Skip to content
EN

Lesson 8 of 8 · 6 min · +100 XP

What can AI do, and what can't it do?

AI is strong at tasks built on patterns: recognising images and speech, translating, summarising, drafting text and spotting trends in data. Its main limitations are that it can't guarantee truth, has no values or feelings of its own, can't take responsibility and struggles in the messy physical world. Use it as a helper and keep a human in charge.

▶ Try it

Why it matters

Knowing where AI is strong and where it is weak is the core of using it well: let it help with patterns, keep people in charge of truth, values and decisions.

In 6 minutes you'll be able to

  • Use AI for patterns: summaries, translations, ideas, finding trends.
  • Check facts yourself; AI can't guarantee that its answers are true.
  • Decisions about people stay with people who can be held responsible.

What you'll practise

  1. Four limits to know

    AI keeps getting better at patterns. These limits come from what it is.

    ✅ Truth
    AI produces likely answers, not checked ones. It can't guarantee that what it says is true.
    🧭 Values
    AI has no values, feelings or sense of fairness of its own. It reflects patterns in its data and the choices of the people who built it.
    ⚖️ Responsibility
    AI can't be held accountable. When a decision affects a person, a person must own it.
    🌍 The physical world
    Folding laundry or fixing a leaky tap is still very hard for machines, even when tests that are hard for people are easy for them.

  2. Shines or stays human?

    Where does AI shine as a helper, and where must a human stay in charge?

    • Turn a long report into five bullet points → AI shines
    • Decide whether to fire an employee → Human in charge
    • Translate a menu into another language → AI shines
    • Comfort a friend whose pet just died → Human in charge
    • Spot unusual payments among millions of transactions → AI shines
    • Choose which news is true before sharing it → Human in charge
    • Suggest ten names for a new café → AI shines
    • Sign off a medical diagnosis → Human in charge

  3. Judge the scene

    1. Lumi wrote a summary for your boss. It reads well. What now?

      • Check it against the original before sending (Answer)
      • Send it, Lumi's summaries are usually fine

    2. A friend asks Lumi whether it's okay to tell a small lie to protect someone. What can Lumi offer?

      • Different viewpoints to think about, not the answer (Answer)
      • The objectively correct moral answer

    3. A robot beats a chess champion but can't reliably fold a T-shirt. How come?

      • Rules-based games are easier for machines than the messy physical world (Answer)
      • The robot just needs a faster processor

Remember

Let AI help with patterns. Keep truth, values and responsibility with people.

Key facts

  • UNESCO's Recommendation on the Ethics of Artificial Intelligence, adopted by its 193 member states in November 2021, stresses that AI can never replace final human responsibility and accountability. · Source: UNESCO, Recommendation on the Ethics of Artificial Intelligence, 2021
  • The EU AI Act requires high-risk AI systems to be designed so that people can effectively oversee them while they are in use, including being able to override or stop them. · Source: EU Artificial Intelligence Act, Regulation (EU) 2024/1689, Article 14
  • Roboticist Hans Moravec observed in 1988 that tasks hard for humans, like playing chess, can be easy for computers, while everyday skills like seeing and moving are very hard. This is known as Moravec's paradox. · Source: H. Moravec, Mind Children, Harvard University Press, 1988
  • AI performance can change a lot with small changes in wording or situation, so an AI that does well on one test may still fail on a slightly different real task.

Updated:

Questions people ask

What are the main limitations of AI?

AI can be confidently wrong, can repeat biases in its data, has no real understanding or values, can't take responsibility for decisions and often struggles with the physical world and situations unlike its training data.

What is AI good at?

Pattern-based work: recognising images and speech, translating, summarising, drafting text, brainstorming and finding trends or unusual items in large amounts of data.

Can AI replace human judgement?

Not for decisions that involve values, responsibility or people's lives. AI can inform a decision, but a person should review it and be accountable for it.

Will AI ever be able to do everything humans can?

Nobody knows. AI keeps improving, and experts disagree about how far it will go. Today it is strong at specific pattern tasks and still weak at many things people find easy.

Related terms