Machine Learning 101: A Youth Guide to Teaching Computers to Think

When Netflix recommends your next favorite show or when your email filters out spam, that’s machine learning at work. Machine learning (ML) is a branch of artificial intelligence that allows computers to learn from data and make decisions without being explicitly programmed. Instead of humans writing every rule, the computer figures out patterns itself and improves as it sees more examples.

For many people, ML feels abstract or intimidating. That’s why, during the curaJOY Fellowship, the Tech Cohort hosted a workshop to explain ML in a simple, youth-friendly way so that young people can see themselves not just as tech consumers, but as future builders of ethical, impactful technology.

Machine Learning Is Everywhere

Machine learning is everywhere today, helping us make sense of the massive amounts of data that humans can’t process on their own. It uncovers patterns we might never notice, and it gets smarter as it learns from more data. From navigation apps and Siri to self-driving cars, medical diagnoses, and even YouTube recommendations, machine learning powers many of the tools we rely on every day.

Types of Machine Learning

In our workshop, we explained five major types of ML:

1. Supervised Learning: Learns from labeled data (e.g., spam vs. not spam emails).
2. Unsupervised Learning: Groups or clusters unlabeled data (e.g., discovering patterns in shopping habits).
3. Semi-Supervised Learning: Combines a small labeled set with lots of unlabeled data (useful when labeling is expensive).
4. Reinforcement Learning: Learns by trial and error, guided by rewards and penalties (e.g., training an AI to play chess or drive a car).
5. Self-Supervised Learning: Creates its own labels from raw data (e.g., predicting missing words in a sentence — how GPT models are trained).

Classification Errors: Why They Matter

We used real-life examples to explain false positives and false negatives:

• In spam detection:
  • A false positive wrongly flags an important email as spam.
  • A false negative misses a spam email, which is annoying but not catastrophic.
• In medical diagnoses:
  • A false positive causes stress and extra tests.
  • A false negative means missing a cancer diagnosis — which can be deadly.

We had a discussion about which classification error was worse for cyberbullying detection:

• A false positive wrongly flags a harmless message, eroding trust and causing censorship concerns.
• A false negative misses real bullying, leaving harm unaddressed. The lack of intervention allows the situation to escalate.

We decided that, while false positives are still harmful, false negatives have a much worse impact. Thus, we should focus on minimizing false negatives. In other words, we have to optimize the recall of our model. Recall measures how many actual positives are correctly identified: if recall is high, then false negatives are low.

The Ethical Questions

Teaching machines to make decisions isn’t just technical — it’s ethical. We posed questions like:

• Who is responsible when AI makes a mistake?
• Should engineers be accountable if an AI chatbot generates offensive responses?
• If an algorithm misdiagnoses a patient, where does the liability lie?
• What are the environmental costs of training huge ML models?

These aren’t easy questions to answer. During the workshop, fellows and guests explored different perspectives, weighing both the benefits and risks. Our goal was to emphasize that every technological choice has human consequences — and that youth voices are essential in shaping how these technologies evolve.

How We’re Using ML in Our Fellowship

At curaJOY, we’re applying ML to one of the most urgent challenges facing young people today: cyberbullying.

So far, we have:

• Begun building a youth-focused dataset of cyberbullying examples, using scraping, AI-driven data augmentation, and exploring crowdsourcing.
• Created a labeling system to classify cyberbullying into categories like:
  • Verbal Violence (insults, threats, harassment, abuse)
  • Visual Violence (harmful images/memes, deepfakes, screenshot shaming)
  • Exclusion (leaving people out of groups, shadowbanning, gossip groups)
  • Other (impersonation, catfishing, hybrid behaviors)
• Integrated an annotation interface so fellows could tag both text and images consistently, building high-quality training data.

Our ultimate goals are to:

1. Train ML models that detect cyberbullying in real time.
2. Build an AI coach that doesn’t just flag harmful messages, but suggests supportive responses and resources — empowering peers to intervene, since research shows that 57% of bullying stops within 10 seconds when someone steps in.

Why This Matters

Most existing online safety tools miss the mark because they don’t reflect how younger generations actually communicate. Memes, slang, emojis, and new platforms evolve faster than adults or legacy filters can keep up. By building a dataset curated by Gen Z and Gen Alpha, we’re creating tools that finally understand our world.

For me, curaJOY isn’t just about applying technical skills. It’s about proving that youth-led teams can combine research, design, and technology to build tools that save lives. Because at the end of the day, it’s our generation — youth with drive, passion, and the ability to work smarter than ever before — who have the power to change the world.