Today's concept: AI vs ML vs deep learning vs generative AI
ChatGPT, Claude, and every chatbot you've used aren't a new species of "AI". They're the newest, narrowest subclass in a chain that starts at the 1956 Dartmouth workshop where the term was coined.
Picture four boxes, each inside the last. Devs: an inheritance chain.
1) AI is the outer box: any system that behaves as if it's intelligent. No learning required. Deep Blue, which beat Kasparov in 1997, searched millions of positions and scored them with an evaluation function hand-tuned with a grandmaster's help. It learned nothing. Still AI.
2) ML (machine learning) narrows it to systems that learn patterns from data instead of rules a human wrote. A spam filter trained on 10,000 labeled emails is the classic: you never write "flag anything saying free money", you supply examples and let the algorithm fit the rule.
3) Deep learning narrows it again: ML built on neural networks stacked many layers deep. Each layer is a function transforming the output of the one before it, and chained deep enough they learn messy, non-linear patterns nobody could hand-code.
4) Generative AI narrows it one last time: deep learning trained to produce new content (text, images, audio, code) rather than output a label or a number. What made it work at scale is the transformer, from the 2017 paper "Attention Is All You Need".
Every generative AI is AI. Almost no AI generates anything.
One thing the chain misses: agentic AI isn't a fifth subclass, it's composition, not inheritance. You wrap a loop around a generative model and give it tools, memory, and a stopping condition. The model is a field in your system, not the base class you extend, which is why "make the agent better" is usually an engineering problem before a model one.
That framing pays off fast. One RAG request crosses three boxes: an embedding model (deep learning) turns text into vectors, nearest-neighbor search pulls candidates (classic ML, zero generation), and only the last step generates. When retrieval returns junk, a smarter LLM won't save you. You're debugging the ML layer.
Same with what practitioners argue about today: mixture-of-experts (each token runs through only a slice of the network), KV caching, test-time compute. Deep-learning and systems levers, not "AI" levers.
Quick check before you scroll: A decision tree that flags spam emails based on patterns learned from 10,000 labeled examples - is it AI? ML? Deep learning? Generative AI?
Full breakdown + the answer: frankduah.me/learnings/2026-08-06-ai-vs-ml-vs-deep-learning-vs-generative-ai
New here? I post a bite-size AI / ML concept like this every day - follow me for the daily drop, and it compounds fast. Why I do it: https://lnkd.in/gK8knHDH
#AI #LLM #AIAgents #MachineLearning
The answer
It's AI and ML (it learns from data instead of hardcoded rules), but not deep learning (no layered neural network) and not generative (it classifies existing emails, it doesn't create new content).