Optimizing Algorithms
Explain to students that algorithms are not programmed like regular computer programs. Instead, they are trained on data and then optimized for a particular goal. That means that they constantly test themselves to see how well they’re doing at meeting that goal, and then make changes to do it better.
For instance, the Netflix algorithm is optimized to help you find something that you want to watch in less than 90 seconds. It does this sorting and choosing thumbnails, as the last activity showed.
Tell students to think of another app they are familiar with that uses algorithmic recommendations. (Examples: TikTok, Instagram, YouTube, Spotify.)
Have them open the Optimizing Algorithms student chapter and go through the different optimization goals. (Point out that these are not the only possible goals, but they are common ones.)
Next, have them rank the optimization goals according to which they feel are the most important to the app or website. (You can have them do this individually, in pairs, or in small groups.)
If students have difficulty ranking the goals, tell them to think about what the algorithm rewards. For example:
- If it shows you more long videos and fewer short videos, it’s trying to boost watch time.
- If it shows you things that get you upset, it’s trying to boost engagement.
- If you worry about missing things if you’re away too long, it’s trying to boost daily active use.
Have students share their ranking with the class:
- How did they decide on their ranking?
- Did students (or groups) who evaluated the same app rank the goals similarly?
- If so, what makes it so clear what the app is optimized for?
- If not, why do they disagree?
Training an algorithm to achieve a particular goal.