How Computers Process Information

The photo that broke everything

It's 2015, and Google Photos just labelled a photo of two Black people as "gorillas" (BBC News, 2015). The internet explodes. Google scrambles to fix it. But here's the thing nobody asked: how does a computer even "see" a photo in the first place?

The answer is: it doesn't. Not the way you do. A computer doesn't see faces, colours, or people. It sees a grid of numbers. And if you don't understand that — if you don't understand that everything inside a computer is just numbers — you'll never understand why AI works the way it does, or why it fails the way it does.

Let's start from the very bottom.

Binary: the language of light switches

Your computer only understands two things: on and off. That's it. Every video you've watched, every song you've streamed, every text you've sent — all of it was processed using just two states.

We write these two states as 1 (on) and 0 (off). This is called binary.

Think of it like a room full of light switches. Each switch can only be up (1) or down (0). One switch isn't very useful — it can only represent two things. But give me 8 switches, and I can represent 256 different combinations. Give me 64 switches, and I can represent more combinations than there are grains of sand on Earth.

A single switch (on/off) is called a bit. Eight bits together are called a byte. Everything in computing — file sizes, internet speed, storage capacity — is measured in bytes.

How computers store text, numbers, and images

Everything is numbers. But different types of information get encoded differently.

Text: every letter has a number

Computers store text using a coding system called ASCII (or its bigger sibling, Unicode). Each letter, number, and symbol gets assigned a unique number.

So the word "Hi!" is stored as: 72, 105, 33. Three numbers. That's all the computer sees.

Numbers: just... numbers

This one's straightforward. Computers store numbers as binary. The number 42 in binary is 101010. The computer doesn't need a translation layer — numbers are its native language.

Images: grids of coloured numbers

Here's where it gets interesting. A digital image is a grid of tiny dots called pixels. Each pixel is described by three numbers: how much red, how much green, and how much blue (RGB).

A 1920x1080 photo = 2,073,600 pixels. Each pixel = 3 numbers. That's over 6 million numbers just to store one photo. And the computer has no idea what the photo is "of" — it just sees a giant spreadsheet of numbers.

This is exactly why AI "sees" the world differently than you do. When an AI model processes an image, it's doing math on millions of numbers. It doesn't see a "cat" — it sees a pattern of numbers that it's learned to associate with the label "cat."

Files and storage: putting numbers in boxes

All those numbers have to live somewhere. That's what files are — containers for data. And files live on storage devices (hard drives, SSDs, the cloud).

Think of your computer's storage like a giant filing cabinet:

The key insight: a .jpg file and a .txt file are both just numbers on a disk. The only difference is how the computer interprets those numbers. A .jpg reader knows to treat the numbers as pixel colours. A .txt reader knows to treat the numbers as letter codes. Open a .jpg file in a text editor and you'll see gibberish — because it's trying to read pixel data as letters.

Databases: the world's most organised spreadsheet

When you need to store lots of related data — like every customer's name, email, and order history — you use a database. Think of it as a giant, super-powered spreadsheet.

Databases are different from spreadsheets in one critical way: they're built for speed and scale. A spreadsheet with 1 million rows grinds to a halt. A database with 1 billion rows answers questions in milliseconds.

Why this matters for AI: Every AI model is trained on data. That data lives in databases (or files that look like databases). When you hear "we trained the model on 10 billion web pages," those web pages were stored in massive databases before being fed to the model. Understanding databases = understanding where AI's knowledge comes from.

Why all of this matters for AI

Here's the punchline of this entire module. Read it twice:

AI models don't process words, images, or sounds. They process numbers.

When you type "What's the weather today?" into ChatGPT, your words get converted to numbers before the model ever sees them. When an AI looks at a photo, it sees a grid of numbers. When it listens to your voice, it sees a waveform converted to numbers.

This is why:

• AI can process any type of data — as long as you can convert it to numbers first
• AI makes mistakes when the number conversion loses important information — like how a blurry photo (low-resolution, fewer numbers) is harder to classify
• AI doesn't "understand" your data — it finds mathematical patterns in numbers. Understanding is something you bring to the table.

Back to Google Photos

Remember that 2015 scandal? The one where Google Photos labelled a photo of two Black people as "gorillas"? Now you know exactly why it happened — and why it wasn't a simple fix.

The AI didn't see people. It saw a grid of numbers — RGB values across millions of pixels. When those numbers matched patterns the model had learned to associate with the label "gorilla," it confidently predicted that label. There was no malice, no awareness of race, no understanding of what the image meant. Just math on numbers that looked similar to other numbers the model had seen before.

Fixing it required not just changing the label, but changing what patterns the model was learning from — which means changing the training data. Google's longer-term fix was to remove the "gorilla" category from image recognition results entirely. Understanding the machine means understanding why the problems are this hard.

Key takeaways

• Everything in a computer is 0s and 1s. Text, images, audio, video — all of it gets encoded as numbers using patterns of on/off switches (bits).
• Images are grids of numbers, text is sequences of numbers, audio is waveforms of numbers. Different data types use different encoding schemes, but they all end up as numbers.
• Databases are organised number stores — like supercharged spreadsheets that can handle billions of records. AI training data lives in databases.
• You can explain why AI "sees" the world differently — because it processes numbers, not meaning. It finds mathematical patterns, not understanding.
• The quality of the number conversion matters. Low-resolution images, poorly encoded text, or noisy audio all make AI's job harder — garbage numbers in, garbage results out.