Four questions on data types, statistical thinking, and the maths that sits behind every AI model.
Question 1 of 4
Question 1 of 4
A hospital reports that the average patient wait time is 12 minutes, but the median is 6 minutes. What does this most likely tell you?
AThe data was collected incorrectly
BA small number of very long waits are pulling the average up
CMost patients wait exactly 12 minutes
DThe median is always lower than the mean
Explanation
The mean is pulled upward by outliers — in this case a few patients who waited hours. The median (the middle value when sorted) is far less sensitive to extremes. In AI, a model optimised to minimise average error can still perform terribly on edge cases. Always ask for the distribution, not just the mean.
Question 2 of 4
A customer's written product review is which type of data?
AStructured data
BLabelled numerical data
CUnstructured data
DCategorical data
Explanation
Unstructured data has no fixed format or schema. Free-form text, images, audio, and video are all unstructured. A product review does not fit neatly into rows and columns — you cannot sort it by price or filter it by SKU without first processing the language. About 80% of all data generated in the world is unstructured.
Question 3 of 4
City A has 50 cinemas and 2,000 crimes per year. City B has 5 cinemas and 200 crimes. What is the most accurate conclusion?
ACinemas cause crime
BBoth are driven by a third variable — likely population size
CCrime causes people to build more cinemas
DThe data is too unreliable to draw any conclusion
Explanation
This is a classic confounding variable. Both cinemas and crime rates scale with population — larger cities have more of both, independently. Acting on this correlation (closing cinemas to reduce crime) would waste resources and miss the real causes. AI models trained on correlated data without understanding causation make exactly this kind of mistake.
Question 4 of 4
In a neural network, what do vectors primarily represent?
AError rates accumulated during training
BThe speed at which a GPU processes calculations
COrdered lists of numbers that encode real-world data like words or images
DThe number of layers in the model architecture
Explanation
Everything a neural network works with must be a number. Vectors are the bridge between the real world and computation. A word like "king" becomes a 300-dimensional vector where each position encodes something about its meaning based on how it was used in training text. An image becomes a vector of pixel values. Vectors are the universal language of neural networks.
🎉
Phase 2 Complete
4
out of 4
Strong statistical instincts. You understand data at the level needed to think critically about what any AI model is actually learning.