Episode 2 – How to explain Machine Learning, NLP, Generative AI, and Predictive Analytics
"They're All Just AI, Right?"
How to explain Machine Learning, NLP, Generative AI, and Predictive Analytics to a team that nods politely and understands nothing.
Marcus Has a Problem
He's got 90 minutes, a room full of evaluators, and a slide deck that says "AI capabilities" at the top. Help him figure out how Priya would actually run this session.
Marcus finds Priya in the kitchen, staring into a mug of tea with the vacant expression of someone who has just read the words "neural network" for the fourteenth time this week. He pulls up a chair.
"Priya, I need help. I'm doing a training tomorrow on ML, NLP, Generative AI, and Predictive Analytics. Every time I explain them, everyone nods. Then someone asks a question and I realise — they all think it's the same thing."
Priya puts down her mug. She has seen this exact look before. It is the look of a person who made the mistake of using the word "algorithm" in a sentence and is now living with the consequences.
Marcus has his laptop open. His coffee is untouched. He is ready. Choose wisely.
The Comparison Table Framework — Unpacked
Marcus's typing stops. He stares at the table on his screen. "So the ethical column isn't a disclaimer — it's the fifth capability?"
Exactly. And now his team won't just know what each tool does. They'll know what each tool costs — in oversight, privacy, and responsibility.
| Capability | What it does | M&E example | Simple analogy | Ethical consideration |
|---|---|---|---|---|
| Machine Learning | Finds patterns in data and improves with experience | Classifying survey responses into outcome categories automatically | A junior analyst who gets better the more files you give them | Requires bias checks — it learns your data's prejudices too |
| NLP | Reads, understands, and processes human language | Coding 500 interview transcripts for themes overnight | A very fast reader who highlights everything but understands nothing | Needs privacy safeguards — it reads every word, including the sensitive ones |
| Generative AI | Creates new text, reports, visuals, or content on request | Drafting a first version of an evaluation report from your findings | A research assistant who writes confidently even when they're guessing | Demands fact-checking — it will hallucinate a citation without blinking |
| Predictive Analytics | Uses past data to forecast future outcomes or flag risks | Identifying which programme participants are most likely to drop out | A very well-read weather forecaster — useful, but not always right | Must be used for early support, never for punitive decisions |
How Priya built Marcus's training in one evening
via ClaudePriya started by making Claude meet the audience, not the subject matter. Evaluators respond to examples, not encyclopaedia entries.
"Create simple, non-technical definitions for Machine Learning, NLP, Generative AI, and Predictive Analytics. Use a concrete evaluation example for each. Avoid the words 'algorithm', 'model', and 'training data'."The table does what a slide deck cannot: it forces side-by-side comparison. Evaluators immediately spot the differences when they see them laid out in columns rather than explained in paragraphs.
"Now create a comparison table with five columns: Capability, What it does, M&E example, Simple analogy, and Ethical consideration for each capability."Abstract capabilities become concrete the moment you show what Monday morning looked like before versus after. Marcus used these as his opening slides — and the room was quiet in the good way.
"For each capability, show a 'before AI' (manual process) and 'after AI' example from real evaluation work. Keep it realistic — not magic, just faster."Marcus created four scenario cards drawn from his team's actual projects. Participants had to identify which capability applied — and defend their answer. Two people argued for eight minutes about whether thematic coding was NLP or ML. (It's NLP. They were both slightly right.)
Priya insisted on this. Not as a disclaimer tucked at the end, but as the fifth column in the table — equal weight to the others.
"For each AI capability, add one specific ethical consideration an evaluator must address before using it. Be concrete — not 'be careful with data' but the actual risk and safeguard."- ML requires bias audits — it learns from flawed data and amplifies the flaws
- NLP needs informed consent — participants must know their words are being processed
- Generative AI demands verification — every fact it produces is a hypothesis until checked
- Predictive Analytics must be used for support, never for exclusion or punishment
Tables Beat Slides
Side-by-side comparison forces the brain to notice differences. Sequential slides let people nod through each one without ever comparing them.
Use Cases Over Definitions
Evaluators don't need to know how NLP works. They need to know that NLP is what codes their transcripts at 2AM so they don't have to.
Scenarios Create Memory
Arguing about whether something is ML or NLP is more memorable than reading about either. The wrong answers teach as much as the right ones.
Ethics as Column Five
Making ethics structural — not a footnote — means every capability discussion automatically includes its responsibilities. It becomes reflex, not reminder.
Your Turn — Build Your Own Capability Table
Works in Claude, ChatGPT, Gemini — paste it in, edit the context, and get your table in under two minutes
Create a comparison table explaining the four main AI capabilities to an M&E team. The audience are experienced evaluators — comfortable with qualitative and quantitative methods, but new to AI. The table must have five columns: 1. Capability (ML, NLP, Generative AI, Predictive Analytics) 2. What it does — in plain language, no jargon 3. M&E example — a specific, realistic evaluation task this enables 4. Simple analogy — something they'd recognise from their existing work 5. Ethical consideration — one concrete risk and how to address it After the table, create: - A "before/after" example for each capability showing manual vs. AI-assisted process - Four scenario cards for a group activity: describe a real evaluation situation and ask the team to identify which capability applies - A summary of the two most commonly confused pairs (ML vs Predictive Analytics, NLP vs Generative AI) and a one-sentence rule for telling them apart
Pro tip: After getting the table, ask: "Now add a row for Computer Vision" — watch your team's faces when they realise there's a fifth capability they haven't accounted for yet.
Ready to Go Deeper?
Both courses cover AI capabilities in the context of real evaluation practice — not tech theory, not marketing copy.
AI for International Development & Humanitarian Practitioners
Specialised for development and humanitarian contexts
Explore Course →Already enrolled? Continue your learning journey →
Marcus needs a strategy. But Priya has three options in mind — only one of them will actually work when the room is full of evaluators who last thought about statistics in 2009.