An interactive master reference of all 19 AI system types — taxonomy, classification, architectures, market data, and cross-system comparisons for 2026.
~18 min read · Interactive ReferenceA comprehensive taxonomy of every major AI paradigm in the 2026 landscape — grouped by primary function.
Autonomous goal-driven agents that plan, use tools, maintain memory, and execute multi-step workflows.
Discovers patterns, surfaces insights, and explains data through BI, augmented analytics, and NLQ.
Self-managing systems — auto-scaling, autopilot, algorithmic trading — within fixed boundaries.
Principled uncertainty quantification — MCMC, GPs, Bayesian deep learning, probabilistic programming.
Neural + symbolic integration — knowledge-guided networks, LLM+KG, neural theorem provers.
Enables natural multi-turn dialogue via chatbots, voice assistants, and conversational search.
Genetic algorithms, neuroevolution, NAS — population-based optimisation inspired by evolution.
SHAP, LIME, Grad-CAM, counterfactuals — making black-box models interpretable.
Train models without sharing raw data — FL, DP, MPC, homomorphic encryption, TEEs.
Creates novel content — text, images, video, audio, code, 3D, molecules — from learned distributions.
Fuses vision, language, audio, depth — VLMs, sensor fusion, any-to-any models.
LP, MIP, constraint programming — mathematical optimisation for logistics, scheduling, routing.
Robots, autonomous vehicles, drones — AI that senses, plans, and acts in the physical world.
Classifies, scores, forecasts, and detects anomalies from structured and unstructured data.
Stateless stimulus-response systems — rule engines, PLCs, game minimax — no memory or learning.
Collaborative filtering, neural retrieval, RAG — personalised content and search.
Learns by trial-and-error in environments — maximising cumulative reward via policy optimisation.
PINNs, GNNs, neural operators — accelerating drug discovery, materials science, climate modelling.
Knowledge graphs, ontologies, expert systems, theorem provers — explicit symbolic reasoning.
Artificial Intelligence
│
├── 1. Agentic AI
├── 2. Analytical AI
├── 3. Autonomous AI (Non-Agentic)
├── 4. Bayesian / Probabilistic AI
├── 5. Cognitive / Neuro-Symbolic AI
├── 6. Conversational AI
├── 7. Evolutionary / Genetic AI
├── 8. Explainable AI (XAI)
├── 9. Federated / Privacy-Preserving AI
├── 10. Generative AI
├── 11. Multimodal Perception AI
├── 12. Optimisation / Operations Research AI
├── 13. Physical / Embodied AI
├── 14. Predictive / Discriminative AI
├── 15. Reactive AI
├── 16. Recommendation / Retrieval AI
├── 17. Reinforcement Learning AI
├── 18. Scientific / Simulation AI
└── 19. Symbolic / Rule-Based AI
Every AI system can be classified along these 9 orthogonal dimensions, revealing its capabilities and limitations.
Supervised · Unsupervised · Self-Supervised · Reinforcement · Few-Shot · Zero-Shot · Meta-Learning
Statistical/Connectionist · Symbolic/Logical · Hybrid Neuro-Symbolic · Probabilistic · Evolutionary
Classification · Regression · Generation · Actions · Explanations · Decisions · Recommendations
Text · Image · Video · Audio · Tabular · Graph · Point Cloud · Molecular · Time Series
Human-in-the-Loop · Human-on-the-Loop · Human-out-of-the-Loop · Fully Autonomous
Cloud API · On-Premise · Edge · Embedded · Hybrid · Federated
White-Box · Glass-Box · Grey-Box · Black-Box · Post-Hoc Explainable
Stateless/Reactive · Session-Stateful · Persistent Memory · Continual Learning
Single-Device · Multi-Device · Cluster · Geo-Distributed · Federated Cross-Silo
AI systems can also be classified along these cross-cutting dimensions:
| Dimension | Options |
|---|---|
| Capability Level | Narrow AI (ANI) → General AI (AGI) → Superintelligent AI (ASI) |
| Learning Paradigm | Supervised → Unsupervised → Semi-supervised → Self-supervised → Reinforcement |
| Deployment Model | Cloud → Edge → On-device → Hybrid |
| Openness | Proprietary → Open-weight → Fully open-source |
| Input Modality | Text → Image → Audio → Video → 3D → Tabular → Molecular → Multimodal |
| Output Modality | Text → Image → Audio → Video → Label → Score → Action → Simulation |
| Regulation Tier (EU AI Act) | Minimal risk → Limited risk → High risk → Unacceptable risk |
| Reasoning Style | Statistical → Logical → Causal → Probabilistic → Hybrid |
| Memory Type | None (Reactive) → In-context → External (RAG) → Long-term persistent |
Side-by-side comparison of all 19 AI system types across core purpose, primary output, and key differentiator.
