1.3. Reflection Checkpoint

Key Takeaways

• AI is the goal, ML and deep learning are the means. Picture nested circles: AI contains machine learning, which contains deep learning. Use the right term for the right level of specificity.
• Modern AI learns patterns from data and predicts on new inputs. Every strength and every limitation — accuracy, bias, hallucination — traces back to this mechanism.
• Generalization is the real measure of a model, not performance on data it already saw. Overfitting is memorization masquerading as success.
• Generative AI predicts content into existence rather than retrieving stored facts, which is why it can be fluent yet wrong.

Connecting Forward

You now know that AI systems behave the way their data and mechanisms dictate. That sets up Phase 2 (Principles of Responsible AI), which is essentially a structured answer to the question "given how these systems actually work, how do we build them so they're fair, safe, and trustworthy?" Every one of the six responsible-AI principles is a direct response to a limitation you just learned about.

Self-Check Questions

• Explain to a non-technical friend why a self-driving car uses deep learning but a thermostat schedule does not, even though both could be called "AI."
• A model gives biased hiring recommendations. Using the first-principles idea from this phase, where would you look first for the cause, and why?