Theme:
- Foundations
- Understanding (gen)AI and Its Limits,
- Appopriate Use
Prereading
Course Content
Expectation setting: what do we want to collaboratively steward in this in-person time together, what exists here, in the time that we have here, and what do we (as a group) agree belongs in offline work between sessions.
How AI systems actually work (demystifying the technology)
Data and Statistics
- Data is structured observation
- Every time you click, buy, type, or move, you generate data. Data and data sets are just observations organized in a manner so that they can be operated upon.
- Statistics finds signal in noise
- Extracting pattern versus identifying noise. Statistics are mathematical operations that summarise the characteristics of large groups of observations.
- Probability is the language of uncertainty
- We can express descriptions and predictions as probabilities against further obesrvations 'There is a 70% chance of rain' is a statistical statement; so is 'this email is probably spam'
- Models are simplified representations
- A statistical model captures the key relationship in data. The simplest example: a straight line through a scatter plot (linear regression).
Machine Learning
- Traditional programming vs. ML
- In traditional software, a human writes every rule: IF purchase > $100 THEN flag as suspicious. ML extracts a statistical representation, of a specified form, from (usually thousands of) examples, effetively writing the rules.
- Supervised learning: learning from labels
- You give the model labelled examples (email = spam / not spam) and it learns to classify new examples. This is the workhorse of modern AI: fraud detection, medical diagnosis, credit scoring.
- Unsupervised learning: finding hidden structure
- No labels, Algorithm discovers clusters and patterns on its own — used in customer segmentation, anomaly detection, and recommendation systems.