1.2 Traditional AI and Generative AI

The previous section described AI as a toolbox. Now we will open it and compare two common kinds of tools: those that judge existing material and those that generate new material.

For an approachable starting point, we can group many earlier or specialized uses under “recognition and decision AI.” These systems classify, predict, recommend, or recognize. They might judge whether an email is suspicious, read text in a photograph, predict an arrival time, or recommend the next song. They often label an existing item, calculate a score, or choose among existing options.

Generative AI can produce new text, images, audio, and other content in response to an instruction. It might draft an email, turn a long document into a summary, create an illustration from a description, or rewrite a message in a friendlier tone.

The two abilities often work together

This is not always a strict either-or boundary. One AI product can recognize objects in an image and then generate a caption. It can identify what a user wants and then compose an answer. For an everyday user, the most useful questions are not “Which perfect category is this?” but “What does it receive, what task does it perform, and what do I need to check?”

Imagine giving AI a rough meeting transcript:

1. It finds dates, names, decisions, and action items in the text.

2. It organizes that information into a new set of meeting minutes.

The first step leans toward recognition and extraction; the second leans toward generation. The result may look complete, but a wrong date in the original can remain wrong, and the AI may overlook an important sentence. Generation does not mean the facts have been verified.

Generation is not retrieving a finished item

Generative AI is more like a system composing an answer from learned language or image patterns than a warehouse handing out one fixed, finished answer. That is why the same request can produce different versions—and why a natural result can still be inaccurate.

Later we will see how a large language model (LLM) produces text. For now, build one practical habit: treat generated output as a draft or suggestion that can be revised, not as a decision that should automatically take effect.