AI Types
Artificial Intelligence can be grouped by capability and deployment context. This page introduces the six types most relevant to modern enterprise and 5IR deployments: Machine Learning & Deep Learning, Generative AI, Agentic AI, Edge AI, Embodied AI, and Vertical AI.
Machine Learning & Deep Learning
The predictive backbone of modern AI for perception and forecasting.
| Component | Examples | Primary Role |
|---|---|---|
| ML / DL Frameworks | PyTorch, TensorFlow, Scikit-learn | Model development, training, evaluation |
| Model Families | CNNs, RNNs, Transformers | Perception, sequence modeling, language |
| Use Cases | Fraud detection, medical imaging, speech | Predictive analytics & pattern recognition |
Generative AI
Systems that create new content—text, code, images, audio, or video.
| Component | Examples | Primary Role |
|---|---|---|
| Model Classes | LLMs, diffusion models, text-to-video | Content generation across modalities |
| Applications | Assistive writing, design, coding | Creative acceleration & automation |
| Key Enablers | Instruction tuning, RLHF, tool-use APIs | Quality, steerability, integration |
Agentic AI
Goal-directed systems that plan, call tools, and iterate toward outcomes with feedback.
| Component | Examples | Primary Role |
|---|---|---|
| Agent Frameworks | LangChain, AutoGen, CrewAI | Task orchestration & chaining |
| Core Capabilities | Planning, tool use, memory | Autonomous, goal-seeking workflows |
| Use Cases | Research pipelines, RPA+, ops runbooks | Continuous execution & hand-offs |
Edge AI
On-device inference for low latency, privacy, and resilience.
| Component | Examples | Primary Role |
|---|---|---|
| Hardware | Jetson Orin, Apple Neural Engine | On-device acceleration |
| Workloads | Vision, speech, anomaly detection | Real-time decisions near sensors |
| Advantages | Low latency, privacy, bandwidth savings | Reliable local autonomy |
Embodied AI
AI that perceives, reasons, and acts through physical bodies—robots and autonomous machines.
| Component | Examples | Primary Role |
|---|---|---|
| Platforms | Humanoids, quadrupeds, UAVs | Physical interaction & dexterity |
| Key Enablers | RL, visuomotor policies, multimodal | Skill learning & adaptation |
| Applications | Manufacturing, logistics, inspection | Labor automation & safety |
Vertical AI
Domain-specialized AI tuned for regulated industries with compliance and auditability requirements.
| Component | Examples | Primary Role |
|---|---|---|
| Domains | Healthcare, finance, legal, climate | Industry-grade solutions |
| Specialization | Clinical NLP, credit risk, eDiscovery | Accuracy, governance, trust |
| Controls | Model cards, audit trails, guardrails | Compliance & safety-by-design |
Quick Comparison
At-a-glance comparison across all six types.
| Type | Maturity | Deployment | Latency Sensitivity | Primary Risks | Enterprise Readiness |
|---|---|---|---|---|---|
| ML / DL | High | Cloud / on-prem | Medium | Data drift, bias | Production-proven |
| Generative AI | High (fast-evolving) | Cloud APIs / fine-tuned | Low–Medium | Hallucination, drift, IP leakage | Ready with guardrails |
| Agentic AI | Medium | Cloud with tools | Medium | Unbounded actions | Pilot?prod with oversight |
| Edge AI | Medium–High | On-device / gateway | High | Model updates, fleet mgmt | Ready for real-time |
| Embodied AI | Medium | Factories / field robots | High | Safety, reliability | Scaling in pilots |
| Vertical AI | High in niches | Industry platforms | Medium | Regulatory failure | Adopted in regulated |
FAQ
How do these types of AI relate to each other?
A: These AI types often stack together. For example, a Generative AI model can be embedded within an Agentic AI system, which might run on Edge AI hardware and be further specialized into a Vertical AI solution.
What is the fastest path for enterprises to move AI into production?
A: Enterprises typically begin with Machine Learning, Deep Learning, or Vertical AI patterns. They can then add Generative AI and Agentic AI capabilities with proper governance and oversight.
What is the difference between Agentic AI and Embodied AI?
A: Agentic AI operates in digital environments by planning and using tools in software, while Embodied AI requires physical sensors and actuators to interact with the real world.
What guardrails are essential for safe AI deployment?
A: Important safeguards include strong data governance, transparent model cards, human-in-the-loop oversight, sandboxed tool use, and detailed audit logs.