Use the Code.org Ocean Activity to train an AI model by labelling different objects. Observe how incorrect, insufficient or biased training data affects the model's predictions. Discuss how AI learns and specifically how supervised learning works.
Click here to open the Ocean activity on Code.org.
Exercise (2) - Unplugged activity
Group the given set of objects in any way that makes sense to you. No labels are given and no groups are named beforehand. Compare your grouping with the rest of the class and discuss how the same set can be grouped correctly in more than one way - by living and non-living, by type, or by where each thing is found. Discuss how this differs from Exercise 1, where every example was labelled before the model learned from it.
Click here to open the activity sheet for this exercise.
Check the availability of the AI tools used in this lesson and read the instruction documents for all the exercises before the class.
Let students see the AI’s correct predictions, mistakes, and how bias can affect its results.
AI continuously adjusts its decisions based on what it is learning.
Discuss about supervised and unsupervised learning
Tell students that more correct and fair data helps the AI make better guesses.
Points to Ponder:
Someone labelled a plastic bottle as a fish by mistake, and the AI started calling bottles fish. Whose mistake was the wrong answer - the AI’s or ours?
The AI got better as you labelled more examples. Was that because it started to understand what a fish is, or because it had seen more of them?
In Exercise 1 there was one right answer. In Exercise 2 your class found several groupings that were all correct. Which kind of problem do you think is harder for AI, and why?