COURSE 01L0100% FREE
Verified 2026-08-10

What AI can and cannot do

Today’s AI is usually **strong on familiar patterns** and **weak on messy edge cases**. It can be fast, tireless, and surprisingly fluent. It can also be confidently wrong, biased by its training data, or brittle when the world shifts.

What it is

Today’s AI is usually strong on familiar patterns and weak on messy edge cases. It can be fast, tireless, and surprisingly fluent. It can also be confidently wrong, biased by its training data, or brittle when the world shifts.

Why it matters

Fear shrinks when limits are named. Hype shrinks when strengths are named honestly. This chapter is the bridge to Course 29 (dangers) and Course 08 (hallucinations).

How it works (plain)

Strengths you will often see:

  • Sorting, ranking, and recommending at huge scale
  • Drafting and transforming text
  • Recognizing patterns in images or audio (within training)
  • Helping programmers with boilerplate (with review)

Limits you should expect:

  • No reliable “understanding” like a careful human expert
  • Weakness on brand-new situations unlike the training data
  • Invented facts (hallucinations)
  • Sensitivity to wording
  • Potential unfair errors across groups (Course 19)

Everyday example

A model can summarize a long PDF you provide, then invent a citation that looks real. Useful draft + mandatory human check.

Try it

Ask a chatbot a question you already know the answer to—then a slightly twisted version. Notice where fluency stays high while accuracy drops.

Myths

⚠️ Myth: If it passed a hard test once, it is generally intelligent.
✓ Reality: Benchmarks are narrow; contamination and coaching effects exist (Course 18).
⚠️ Myth: AI cannot be dangerous because it “isn’t conscious.”
✓ Reality: Harm comes from misuse and mistakes, not from movie consciousness (Course 29).

Sources