| # | AI System Type | Core Purpose | Primary Output | Key Differentiator |
|---|---|---|---|---|
| 1 | Agentic AI | Autonomous goal execution | Actions, tool calls, workflows | Plan → Act → Observe loop with tool use |
| 2 | Analytical AI | Discover & explain patterns | Insights, dashboards, narratives | Augmented analytics & NLQ interfaces |
| 3 | Autonomous AI | Self-management | Automated operations | Operates within fixed authority boundaries |
| 4 | Bayesian AI | Reasoning under uncertainty | Posterior distributions | Principled uncertainty quantification |
| 5 | Cognitive AI | Neuro-symbolic reasoning | Grounded inferences | Neural perception + symbolic logic |
| 6 | Conversational AI | Natural dialogue | Text/voice responses | Multi-turn context-aware conversation |
| 7 | Evolutionary AI | Population-based search | Optimised solutions | Selection, crossover, mutation over generations |
| 8 | Explainable AI | Model interpretability | Explanations, attributions | Post-hoc or inherent transparency |
| 9 | Federated AI | Privacy-preserving learning | Shared models | Data never leaves local device |
| 10 | Generative AI | Create novel content | Text, images, video, audio, code | Produces new artifacts from learned distributions |
| 11 | Multimodal AI | Cross-modal understanding | Fused representations | Multiple modalities jointly processed |
| 12 | Optimisation AI | Mathematical optimisation | Optimal decisions | Exact solvers for LP/MIP/CP formulations |
| 13 | Physical AI | Real-world embodiment | Motor commands, navigation | Sense-plan-act in physical world |
| 14 | Predictive AI | Classify and forecast | Scores, labels, predictions | Discriminative modelling on labelled data |
| 15 | Reactive AI | Stimulus-response | Immediate actions | Stateless — no memory, no learning |
| 16 | Recommendation AI | Personalised ranking | Ranked item lists | Collaborative + content-based filtering |
| 17 | RL AI | Sequential decision-making | Policies, actions | Trial-and-error reward maximisation |
| 18 | Scientific AI | Accelerate discovery | Predictions, simulations | Physics-informed architecture |
| 19 | Symbolic AI | Formal reasoning | Logical conclusions, proofs | Explicit knowledge representation & rules |
| # | AI System Type | Core Purpose | Primary Output | Key Differentiator |
|---|---|---|---|---|
| 1 | Agentic AI | Pursue goals autonomously | Actions, decisions, workflows | Plans, reasons, uses tools, self-corrects |
| 2 | Analytical AI | Extract insights from data | Dashboards, explanations, reports | Explains and surfaces meaning in data |
| 3 | Autonomous AI | Self-directed domain tasks | Domain-specific task completion | Narrow autonomy without goal decomposition |
| 4 | Bayesian / Probabilistic AI | Reason under uncertainty | Distributions, credible intervals | Explicitly models what is known vs. unknown |
| 5 | Neuro-Symbolic AI | Combine learning + reasoning | Structured reasoning outputs | Bridges neural and symbolic approaches |
| 6 | Conversational AI | Enable natural dialogue | Text or speech responses | Manages multi-turn human-machine dialogue |
| 7 | Evolutionary / Genetic AI | Evolve optimal solutions | Designs, architectures, trade-offs | Population-based, gradient-free search |
