- Artificial Intelligence — any system that performs tasks requiring 'intelligence'
- Machine Learning — systems that improve from data instead of hand-written rules
- Deep Learning — ML using layered neural networks
AI is an umbrella term for any software that performs tasks we associate with intelligence: playing games, recognizing images, understanding language, planning routes. It says nothing about *how* the task is solved — a chess engine built from hand-coded rules in 1997 and a 2024 chatbot are both 'AI'.
Two very different eras of AI
Rule-based AI (1960s–2000s)
- Humans write explicit if/then logic
- Works great for narrow, well-specified tasks
- Brittle: fails outside the rules it was given
- Example: expert systems, symbolic chess engines
Learning-based AI (2010s–now)
- System learns patterns from examples
- Generalizes to inputs it never saw exactly
- Needs lots of data and compute
- Example: image classifiers, LLMs
Modern AI systems you interact with — search ranking, recommendation feeds, voice assistants, chatbots — are almost all built with machine learning, and increasingly with deep learning specifically. This course is about that modern branch.
Why it matters that you know the distinction
People use 'AI' to mean wildly different things — a thermostat with a temperature curve, a spam filter, and GPT-4 all get called 'AI-powered.' When someone says a product 'uses AI', ask: is it a fixed rule, a small trained model, or a large deep learning system? The engineering, cost, and failure modes are completely different.
- Narrow AI: solves one specific task (translate text, detect fraud). Everything that exists today.
- General AI (AGI): hypothetical system matching human-level competence across virtually all tasks. Does not exist yet; a subject of active debate, not engineering.
The rest of this constellation walks down that stack: what machine learning actually is, then what deep learning adds, then the math and code underneath both.