What is AI? Narrow vs General Intelligence (Study Notes)

-

0.01 What is AI? Narrow vs general intelligence: detailed study notes

Learn AI by Story · Module 0: What is AI · Level: Beginner

The idea in one sentence

Artificial intelligence (AI) is software that does tasks that normally need human intelligence, and every AI that exists today is narrow: very good at one job, and helpless at the others.

The story from the video

Asha's phone unlocks with her face, translates her words and picks her music. She asks whether that is really AI. It is. But a chess program that beats world champions cannot spot a cat or make tea, and that is exactly what "narrow" means.

Key definitions

Term

Definition

In plain English

Artificial intelligence (AI)

The field of building computer systems that perform tasks that normally require human intelligence, such as perception, language, reasoning and decision making.

Software that does smart-looking things.

Intelligence

The ability to learn from experience, solve problems and adapt to new situations.

Being able to work things out and cope with new problems.

Task

A specific job a system must do, such as recognising a face or choosing a route.

One clear job.

Data

Recorded information: photos, sounds, text, numbers, clicks.

The raw material an AI works with.

Pattern

Something that repeats or goes together in data, for example the shapes that usually make a face.

A regularity the AI can spot.

Narrow AI (weak AI, ANI)

AI built for one task or a small family of tasks.

A specialist.

General AI (AGI)

A hypothetical AI that can learn and perform any intellectual task a human can.

A flexible all-rounder, like a person. It does not exist yet.

Superintelligence

A hypothetical AI far smarter than the best humans in nearly every area.

Beyond human level. Only a possibility people debate.

Algorithm

A step-by-step procedure a computer follows.

A recipe.

Model

The learned program that turns an input into an output.

The trained "brain" for one task.

Detailed explanation

1. What counts as AI?

The word AI is used loosely, so it helps to ask a simple question: is this software doing something that, if a person did it, we would call intelligent? Recognising a face, understanding a spoken sentence, choosing a good move in a game, translating between languages and recommending a film all pass that test. A calculator adding numbers does not feel like AI today, even though it once would have, which leads to a useful idea called the AI effect: once a machine can do something reliably, people stop calling it AI.

2. The sense, think, act loop

Most AI systems follow the same loop:

  1. Sense: take in data, such as a photo, a sound or a sentence.
  2. Think: find patterns in that data and compare them with what the system has learned or been given.
  3. Act: produce an output, such as an answer, a label or a move.

Face unlock senses a camera image, thinks by comparing the face pattern with the stored one, and acts by unlocking or not.

3. Narrow AI

Narrow AI is designed and trained for one task or a tight group of tasks. It can be better than any human at that task, for example at chess or at spotting certain features in medical scans, but it cannot move its skill to a different area. A chess engine does not know what a cat is, and a cat detector cannot play chess. Every AI product in use today is narrow: face unlock, map routing, spam filters, recommendation feeds, voice assistants and translation.

4. General AI

General AI, also called artificial general intelligence or AGI, would be able to learn nearly any intellectual task a human can, transfer what it learns between areas, and cope with situations it has never seen. Nobody has built it. Researchers disagree about how far away it is, what it would need, and even how to test for it, so treat any confident date with caution.

5. Where do chatbots fit?

Large language models look wide because they can write, summarise, translate and answer questions about many topics. They are much broader than older narrow systems, but they still make confident mistakes, struggle with some kinds of reasoning, and do not learn new skills from a single experience the way a person does. Most experts still classify them as narrow in the sense that they are not human-level across everything.

6. Strong and weak: a note on naming

You may meet the terms weak AI (narrow) and strong AI (general). "Weak" does not mean bad: a narrow AI can be extremely powerful at its job. It only means limited in scope.

What the video did not explain

  • AI is a field, not one technology. It includes rule-based systems, machine learning, deep learning, search and planning, robotics and more. Video 0.02 shows how machine learning and deep learning nest inside AI, and 0.04 compares writing rules with learning from data.
  • A short history. The term was coined at the 1956 Dartmouth workshop, and AI has gone through hype cycles and winters. This is the topic of video 0.03.
  • The Turing test. Alan Turing proposed in 1950 that a machine could be called intelligent if a person chatting with it could not reliably tell it from a human. It is a famous idea, but today many researchers consider it a poor measure of real intelligence.
  • Types of AI by capability. Reactive machines respond only to the present input (a chess engine). Limited-memory systems use past data (self-driving cars). "Theory of mind" and "self-aware" AI are theoretical categories that do not exist.
  • Limits of today's AI. AI can be confidently wrong, can inherit bias from its training data, needs large amounts of data or computing power, and usually cannot explain its reasoning.
  • Why the narrow versus general split matters. It sets realistic expectations: use AI as a powerful tool for specific jobs, and be careful with claims that it understands or thinks like a person.