| 8 | Explainable AI (XAI) | Make AI transparent | Explanations, attributions | Answers "why" the AI decided what it did |
| 9 | Federated / Privacy AI | Learn without centralising data | Distributed model updates | Data stays local; privacy by design |
| 10 | Generative AI | Create new content | Text, image, video, audio, code | Produces novel original content |
| 11 | Multimodal Perception AI | Understand multi-sensory input | Classifications, transcriptions | Perceives the world — does not generate it |
| 12 | Optimisation / OR AI | Find optimal constrained solutions | Schedules, routes, allocations | Solves explicit mathematical formulations |
| 13 | Physical / Embodied AI | Act in the physical world | Physical actions via robots | Has a body; operates in real-world space |
| 14 | Predictive AI | Classify and forecast | Labels, scores, probabilities | Predicts from historical patterns |
| 15 | Reactive AI | Respond to current input | Fixed, deterministic responses | No memory, no learning, purely stateless |
| 16 | Recommendation / Retrieval AI | Surface relevant items | Ranked lists, retrieved passages | Surfaces existing items — does not create |
| 17 | Reinforcement Learning AI | Learn via reward signals | Optimal policy / strategy | Learns from interaction, not labelled data |
| 18 | Scientific / Simulation AI | Accelerate scientific discovery | Predictions, models, simulations | Purpose-built for formal scientific domains |
| 19 | Symbolic / Rule-Based AI | Reason from explicit rules | Inferences, decisions | Fully logic-based and interpretable |
2024 market sizes and 2030 projections across all major AI system categories.
Modern AI products rarely use a single paradigm — they combine multiple system types. Here are the most common real-world combinations.
Combines: Generative + Conversational + Agentic + RL (RLHF)
Foundation model generates; conversational layer manages dialogue; agent layer uses tools; RLHF aligns behaviour.
Combines: Physical + Multimodal + Predictive + RL + Reactive
Sensor fusion perceives; predictive models forecast; RL optimises policy; reactive controls handle real-time safety.
Combines: Recommendation + Predictive + RL + Analytical
Collaborative filtering recommends; predictive models score; RL explores; analytics measure engagement.
Combines: Scientific + Generative + Evolutionary + Bayesian
GNNs predict structure; generative models design candidates; evolutionary search explores space; Bayesian uncertainty guides selection.
Combines: Generative + Conversational + Recommendation (Retrieval) + XAI
Retrieval finds documents; generator produces answers; conversational layer maintains dialogue; XAI provides citations.
Combines: Physical + Autonomous + Analytical + Predictive + Reactive
Robots execute; autonomous systems self-manage; analytics surface insights; predictive models forecast failures; PLCs react.
Combines: Predictive + XAI + Federated + Conversational + Analytical
Models predict risk; XAI explains decisions; federated protects patient privacy; NLQ surfaces insights.
Combines: Reactive (rules) + Predictive (ML) + Autonomous (auto-block) + XAI + Analytical
Rules catch known patterns; ML detects novel fraud; autonomous layer blocks in real-time; XAI explains to analysts.
