Introduction
AI, ML, deep learning, and generative AI are nested
The story: Think of a set of nesting dolls. The biggest doll is "anything that acts smart". Open it and there is a smaller doll, "things that get smart by practicing". Inside that is "things that practice with a brain-like network". The smallest doll, right in the middle, is "things that make brand new stuff". Every inner doll is still part of the dolls around it.
In AI/AWS terms:
| Doll | Real name | What it means | Example |
|---|---|---|---|
| Biggest | Artificial intelligence (AI) | Any system that does tasks that normally need human intelligence: seeing, reasoning, learning, solving problems, deciding | Any intelligent system |
| Second | Machine learning (ML) | AI that learns from data and gets better at a task, instead of following hand-written rules | A model that flags fraudulent card payments |
| Third | Deep learning (DL) | ML that uses layered neural networks, loosely modeled on brain neurons and synapses | Amazon Rekognition analyzing photos and video |
| Smallest | Generative AI | Deep learning that creates new content from patterns it learned in training | New text, images, audio, or code |
For the exam: Each field is a subset of the one before it: AI ⊃ ML ⊃ deep learning ⊃ generative AI.
Analyzing versus creating
The story: A food critic tastes a dish and tells you if it's good, what's in it, and whether you'll like it. A chef, after tasting thousands of dishes, cooks you a new one nobody has made before. Both learned from food, but one judges and the other creates.
In AI/AWS terms: Traditional AI/ML is the critic: it analyzes and interprets existing data (is this email spam? what's in this photo?). Generative AI is the chef: it creates new content. And because the chef already knows cooking in general, you can ask for a new cuisine without sending them back to culinary school. Generative AI can adapt a deep learning model to new tasks without retraining or fine-tuning it.
For the exam: Traditional ML analyzes and predicts. Generative AI creates new content, and it can take on new tasks without retraining or fine-tuning.
Models, uses, and responsibility
The story: A big kitchen has specialist chefs: one for pastry, one for curry. Their food can be made for bread, soup, drinks, or desserts. But a restaurant still has to worry about who it serves, what goes into the food, and whether it's safe to eat.
In AI/AWS terms:
- The specialist chefs are models. Amazon Titan and Anthropic Claude make text. Stable Diffusion makes images.
- The menu is the uses: text generation, image generation, speech synthesis, and code generation.
- The food safety worries are bias, privacy, and responsible use. Generative AI can repeat unfair patterns from its training data, leak private details, or be used to cause harm.
For the exam: Titan and Claude are text models, Stable Diffusion is an image model. Generative AI raises bias, privacy, and responsible use concerns.