💡 Inspired by: Sal Khan (Khan Academy) — champion of making complex knowledge accessible to everyone
What Is AI? Demystifying the Buzzword
We begin at the beginning. What actually is Artificial Intelligence? How is it different from Machine Learning and Generative AI?
This lecture cuts through the hype with everyday analogies and exposes both the power and the limitations of AI — showing you how it's
already quietly woven into your daily life.
💭
1. Concept
🛠️
2. Hands-On
🗺️
3. Takeaway
Concept Discussion — 20 Min
What actually is AI?
Artificial Intelligence is not a conscious mind. It is applied mathematics. It is the science of making machines do things that would require intelligence if done by humans.
1950
Alan Turing asked: "Can machines think?"
2017
Google invents the "Transformer" architecture
"AI is the new electricity. Just as electricity transformed almost everything 100 years ago, today I actually have a hard time thinking of an industry that I don't think AI will transform in the next several years."
— Andrew Ng, Stanford University
The AI Matryoshka Doll
It's important to understand the hierarchy of AI terminology. They are not interchangeable buzzwords.
Artificial Intelligence (AI): The broad concept of machines performing tasks normally requiring human intelligence.
Machine Learning (ML): A subset of AI. Instead of giving a computer rules, you give it data, and it learns the rules. (e.g., YouTube algorithm).
Deep Learning (DL): ML using multi-layered artificial neural networks inspired by the human brain.
Generative AI: The newest breakthrough. Deep Learning models that create original content (text, audio, images) rather than just categorizing existing data.
How LLMs Learn: Pre-Training vs. Fine-Tuning
Why does ChatGPT act like a helpful assistant rather than just blurting out random internet text? It goes through two distinct phases of learning:
1. Pre-Training (The Generalist): The model reads terabytes of data (Wikipedia, Reddit, books) to learn grammar, facts, and reasoning. It simply learns to predict the next word.
2. Fine-Tuning (RLHF): Reinforcement Learning from Human Feedback. Human workers rate the AI's responses, teaching it to be polite, helpful, and safe. This turns a raw text-predictor into a conversational agent.
The Epistemological Illusion
Large Language Models (LLMs) operate on probabilistic token prediction rather than semantic comprehension. They map the statistical relationships between words within a massive multi-dimensional dataset.
The "Hallucination" Phenomenon
Because they optimize for syntactic plausibility rather than objective truth verification, models inherently generate falsities. In academic contexts, this means the generated text is a reflection of the consensus shape of its training data, not a factual database.
Heuristic: AI is a structural reasoning engine, not an epistemological search engine.
Local Context: AI in Bangladesh
The Government's Aspire to Innovate (a2i) program has already deployed AI-driven solutions to public services.
Muktopaath & AI: Personalized learning pathways for millions of teachers and youth.
National Helpline (333): AI-assisted triage for millions of citizen requests regarding health, disaster response, and legal aid.
In developing nations, AI is not just a productivity tool; it is a critical lever for bypassing traditional infrastructure bottlenecks in health and education.
🎥 More on AI Foundations
Hands-On Activity — 20 Min
📝 Activity 1: The Domain Stress Test
Open ChatGPT or Gemini on your phone. Push the AI into highly specific, localized, or nuanced knowledge testing from your faculty:
Law: "Explain the contradictions in the Digital Security Act of Bangladesh regarding journalistic freedom."
Arts: "Write a critical analysis of Kazi Nazrul Islam's 'Bidrohi' focusing on its anti-colonial themes."
Business: "Calculate the break-even point for a rickshaw garage operating in Dhaka, assuming standard local costs."
🔎 Activity 2: The Hallucination Hunt
Ask the AI to provide biographical details about a niche, lesser-known professor at the University of Dhaka, or request citations for specific academic papers in your field. Fact-check every claim. Report the fabricated details back to your group.
🗺️
Your Takeaway: Personal AI Trust Map
Create a 2x2 matrix on a piece of paper. The axes: Low Consequence vs High Consequence and High Accuracy Needed vs Low Accuracy Needed. Plot 10 tasks you do weekly (e.g., drafting emails, researching medical symptoms, deciding what movie to watch). Where does AI belong safely? This is your first portfolio deliverable.
