The
Technology
of Media
Your phone turns light into numbers, squeezes those numbers down, splits them into packets and sends them through cables under the ocean. Then software decides who sees it. This module opens up the machinery, including the AI tools that can now fake almost anything, and how to check what's real.
- 00First Principlesp.2
- 01Phone camera sensorsp.4
- 02Compression & codecsp.5
- 03Streaming & the internetp.6
- 04Platforms & recommendationsp.7
- 05AI media & deepfakesp.8
- ✦Self-check · Make it · 5 Questions · Side Questp.9
You'll need: pencil, calculator, a phone with a camera, a web browser and some squared paper.
Principles
A first principle is a basic truth you can build on. All digital media is numbers: captured, squeezed, moved and sorted. Start with what you already know about the tech you use every day.
What do I already know?
What is necessary?
Everything in this module is built from four bricks. Rate yourself on each one, then check the example to see if your rating holds up.
What is the objective?
By the end, you'll be able to explain how media tech works, from sensor to screen, and check whether media is real.
- 01Explain how a camera sensor and Bayer filter capture colour.
- 02Compare lossless and lossy compression and calculate file sizes.
- 03Describe how streaming video reaches your screen.
- 04Explain how recommendation systems rank content.
- 05Use SIFT and other tools to check AI and fake media.
- 06Fact-check three real posts and make a fact-check card.
The list on the left is ours. Now make it yours. What do you want out of this?
How much do you know about the technology of media right now? Shade the boxes in pencil. You'll come back at the end (p.9) and shade it again in pen.
Why start this way? When you know what you already have (1), what you need (2) and where you're going (3), you learn faster and you notice the gaps. Every module on the Media Path starts with these same three questions.
Inside the sensor
Behind your phone's lens is an image sensor: a chip covered in millions of tiny light-catchers called photosites. Each one measures only how much light hits it, not what colour. So how do we get colour?
A grid of colour filters sits on top, so each photosite sees only red, green or blue light. The most common pattern is the Bayer filter, patented by Bryce Bayer at Kodak in 1976. It has twice as many green filters as red or blue, because our eyes are most sensitive to green.
Software then guesses the two missing colours at each spot using its neighbours (called demosaicing). Phones also snap several frames in a split second and blend them, which is how night mode and HDR usually work. That's called computational photography.
A 12-megapixel sensor might be 4,000 photosites wide and 3,000 tall.
In a dim room, take the same photo with night mode on and off (or with and without flash). Zoom in on both. What's different: brightness, noise (grain), blur?
After demosaicing, each of the 12,000,000 pixels stores red, green and blue, 1 byte each. How many megabytes is that? Most phone photos are only a few MB. What do you think is going on? (Next page!)
Phone sensors are tiny, so each photosite catches little light, which means more grain (noise) in the dark. Many phones have two or three cameras, each with its own lens and sensor: wide (main), ultra-wide and telephoto (zoom). The phone switches between them, or blends them, as you zoom.
Making files smaller
Raw media is huge. Compression makes it smaller. Lossless compression shrinks a file and gets back every bit (like ZIP or PNG). Lossy compression throws away detail you probably won't notice (like JPEG photos and MP3 or AAC audio).
A simple lossless trick: instead of listing every pixel, count runs of the same colour.
Nothing lost. Works great on images with big flat areas; badly on noisy ones.
One uncompressed 1080p frame: 1,920 × 1,080 pixels, 3 bytes each.
A codec (coder-decoder) is the recipe for compressing and playing video. Common ones are H.264, H.265 (HEVC) and AV1. A big trick: store a full picture only now and then (a keyframe), and for the frames in between, store only what changed.
On squared paper, draw an 8 × 8 pixel picture (a heart, a face) in black and white. Write each row as run-length code. How many symbols did you save in total? Which rows saved the most?
Raw 1080p video is about 187 MB/s. Suppose a stream uses 5 megabits per second (5 Mbps). There are 8 bits in a byte. How many MB/s is the stream, and roughly how many times smaller than raw?
How streaming works
When you press play, the video doesn't arrive as one file. It's cut into chunks a few seconds long, and each chunk is split into packets that find their own way across the internet.
Most data that crosses oceans travels as light through undersea fibre-optic cables, not satellites. Big platforms keep copies of popular videos on content delivery networks (CDNs): servers spread around the world, so your video comes from somewhere nearby.
