Artificial Intelligence
What AI is, what it is not, and how to tell the difference.
Read Week One Free Download the 244-page workbookWhy Artificial Intelligence?
Everybody is being sold AI. Very few people can say whether a given problem needs it.
This course is an introduction, which means it starts before the models: with what artificial intelligence actually covers, how rules, search and learning differ, and how to measure any of them so the answer means something. It then works through language, vision, fairness, explainability and governance, and finishes with a capstone whose first question is whether the problem needs AI at all.
What Makes This Course Different?
The comparisons in this course were run rather than asserted, and several of them favour the unglamorous option. Those results are printed on the lesson page.
What This Programme Delivers
Skills You Will Build
Technical Skills
- Turning text into numbers, and why that step decides most of the outcome
- Entity extraction, keywords, topics, similarity and sentiment
- Breadth-first, depth-first and A* search, with expansion counts compared
- Embeddings, from counting to compression
- Transformers and large language models, attention worked through by hand
- Images as arrays, and the classical features built from them
- Six image classifiers compared on the same images
- Detection, segmentation and transfer learning
- Fairness metrics on data whose truth is known
- Importance, counterfactuals and partial dependence
- Re-identification, model leakage and model cards
Judgement
- Deciding whether a problem needs AI or a rule
- Reading a confusion matrix instead of an accuracy score
- Spotting the four ways a good result turns out not to be one
- Knowing which definitions of fairness cannot hold at the same time
- Recognising where an explanation is misleading you
- Placing the human in the finished system
Does This Problem Actually Need AI?
Week 3 takes all three approaches apart on the same problem. The right answer is often the cheapest one.
| Consideration | Rules | Search | Learning |
|---|---|---|---|
| Use when | The logic is known and stable | The goal is known, the path is not | The pattern exists but nobody can write it down |
| Data needed | None | A defined state space | Labelled examples, usually many |
| Cost to build | Hours | Days | Weeks, plus data collection |
| Explaining a decision | Trivial | Show the path | Needs its own techniques |
| Fails when | The world changes | The space is too large | The data does not match reality |
| Course coverage | Week 1 | Week 3 | Weeks 4 to 10 |
14-Week Learning Roadmap
View Detailed Week-by-Week Curriculum →
Foundations
Weeks 1 to 3. What AI actually is, rules against search against learning, measuring a model honestly, and the four ways a good result turns out not to be one.
Language
Weeks 4 to 7. Text as data, the standard language tasks with the measurement each one needs, embeddings, then transformers and large language models.
Vision
Weeks 8 to 10. What a picture looks like to a computer, six classifiers compared honestly, then detection, segmentation and transfer learning.
Responsibility
Weeks 11 to 13. Bias and fairness measured on known truth, explaining what a model did and where that misleads, then privacy, safety and governance.
Capstone
Week 14. One problem worked end to end, from whether it needs AI at all to where the human belongs in the finished system.
What You Build Along The Way
A Classifier Against A Keyword Rule
A trained text classifier measured against a handful of keyword rules, on a corpus split by phrasing so the test set cannot leak the answer.
scikit-learnA Maze, Three Ways
Breadth-first, depth-first and A* run on the same maze with steps and nodes expanded counted: 22 steps and 108 expansions against 48 and 64.
SearchAn Embedding Space You Can Query
Words as arbitrary symbols turned into words with relationships, built up from counting and compression rather than downloaded.
EmbeddingsA Fairness And Explainability Audit
A model examined on data whose ground truth is known, including the fairness definitions that cannot all be satisfied at once.
GovernanceWhat The Capstone Asks Of You
Framing
Does this need AI at all?
The first question of the capstone, answered with a measured rule-based alternative rather than an opinion.
Who is affected
Name the people on the receiving end of the decision before you build anything that makes it.
Measurement
The metric that fits the problem
Chosen and justified before the model, with a confusion matrix rather than a single accuracy figure.
Where it fails
Per-group performance, and the failure mode the headline number is hiding.
Responsibility
A model card
What the system does, what it was trained on, where it should not be used.
Where the human belongs
The point in the workflow where a person reviews, overrides or is accountable.
Finish With
Where This Material Leads
The roles this course is written for:
Also written for managers and product people who have to judge an AI claim without building the thing themselves.
Tools & Technologies
Core Tools
Also Covered
Every tool used is free and open source. Nothing on this course needs a paid licence.
Who This Is For
Complete Beginners
No maths background, no prior AI
Managers And Product People
Enough depth to judge an AI claim
Choosing A Specialism
Take this before Machine Learning or Deep Learning
No maths background required and no prior AI. Every code example is explained line by line.
Certificate of Completion
Certificate of Completion in Artificial Intelligence
Issued by EDUSHARK TRAINING once you finish the fourteen weeks and submit your capstone. It records what you built, not a grade.
Start At The Beginning
Fourteen weeks that start with what AI is, and end with whether your problem needs it.