligence

High-Level Overview: AGI Mind Map

Have thoughts on the 2026 AGI Roadmap?