Module 1 Summary: Describe Before You Label

Consolidate the distinctions, evidence questions, and TRACE method you will use throughout the rest of the course.

What you can do now

At the start, we looked at a friendly assistant, a scoring model, and a rule-based service.

You now know that a product's look and name do not tell us how it works.

You can ask:

• What is the goal?
• What goes into the system?
• What method does it use?
• What comes out?
• What can change because of the result?

Then you can name only the parts supported by facts.

Main ideas

Rules and machine learning

Rules are written by people. Machine-learning models are trained from data or examples. One product may use both.

Machine learning and deep learning

Deep learning is one type of machine learning. It does not always create content.

Generation and retrieval

Generation creates content. Retrieval finds existing material.

Models and full systems

A model is one part. The full system may also use rules, search, tools, saved data, access controls, approval, logs, and people.

Assistants and agents

An assistant may write one answer and stop. An agent can work towards a goal through several steps and actions.

Training and use

Training builds or changes a model. Inference uses the trained model on a new input.

Your checklist

When you see an AI-powered product, use these steps.

Step 1: Describe it

• Goal.
• Inputs.
• Method.
• Output or action.
• Effect.

Step 2: Use TRACE

• Task: What work does it do?
• Result: What does it produce or change?
• Actual method: What rules, models, search, tools, or steps are proven?
• Controls: What access limits, approval, review, logs, and stop rules exist?
• Evidence gap: What important fact is missing?

Step 3: Name the supported parts

You may use more than one label:

• Fixed rules or calculation.
• Retrieval or search.
• Machine learning.
• Deep learning.
• Generative AI.
• Tool-using system.
• AI agent.
• System with many combined parts.
• Human-reviewed system.
• Not enough information.

Step 4: Ignore weak proof

Do not classify a system only from:

• Its name.
• Human-like words.
• Speed.
• A chat screen.
• A useful answer.
• A claim that it is autonomous.

The knowledge check

The check has ten questions. It tests your judgement, not your memory of exact sentences.

You need 8 out of 10. You can try again after reading the feedback.

Take your time. If an answer feels obvious, ask yourself which fact supports it.

What comes next

In Module 2, we will look inside generative AI. We will learn about tokens, context, prediction, and tools.

We will also answer an important question:

Why can an AI answer sound clear and confident but still be wrong?

Keep our main rule:

Describe the system before you give it a label.