Skip to main content

Lesson 1: The AI Landscape

Topics Covered
  • The Hierarchy: How AI, Machine Learning, and Deep Learning nest inside each other.
  • Machine Learning Paradigms: Supervised, Unsupervised, and Reinforcement Learning.
  • Deep Learning: Why neural networks changed how we handle unstructured data.
  • Foundation Models: The economic shift of "train once, adapt everywhere."

To understand modern AI, we must first understand where it came from. The terminology is not just a list of buzzwords; it represents the history of how we taught machines to think.

1. The Hierarchy of Intelligence​

It is helpful to visualize the field as a set of Russian nesting dolls. Each layer is a specialized subset of the one before it.

2. Artificial Intelligence (AI)​

This is the master term. It simply describes any technique that enables computers to mimic human behavior. This includes everything from the complex logic governing video game enemies in the 1980s (hard-coded rules) to the ChatGPT of today.

3. Machine Learning (ML)​

In classical programming, humans write the rules (if x > 5, do y). In Machine Learning, the machine learns the rules from data. We give it the inputs and the desired outputs, and it figures out the math in between.

There are three primary ways machines learn:

  • Supervised Learning: The machine is taught with an "answer key."
    • Example: Showing a model 10,000 emails labeled "Spam" or "Not Spam." It learns to distinguish them.
  • Unsupervised Learning: The machine is given raw data and asked to find patterns.
    • Example: Feeding customer purchase history into a model. It might group customers into "Bargain Hunters" vs. "Big Spenders" without being told those categories exist.
  • Reinforcement Learning (RL): Learning by trial and error.
    • Example: An AI playing Chess. It gets points for winning and loses points for losing, eventually learning strategies that maximize its score.

4. Deep Learning (DL)​

This is where the modern AI boom began. Traditional Machine Learning struggled with "messy" data like images, audio, or free text. Humans had to manually tell the computer what to look for (e.g., "look for two round objects to find eyes").

Deep Learning uses Neural Networks—layers of algorithms inspired by the human brain—to automate this feature extraction.

  • The Shift: We no longer tell the computer "look for eyes." We feed it a million photos of faces, and the neural network figures out that eyes are a defining feature on its own.

5. Foundation Models​

This is the current era. Before 2018, if you wanted a model to translate English to French, you built a "Translation Model." If you wanted one to summarize text, you built a "Summarization Model."

Foundation Models changed the economics of AI.

  • The Concept: Train one massive model on a huge amount of data (basically the entire internet). This model learns "general patterns" of language or vision.
  • Adaptability: This single model can then be adapted (fine-tuned) to do translation, summarization, or coding. You don't need to rebuild the brain every time; you just teach it a new skill.

6. Large Language Models (LLMs)​

LLMs are a specific type of Foundation Model trained on text. They are "Large" in two ways:

  1. Data: Trained on petabytes of text.
  2. Parameters: They contain billions of parameters (internal settings). Think of parameters like the knobs on a sound mixing board. During training, the AI adjusts these billions of knobs to get the output just right.

Training vs. Inference​

It is vital to distinguish between these two modes:

  • Training: The intensive, expensive process of teaching the model (adjusting the knobs). This happens once at the AI lab (like OpenAI or Google).
  • Inference: The act of using the model. When you send a prompt to ChatGPT and it replies, that is inference.

7. Generative AI (GenAI)​

While Foundation Models are the technology, Generative AI is the capability. It refers to AI that doesn't just analyze existing data (like a spam filter) but creates new data: writing a poem, drawing a logo, or generating code.