AI · Grade 10 · Chapter 4
AI Applications You Know
AI is already part of daily life — far more than most people realise. This chapter breaks it down category by category.
Learning objectives
What We'll Cover
- Map where AI appears in everyday technology
- Understand how recommendation engines, computer vision, and NLP work at a conceptual level
- Analyse AI in industry: healthcare, transport, finance, creative tools
- Evaluate the trade-offs and risks of widespread AI adoption
- Connect AI applications to the underlying techniques from Ch. 3
A day in the life
Your AI-Saturated Morning
7:00 am — smart alarm adapts to your sleep cycle. 7:10 — face unlock. 7:15 — news feed curated by algorithms. 7:30 — navigation app avoids traffic. 8:00 — spam filter cleans email. All before school starts.
Class count: How many AI touchpoints can you identify in your own morning? Write them down individually, then share. Who has the most?
Category 1
Recommendation Systems
These are the most commercially important AI systems in existence. They decide what you see next — on YouTube, Netflix, TikTok, Instagram, Spotify, Amazon.
- Collaborative filtering: "People like you also liked..."
- Content-based filtering: "This is similar to what you watched before"
- Hybrid models: Most modern systems combine both
How recommendations work
The Feedback Loop
1You watch a video
2System records behaviour (watch time, likes, shares)
3Model updates your "interest profile"
4Next recommendations are recalibrated
Key insight: The goal of the recommendation system is not your happiness — it's maximising engagement time. This can lead to filter bubbles and extreme content rabbit holes.
Category 2
Computer Vision
AI systems that can "see" and interpret images and video.
Face Recognition
Phone unlock, airport e-gates, photo tagging on social media
Object Detection
Self-driving cars identifying pedestrians, traffic signs, other vehicles
Medical Imaging
Detecting tumours in X-rays and MRI scans with near-specialist accuracy
Category 3
Natural Language Processing (NLP)
AI that understands, generates, and translates human language.
Voice Assistants
Siri, Alexa, Google — understand spoken requests and respond naturally
Spam Filters
Email providers classify millions of emails per second as spam or legitimate
Machine Translation
Google Translate, DeepL — modern neural translation outperforms rule-based systems
Chatbots & LLMs
Customer service bots, ChatGPT, Gemini — conversational interfaces at scale
Category 4
AI in Serious Industries
Healthcare
Drug discovery, diagnostic imaging, personalised medicine, epidemic modelling
Finance
Fraud detection (real-time), credit scoring, algorithmic stock trading
Transport
Navigation (Waze/Google Maps), autonomous vehicles (Tesla FSD, Waymo), logistics optimisation
Category 5
AI Creative Tools
The newest wave: AI that generates text, images, audio, video, and code.
- Text: ChatGPT, Gemini, Claude — writing, summarising, coding
- Image: Midjourney, DALL-E, Stable Diffusion — art generation from prompts
- Music: Suno, Udio — AI-generated songs in any genre
- Video: Sora (OpenAI), Runway — generating video from text descriptions
- Code: GitHub Copilot — autocompleting and generating code
Analysis
The Trade-Offs
Every AI application comes with benefits AND concerns. Let's think critically.
Recommendation Systems
✅ Discover great content
⚠️ Filter bubbles, addiction design, radicalization risks
Face Recognition
✅ Convenient security
⚠️ Privacy invasion, misidentification bias (especially for darker skin tones)
Medical AI
✅ Earlier, cheaper diagnosis
⚠️ Biased training data, liability questions
AI Creative Tools
✅ Democratise creativity
⚠️ Copyright, misinformation (deepfakes)
Scenario discussion
Case Study: YouTube's Recommendation Engine
Research has shown that YouTube's recommendation algorithm, optimising purely for watch time, was repeatedly found to guide users from mainstream content toward increasingly extreme content — because extreme content provokes stronger emotional reactions and keeps people watching longer.