Most real-world systems combine multiple AI types simultaneously:
| Real-World System | AI Types Combined |
|---|---|
| Tesla FSD | Physical AI + Autonomous AI + RL + Multimodal Perception |
| ChatGPT (with tools) | Generative AI + Conversational AI + Agentic AI |
| AlphaFold 3 | Scientific AI + Neuro-Symbolic AI + RL |
| Waymo | Physical AI + Autonomous AI + Predictive AI + Multimodal Perception |
| Salesforce Agentforce | Agentic AI + Conversational AI + Predictive AI |
| Google Search | Predictive AI + Analytical AI + Generative AI |
| Apple Face ID | Multimodal Perception AI + Federated / Privacy AI |
| FICO Credit Score | Predictive AI + Explainable AI |
| AlphaGo / AlphaZero | Reinforcement Learning AI + Neuro-Symbolic AI |
| NVIDIA Omniverse | Scientific AI + Physical AI + Generative AI |
| GitHub Copilot | Generative AI + Agentic AI + Neuro-Symbolic AI |
| Siri / Alexa | Conversational AI + Reactive AI + Predictive AI |
| Netflix | Recommendation AI + Predictive AI + Analytical AI |
| Amazon Supply Chain | Optimisation AI + Predictive AI + Analytical AI |
| Bayesian Clinical Trials | Bayesian AI + Predictive AI + Scientific AI |
| Google NAS (AutoML) | Evolutionary AI + Predictive AI + Optimisation AI |
Essential terms spanning all 19 AI system types.
| Term | Definition |
|---|---|
| Foundation Model | A large model trained on broad data at scale, adaptable to many downstream tasks |
| Fine-Tuning | Adapting a pre-trained model to a specific domain or task |
| Inference | Running a trained model to produce outputs on new input data |
| Hallucination | When an AI generates confident but factually incorrect outputs |
| Grounding | Connecting AI outputs to real, verifiable facts or data sources |
| Latent Space | The compressed internal representation a model learns during training |
| Transfer Learning | Applying knowledge learned in one domain to a different but related domain |
| Multimodal | Involving or processing more than one type of data (e.g., text + image) |
| Alignment | Ensuring AI behaviour matches human values, intentions, and safety requirements |
| Emergent Behaviour | Capabilities that arise in large models not explicitly trained for |
Animation infographics for The Complete AI Systems Landscape — Master Reference — overview and full technology stack.
Animation overview · The Complete AI Systems Landscape — Master Reference · 2026
Animation tech stack · Hardware → Compute → Data → Frameworks → Orchestration → Serving → Application · 2026
Each card links to a comprehensive interactive deep-dive — complete with architecture diagrams, technology stacks, use cases, benchmarks, market data, and more.
Autonomous goal-driven agents that plan, use tools, maintain memory, and execute multi-step workflows.
Discovers patterns, surfaces insights, and explains data through BI, augmented analytics, and NLQ.
Self-managing systems — auto-scaling, autopilot, algorithmic trading — within fixed boundaries.
Principled uncertainty quantification — MCMC, GPs, Bayesian deep learning, probabilistic programming.
Neural + symbolic integration — knowledge-guided networks, LLM+KG, neural theorem provers.
Enables natural multi-turn dialogue via chatbots, voice assistants, and conversational search.
Genetic algorithms, neuroevolution, NAS — population-based optimisation inspired by evolution.
SHAP, LIME, Grad-CAM, counterfactuals — making black-box models interpretable.
Train models without sharing raw data — FL, DP, MPC, homomorphic encryption, TEEs.
Creates novel content — text, images, video, audio, code, 3D, molecules — from learned distributions.
Fuses vision, language, audio, depth — VLMs, sensor fusion, any-to-any models.
LP, MIP, constraint programming — mathematical optimisation for logistics, scheduling, routing.
Robots, autonomous vehicles, drones — AI that senses, plans, and acts in the physical world.
Classifies, scores, forecasts, and detects anomalies from structured and unstructured data.
Stateless stimulus-response systems — rule engines, PLCs, game minimax — no memory or learning.
Collaborative filtering, neural retrieval, RAG — personalised content and search.
Learns by trial-and-error in environments — maximising cumulative reward via policy optimisation.
PINNs, GNNs, neural operators — accelerating drug discovery, materials science, climate modelling.
Knowledge graphs, ontologies, expert systems, theorem provers — explicit symbolic reasoning.