Module 1: Foundations of AI · BoK IV.A
The AI family tree
Each layer is a subset of the one above: GenAI ⊂ DL ⊂ ML ⊂ AI. Agentic AI is the odd one out - it can be comprised of all categories of AI, leveraging different models depending on the task.
Each layer is a subset of the one above. The nesting order is GenAI ⊂ DL ⊂ ML ⊂ AI. Agentic AI is the odd one out - it can draw on all of them.
Machine learning (ML) → algorithms that learn patterns from data and improve over time without explicit programming.
Deep learning (DL):
- ML using multi-layered neural networks simulating the human brain
- Wins over classic ML → processes unstructured data, finds hidden patterns, can learn unsupervised
- Reduces manual feature engineering, extracts features from raw data itself
- Needs large high-quality data + heavy compute
- Examples → Google DeepMind, Tesla Autopilot
Generative AI (GenAI):
- DL models that generate new content → text, images, video
- Output is representative of training data but distinctly unique (learns "cat", draws a brand-new cat)
- Ethical concern → misuse for misinformation
- Examples → ChatGPT, Gemini, GitHub Copilot, Adobe Firefly, Claude, Microsoft Copilot
Agentic AI:
- Goal-oriented → autonomously decides, plans, executes and adapts with minimal human guidance
- Solves multistep problems with limited supervision
- Relies on patterns and likelihoods to act
- Highly adaptable → improves through reinforced learning
- Can be comprised of all categories of AI, leveraging different models depending on the task
Key terms - quick answers
What is “Machine learning (ML)”?
Algorithms that learn patterns from data and improve over time without explicit programming.
What is “Deep learning (DL)”?
ML using multi-layered neural networks; processes unstructured data and reduces feature engineering.
What is “Generative AI (GenAI)”?
DL models that generate new content representative of training data but distinctly unique.
What is “Agentic AI”?
Goal-oriented AI that autonomously plans and executes, drawing on all AI categories per task.