CS50's Introduction to AI with Python

Learning plan

About 80 hours of focused study, split into 5 phases. Pick your weekly pace and the timeline adjusts.

Estimated finish: 8 weeks

What the exam covers

Source: official exam guide. Weights can change, so check before you start.

Before you start

• Comfortable Python: functions, classes, recursion, dictionaries and sets. CS50P or CS50x week 6 level is enough
• A GitHub account to submit projects with submit50, and a local Python 3 install or the CS50 VS Code codespace

1. Weeks 1–2 · 18 h

   Search and knowledge (weeks 0–1)

   • Implement BFS, DFS, greedy best-first and A* search, and explain why A* with an admissible heuristic is optimal
   • Implement minimax with alpha-beta pruning for an adversarial game
   • Encode propositional logic and use model checking to infer facts

   Hands-on: Complete Degrees (shortest path between actors using BFS) and Tic-Tac-Toe (an unbeatable minimax player), then Knights and Minesweeper. Each must pass check50 before you submit.

   Use: CS50 AI Lecture 0 notes (Search) · Project 0a – Degrees · Project 0b – Tic-Tac-Toe · Project 1a – Knights · Project 1b – Minesweeper

2. Weeks 2–4 · 16 h

   Uncertainty and optimisation (weeks 2–3)

   • Compute probabilities with Bayes' rule, joint distributions and Bayesian networks, and estimate PageRank both by sampling and by iteration
   • Model a constraint-satisfaction problem with node and arc consistency (AC-3) and solve it with backtracking search
   • Compare hill climbing, simulated annealing and linear programming

   Hands-on: Complete PageRank, Heredity (inferring gene distributions in a family) and Crossword (a CSP with arc consistency and backtracking).

   Use: Project 2a – PageRank · Project 2b – Heredity · Project 3 – Crossword

3. Weeks 4–6 · 24 h

   Learning and neural networks (weeks 4–5)

   • Train and evaluate a k-nearest-neighbour classifier with scikit-learn and report sensitivity and specificity
   • Implement Q-learning for a game with an exploration/exploitation trade-off
   • Build and tune a convolutional network in TensorFlow/Keras, experimenting with layers, dropout and filter sizes

   Hands-on: Complete Shopping, Nim and Traffic. For Traffic, keep a short log of each architecture you tried and its validation accuracy on the German traffic-sign dataset, because the project asks you to describe your experimentation.

   Use: Project 4a – Shopping · Project 4b – Nim · Project 5 – Traffic · TensorFlow tutorials

4. Weeks 6–7 · 12 h

   Language (week 6)

   • Write a context-free grammar and use NLTK to parse sentences and extract noun phrases
   • Explain attention and transformers, and visualise BERT attention heads to interpret what they learn

   Use: Project 6a – Parser · Project 6b – Attention

5. Week 8 · 10 h

   Resubmissions and certificate check

   • Check your gradebook and resubmit any project scoring under 70%
   • Re-implement one search and one learning project from scratch without notes to lock in the concepts

   Use: CS50 Certificate requirements · CS50 AI academic honesty policy

You're ready to book when…

• All 12 projects are submitted and each shows a score of at least 70% in your gradebook
• You can explain minimax, A*, AC-3, Q-learning and backpropagation to someone else without notes
• Your Traffic CNN reaches high validation accuracy and you can justify the architecture you chose

Exam tips

• The free CS50 Certificate needs a score of at least 70% on every project. The edX verified certificate is a separate, paid option.
• The academic honesty policy bans copying solutions or using AI tools other than CS50's own duck. Violations can cost you the certificate.