Discuss: Should AI companies be responsible for what their algorithms recommend? Should there be government regulations? What would you change?
Quick check · 1
Which best describes how a recommendation system works?
AIt randomly shows you content
BIt learns from your past behaviour to predict what you'll engage with next
CA human editor chooses content for each user
DIt shows the most recently uploaded content first
Click to reveal answer
Quick check · 2
Which AI application would best help detect a tumour in an X-ray?
ANatural Language Processing
BComputer Vision
CRecommendation System
DSpam Filter
Click to reveal answer
Class activity
AI Application Audit
As a class, we'll build a visual map on the board.
Activity: Students call out apps/services they use daily. For each one, identify: (1) What AI task is it doing? (2) What data does it need? (3) What's a potential risk? Build a table together.
Algorithmic bias
When AI Gets It Wrong — Systematically
AI is only as fair as its training data. Examples of real-world algorithmic bias:
- Face recognition systems with 99% accuracy on white male faces — but only 65% on dark-skinned female faces (MIT Media Lab, 2018)
- Hiring algorithm penalising CVs that included the word "women's" (Amazon, abandoned 2018)
- Risk assessment tools in US criminal justice scoring Black defendants as higher risk at higher rates
Lesson: Biased data → biased model. Garbage in, garbage out.
Quick check · 3
Why might a facial recognition system perform worse on darker skin tones?
AThe camera hardware is different
BThe training dataset contained more images of lighter-skinned people, so the model learned less from darker-skinned examples
CDarker skin blocks facial features
DThe algorithm was deliberately designed this way
Click to reveal answer
AI and the future of work
What Jobs Does AI Affect?
AI automates tasks — not just jobs. Any task involving pattern recognition, repetition, or data processing is a candidate for automation.
At risk (routine tasks)
Data entry, basic customer service, routine legal/accounting work, basic journalism, radiology
More resilient (human judgment)
Creative direction, empathetic counselling, physical trades (plumbing/electrical), complex negotiation, research leadership
Critical thinking
Is AI a Tool or a Replacement?
Two views exist:
Augmentation View
"AI makes humans more capable. A doctor with AI can diagnose 10× more patients. AI assists, humans decide."
Displacement View
"AI replaces entire categories of work. The people displaced may not be able to retrain fast enough."
Discuss: Which view do you think is more accurate? What evidence would change your mind?
Quick check · 4
True or False: AI always makes better decisions than humans.
ATrue — AI processes more data than any human can
BFalse — AI can be biased, wrong, and lacks the contextual understanding humans have
CTrue — AI is always objective because it uses maths
DFalse — AI never makes correct decisions
Click to reveal answer
Application activity
Design an AI Application
In pairs, design a hypothetical AI application for your school or local community.
- What problem does it solve?
- What AI technique does it use? (recommendation, CV, NLP, prediction…)
- What data does it need to train on?
- What are two potential risks?
- How would you address those risks?
Quick check · 5
Which of these is an example of Natural Language Processing?
AA self-driving car recognising a stop sign
BNetflix recommending a show
CA voice assistant understanding your spoken question and replying
DA spam filter blocking email based on sender IP
Click to reveal answer
Exit reflection
Your AI Literacy Score
Rate yourself honestly (thumbs up / sideways / down) on each:
- I can explain how a recommendation engine works
- I can name three real-world AI applications and their AI technique
- I can identify a trade-off or risk in an AI application
- I understand what algorithmic bias is and how it arises
Discuss: Which item needs the most work? What would help you understand it better?
Before you go
Today We Learned...
- AI is woven into daily technology: recommendations, vision, NLP, and creative tools
- Recommendation systems optimise for engagement, not user wellbeing
- Computer vision enables face recognition, autonomous vehicles, and medical imaging
- AI bias comes from biased training data — it is a real, documented problem
- AI augments and disrupts work — the picture is nuanced, not simple
Next chapter: We start writing Python — our first steps toward understanding how these systems are actually built.