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.