Worked example

Take a chess engine, a cat detector and a translation app, and score each from 0 (useless) to 10 (expert) on four tasks. The scores below are illustrative.

System

Chess

Spot a cat

Translate

Make tea

Chess engine

10

0

0

0

Cat detector

0

9

0

0

Translation app

0

0

8

0

Each row is high in exactly one column. That is narrow AI. A general AI would score well across the whole table and could pick up a new column, such as making tea, without being rebuilt. No such system exists.

Common mistakes and myths

  • "AI thinks like a person." No. Today's AI finds patterns in data and does not have human understanding, feelings or common sense.
  • "AI is always right." No. It can be wrong, and sound very sure while it is wrong.
  • "General AI is already here." No. Nobody has built it.
  • "A smart-looking answer means understanding." Not necessarily. Fluent output can come from pattern matching.
  • "AI and robots are the same thing." No. Most AI is pure software with no body. A robot is a machine that can sometimes use AI.

Where it is used

  • Phone: face unlock, photo search, keyboard prediction, voice assistant.
  • Navigation: route and traffic prediction.
  • Entertainment and shopping: recommendations for videos, music and products.
  • Communication: spam filtering, translation, autocorrect.
  • Health and industry: reading scans, spotting faults on production lines, fraud detection in banks.

All of these are narrow AI systems, each doing one job well.

Self-check questions

  1. Define artificial intelligence in one sentence. Software that does tasks that normally need human intelligence.
  2. What are the three steps of the sense, think, act loop? Take in data, find patterns, produce an answer or action.
  3. What does "narrow" mean in narrow AI? Built for one task or a small group of tasks, and unable to transfer the skill elsewhere.
  4. Does general AI (AGI) exist today? No. It is hypothetical.
  5. Why can a chess engine that beats champions not spot a cat? It was built and trained for chess only, and has no skill outside that job.
  6. Is "weak AI" a bad AI? No. It means limited in scope, not poor in quality.
  7. Name three AI systems on your phone. For example face unlock, voice assistant, photo search, keyboard prediction.
  8. Why is it risky to trust an AI answer blindly? AI can be confidently wrong and can inherit bias from its data.
  9. What is the AI effect? Once a machine does something reliably, people stop calling it AI.
  10. A system scores 9 in translation and 0 in everything else. Narrow or general? Narrow.

Key takeaways

  • AI is software that does tasks that normally need human intelligence.
  • Most AI senses data, finds patterns and acts.
  • Every AI today is narrow: strong at one task, helpless at others.
  • General AI and superintelligence are hypothetical.
  • Be careful about claims that AI understands or thinks like a person.

Go deeper and what next

  • Read about the Turing test and the Dartmouth workshop.
  • Think of five apps you use and write the single task each one does well.
  • Next in the roadmap: 0.02 AI vs ML vs DL vs generative AI: how they nest.

Comments

Post a Comment

Popular posts from this blog

The five kinds of learning: study notes

-
0.05 The five kinds of learning: detailed study notes Learn AI by Story · Module 0: What is AI · Level: Beginner The idea in one sentence Machines learn in five main ways, depending on the feedback they get: labels, no labels, a few labels, labels made from the data itself, or rewards. The story from the video Neha is learning to cook and meets all five styles: a teacher tasting every dish (supervised), sorting spices into groups herself (unsupervised), a few labelled dishes among many unlabelled ones (semi-supervised), guessing a hidden word in a recipe (self-supervised), and learning from rewards when a dish turns out great (reinforcement). Key definitions Term Definition In plain English Learning (in ML) Improving performance on a task by using data or experience. Getting better from examples. Label The correct answer attached to an example (for instance cat or not cat). The answer key. Supervised learning Learning a mapping from inputs to outputs using examples that all have labels...
Read more

Rules vs learning: study notes

-
0.04 Rules vs learning: two ways to build smart software: detailed study notes Learn AI by Story · Module 0: What is AI · Level: Beginner The idea in one sentence There are two ways to build smart software: write the rules yourself, or show the machine examples and let it learn the rules, and real products usually mix both. The story from the video Gopal's inbox is full of junk mail and he wants a spam filter. First he writes rules (if it says lottery, it is spam), but spammers write "l0ttery" and a friend's email about the office lottery is blocked by mistake. Then he collects ten thousand emails, labels each one spam or not spam, and lets an algorithm learn the patterns. The learned filter keeps improving as it sees more examples. Key definitions Term Definition In plain English Rule-based system Software whose behaviour is defined by explicit rules written by a person. If this, then that. Machine learning Methods in which a system learns a rule or function from dat...
Read more