B2 · How Artificial Intelligence Works: Data and Learning

Types of Data

Data comes in several types: numbers, text, images, audio and video. It can be structured, meaning arranged in fixed rows and columns like a table, or unstructured, meaning it has no fixed format, like photos, messages and recordings.

8 min Beginner Jump to the activity

What you need: Nothing. This page starts from the beginning, with no programming and no formulas.

Definition

Is a selfie data? Is a voice note?

Data comes in several types: numbers, text, images, audio and video. It can be structured, meaning arranged in fixed rows and columns like a table, or unstructured, meaning it has no fixed format, like photos, messages and recordings.

In simple words

A marks register is structured: every student has a row and every subject a column. A pile of WhatsApp messages is unstructured: each one is a different length and about something different. Most data in the world is unstructured, which is why AI that reads text and sees pictures matters so much.

Why it matters

Different types of data need different kinds of AI. Tables suit simple prediction models, photos need computer vision, and text needs language AI. Knowing the type of data is the first step in choosing how to use it.

Key terms

Numerical data
Data made of numbers that you can do arithmetic with, such as height or price.
Categorical data
Data that names a group or kind, such as blood group, city or colour.
Structured data
Data arranged in fixed rows and columns, like a table.
Unstructured data
Data with no fixed layout, such as photos, voice notes and free text.
Metadata
Data about data, such as the date, place and camera settings saved with a photo.
Modality
The form data takes: text, image, audio, video or numbers.

Explained in detail

Two questions to ask about any data

You can describe any piece of data by answering two questions. First, what form does it take: numbers, text, image, audio or video? Second, how is it arranged: neatly in a table, or loose with no fixed layout? The answers tell you what tools are needed to work with it.

Numbers and categories

Inside a table, values come in two main kinds. Numerical data is made of numbers you can add, average or compare: marks, heights, prices, temperatures. Some numbers are counted in whole steps, such as the number of students in a room. Others are measured and can take any value in between, such as a weight of 42.6 kilograms. Categorical data names a group: a blood group, a state, a favourite sport. You cannot average categories; the average of "cricket" and "kabaddi" has no meaning. You can only count how many fall into each one.

Be careful with numbers that are really labels. A pin code, a phone number and a jersey number are written with digits, but adding two pin codes together tells you nothing. They are categories dressed as numbers.

Structured, unstructured and in between

Structured data sits in rows and columns where every row has the same fields: a railway timetable, a marks register, a bank statement. It is easy for a computer to search and add up. Unstructured data has no such layout: a photograph, a recorded phone call, a handwritten letter, a long review. It is commonly estimated that most of the world's data, perhaps four-fifths of it, is unstructured. Some data lies in between and is called semi-structured. An email is an example: it has fixed fields for sender, date and subject, and then a free body of text.

For most of computing history, unstructured data was very hard for machines to use. A computer could store a million photos but could not tell you which ones showed a dog. Modern AI, especially deep learning, changed this. It can find patterns directly in pixels, sound waves and sentences, which is why image search, voice typing and chatbots became possible.

Data about data

Almost every file carries hidden extra data called metadata. A photo from a phone usually stores the date and time, the model of the phone and often the exact place where it was taken. A message stores who sent it and when. Metadata is very useful for organising files, and it can also reveal more than you intended, such as your home location from a photo posted online.

Type of dataKind of AI that works with it
Tables of numbers and categoriesPrediction models, such as estimating a price
Images and videoComputer vision
TextNatural language processing
AudioSpeech and sound recognition
Several types togetherMultimodal AI, such as a chatbot that reads a photo

How it works

TypeExamplesUsually
NumbersMarks, prices, temperaturesStructured
TextMessages, reviews, booksUnstructured
ImagesPhotos, X-rays, satellite picturesUnstructured
AudioVoice notes, songs, bird callsUnstructured
VideoFilms, CCTV, sports clipsUnstructured

Try it yourself

One phone, many kinds of data

Follow what a phone stores on a normal day.

Sort the data cards into types

Which type of data is each card?

  • A product review on a shopping site

  • Daily rainfall in millimetres

  • Your height in centimetres

  • A cricket match highlight

  • A selfie

  • A voice note to a friend

Structured or unstructured?

Does the data sit neatly in rows and columns?

  • Comments under a video

  • A spreadsheet of monthly expenses

  • A folder of holiday photos

  • A train timetable

Common mistakes

  • Thinking only tables are data.
  • Thinking unstructured means useless. It is the most common kind, and AI can learn from it.
  • Forgetting that one item can hold several types, such as a video with sound and captions.

Check yourself

Question 1

1. A list of students' marks in a table is…

Question 2

2. Which type is a podcast?

Question 3

3. Why does the type of data matter?

Summary

  • Data types: numbers, text, images, audio and video.
  • Structured data fits rows and columns; unstructured data does not.
  • The type of data decides what kind of AI can learn from it.

Finished reading?