Adaptive bitrate streaming keeps several versions of each chunk at different qualities. If your connection slows down, the player grabs the next chunk at lower quality. That's why video sometimes goes blurry instead of stopping.
Internet speed is in megabits per second (Mbps). File sizes are in megabytes (MB). 1 byte = 8 bits.
On YouTube on a computer, right-click a playing video and choose "Stats for nerds." Write down the resolution and connection speed. Change the quality setting. What changes?
A phone plan includes 10 GB of data a month. Streaming uses about 2.25 GB per hour (from above). How many hours of video is that? If you stream on cellular for 30 minutes a day, when do you run out?
How a feed is built
A platform is apps, servers and databases, plus recommendation systems that pick what to show. They work roughly like a funnel: from huge numbers of posts, pick a few thousand candidates you might like, then rank them.
Common methods: collaborative filtering ("people who liked what you liked also liked this") and content-based matching ("this is similar to what you watched"). Signals include watch time, likes, shares, follows and skips.
Engineers constantly run A/B tests: show version A to some users and B to others, and keep whichever performs better on what they measure. What a platform chooses to measure shapes what you see.
Score = 3 × shares + 2 × comments + 1 × likes
Y ranks higher with fewer likes. Comments and shares count more, so posts that start arguments can win.
Made-up data. You liked videos A, B, C. Jordan liked A, B, C, D. Priya liked E, F. Using collaborative filtering, what should the app suggest to you, and why?
Write your own scoring rule for a feed that's good for its users, not just for time spent. Which signals would you use (maybe "saved for later" or "said this was helpful")? Which would you leave out? Score Posts X and Y with your rule.
Check before you trust
Generative AI can now create realistic images, voices and video from a text prompt. A deepfake is media made or changed with AI to show someone saying or doing something they never did. Old tips like "look at the hands" are getting less reliable as the tools improve, so the best check is about where it came from, not just how it looks.
S · Stop. Notice your reaction before you share.
I · Investigate the source. Who posted it? What's their track record?
F · Find better coverage. Are trusted outlets reporting it?
T · Trace it to the original. Where did the image or quote first appear?
Tools that help
Reverse image search (Google Lens, TinEye) finds older copies of a picture.
Content Credentials (the C2PA standard) can show how a file was made and edited, if the creator's tools added them. Missing credentials don't prove a fake.
MediaSmarts, Canada's centre for digital media literacy, has free guides like "Break the Fake."
"A moose just walked into West Edmonton Mall!! 😱" (photo attached)
Pick a dramatic photo from a news site or your feed. Run a reverse image search. When and where does it first appear?
If AI can make anything look real, then "it looks real" isn't evidence any more. What is good evidence? Write three rules you'd teach a friend or family member.
Quick self-check
Six questions, covering every stop. Show your thinking. Answers are upside down at the bottom; no peeking till you're done.
1 · Why does a Bayer filter have twice as many green squares?
2 · Is a ZIP file lossless or lossy? What about a JPEG?
3 · How long does a 30 MB file take to download at 60 Mbps?
4 · What does a CDN do?
5 · Score a post with 6 likes, 1 comment and 3 shares using 3 × shares + 2 × comments + likes.
6 · What do the four letters in SIFT stand for?
Re-shade your knowledge meter
How much do you know about the technology of media now? Flip back to p.3 and shade the meter again in pen (or shade this one). Did any of the four bricks on p.2 move from "Not yet" to "Got it"?
Be the fact-checker
Professional fact-checkers don't trust their eyes; they check sources. Verify three real posts from your feed, then make a fact-check card for each.
The job
- Find 3 posts that make a claim (a stat, a photo, a quote)
- Run SIFT on each one
- At least one reverse image search
- Fill in a fact-check card for each
Then share one
Turn your best card into a 30-second video or a slide that teaches someone else how you checked it. Be fair: say what you couldn't confirm.
Stay safe: don't comment on, argue with, or report people while you investigate. Just observe and record. Never share personal information about the people who posted.
5 Questions
I Still Have
This module didn't answer everything. Good. Write 5 new questions it didn't answer, one of each type below. There are no wrong questions, only ones nobody's asked yet.
★ Star the one question you'd most like answered. You'll need it on the next page.
1What I'll find out
Break your big question into 2 or 3 smaller ones you could actually answer.
2Where I'll look
3How I'll share it
Who I'll share it with:
Done by: