AI curriculum

Course outline for Artificial Intelligence Foundations

Explore how AI solves problems through explicit rules, statistical learning, neural networks, and generation. Learn the difference between supervised and unsupervised learning and between regression, classification, and clustering. Compare linear models, decision trees, and classic text-processing techniques before placing deep learning and generative AI in the broader picture. Use small experiments to distinguish deterministic execution, probabilistic prediction, and sampled generation, then choose and check an appropriate approach.

Artificial Intelligence Foundations Course Objectives

  • Distinguish symbolic AI, statistical learning, neural networks, and generative tasks.
  • Identify supervised and unsupervised problems and suitable learning tasks.
  • Explain regression, classification, clustering, and representative non-neural methods.
  • Explain classic NLP representations and a simple text-classification pipeline.
  • Distinguish training, validation, testing, and inference through examples.
  • Choose rules, probabilistic models, or people and match checks to the task.

Pre-requisites

  • Basic computer literacy and comfort with tables, graphs, and simple arithmetic.
  • No programming, calculus, prior AI vocabulary, or model account is required.

Before this course

Complete the required courses or bring equivalent experience. Recommended courses add useful depth; they are not enrolment gates.

No prerequisite course is required. The entry skills above still apply.

Readiness checks

  • Basic computer literacy and comfort with tables, graphs, and simple arithmetic.
  • No programming, calculus, prior AI vocabulary, or model account is required.

Basic computer literacy and comfort with tables, graphs, and simple arithmetic; no programming, calculus, prior AI vocabulary, or learner model account is required.

Find your place in the learning map

Lab Setup

Delivery requirements: the following environment, materials, access, and exercise timing must be prepared and verified before delivery. This outline does not certify that a lab package is ready.

A browser-based teaching interface or prepared notebook with small tabular and text datasets, non-neural models, and visible evaluation results. Controls must permit actual input or model-setting changes without requiring learners to write code. The instructor must verify this environment before delivery. No paid model account or specialized hardware is required. A worksheet alone cannot substitute for observing model fitting and evaluation. The lab must include a small actual generation comparison; no paid learner account is required, but the lab provider must arrange verified model access.

Detailed Course Outline

Module 1: From explicit rules to learning from data (2 hours)

  • Symbolic AI and rules: AI, symbolic reasoning, search, and explicit inference rules.
  • Deterministic execution: Deterministic execution: the same complete inputs and fixed rules give the same result.
  • Learning, prediction, and sampling: Statistical learning versus written rules; uncertain predictions versus sampled outputs.
  • Learning inputs and parameters: Identify features, targets, labels, examples, and learned parameters in a dataset.
  • Learning setups: Supervised and unsupervised learning; a brief reinforcement-learning orientation.
  • Training and inference stages: Distinguish training, validation, testing, and inference through small learning examples.
  • Data quality and generalization: Explain data quality, leakage, overfitting, and generalization through observed results.

Practical outcome: Distinguish fixed calculations from uncertain predictions and identify valid learning inputs.

Module 2: Regression and classification without neural networks (2 hours)

  • Prediction tasks: Regression for numerical prediction and classification for category prediction.
  • Linear regression: Use linear regression to examine fitted relationships and residual prediction errors.
  • Logistic regression: Use logistic regression for classification rather than numerical regression, despite its name.
  • Decision trees: Decision trees, split rules, and controlling tree complexity.
  • Fixed-model inference: Fixed-model inference can be deterministic despite uncertain predictive accuracy.
  • Baselines and held-out evaluation: Compare simple baselines with models using held-out examples rather than training results.
  • Prediction errors and trade-offs: Numerical error, confusion matrices, precision, recall, and error trade-offs.

Practical outcome: Compare simple models and explain their errors on unseen cases.

Module 3: Clustering and classic natural-language processing (2 hours)

  • Unsupervised clustering: Explore clustering without target labels and interpret the resulting exploratory groups.
  • K-means and distance: Use K-means to examine feature scaling, distance, and cluster interpretation.
  • Classic text preparation: Prepare text with NLP tokenization, normalization, and word-based input features.
  • Bag-of-words and TF-IDF: Compare bag-of-words and TF-IDF representations as inputs to classic text models.
  • Non-neural text classification: A classic text classifier using Naive Bayes or a linear model.
  • Text representation limits: Lexical matching limits, context, negation, and domain changes.

Practical outcome: Interpret a clustering result and test a word-feature text classifier.

Module 4: Neural networks, generation, and choosing an approach (2 hours)

  • Neural representations: Explain neural-network weights, layers, and learned representations through small examples.
  • Deep learning within machine learning: Deep learning within machine learning; relationships among AI approaches.
  • Generative tasks and context: Relate generative tasks, foundation models, and context supplied to a model.
  • Tacit knowledge: Tacit knowledge and the limits of fully specified rules.
  • Application decisions: Observation, interpretation, evaluation, and action in an application.
  • Choosing rules or models: Choose rules for exact requirements; models for uncertain patterns or useful generation.
  • Valid variation and constraints: Recognize several valid answers while requiring facts, constraints, and safe actions.
  • Human judgment: Keep human judgment where evidence, checks, or acceptable risk are insufficient.

Practical outcome: Explain the relationships and choose a justified approach for a new task.

Practical exercise

  • Use a small service-desk dataset to frame numeric prediction, ticket classification, and exploratory clustering as different tasks. Compare a simple baseline with a non-neural model on held-out examples. Change one supported setting or input feature and inspect the result. Test a bag-of-words or TF-IDF text classifier on a new message. Finish with a short method-selection table for calculation, prediction, summary, rule-based recommendation, and approval. Run a fixed calculation twice and compare several sampled summaries of the same source. Explain why wording can vary while facts and required content must still be correct. Contrast exact-result checks, prediction errors on unseen examples, and a simple generation-quality rubric.

What you will produce

Compare rules, classical statistical models, neural methods, and generation, then select an approach and appropriate evidence checks.

How the work is checked

  • The learner distinguishes a task, a learning setup, and a model family.
  • Regression and classification use labels; clustering does not assume target labels.
  • Linear regression and logistic regression are assigned to appropriate tasks.
  • Evaluation uses unseen examples and explains a relevant error measure.
  • The NLP explanation connects word features with a classifier and its limitations.
  • Neural, deep-learning, and generative categories are not treated as disjoint branches.
  • A method choice considers data, uncertainty, consequences, and a simpler alternative.
  • The learner does not equate AI with randomness or deterministic execution with correctness.
  • The learner accepts valid wording differences but detects incorrect facts or broken constraints.

Eight guided hours include demonstrations, actual learning experiments, practice, feedback, and questions. Breaks and pre-course setup are outside the total. This is a broad conceptual and experimental foundation, not mathematical derivation or independent ML development. Reinforcement learning receives orientation only. Classic NLP includes word representations and a classifier, not a complete NLP course. Fixed-model inference can be deterministic while predictive accuracy remains uncertain; sampled generation can produce several valid answers. Changing inputs or external state alone does not establish stochastic behavior. LLM mechanisms receive deeper treatment in How Large Language Models Work. Demonstrations, model access, and pacing require verification before delivery.

When another approach is better

  • Use exact rules when the requirement is fully specified and exact results matter.
  • Keep human judgment where data, evidence, checks, or acceptable risk are insufficient.
  • Use specialist ML or NLP development training for mathematical or independent implementation depth.

How we train

Contact us for full course details, including duration, delivery options and lab requirements.

Ask about exercises, instructor feedback and the prior knowledge you need. Tool-specific courses marked provisional may become modules in a broader course.

Discuss your learning goals