The Cold Start Problem

Explain to students that optimization goals are just half of how algorithms make decisions about what to show you.

The other half is your inputs: what the algorithm knows about you. For example, if the algorithm is optimized for watch time, it need to know what will make you watch longer.

There are two kinds of inputs: explicit and implicit. Explicit inputs are ones you know you are giving to the algorithm, while implicit inputs are ones you don’t know you’re sending. These come from the app or website watching what you do or getting information about you from someone else. For example, YouTube can use your Google search history, because they are owned by the same company. Other apps or websites may buy information from data brokers.

Have students complete the sorting exercise in the student chapter The Cold Start Problem. Go through the inputs and point out that these are by no means all the inputs that algorithms use, just some examples.

Once students have sorted the inputs into explicit and implicit ones, ask:

After students have discussed for a few minutes, explain that algorithms rely on implicit inputs to overcome the Cold Start Problem. This just means that when you start using something that has algorithmic recommendations, the algorithm doesn’t know much about you, so its recommendations aren’t very accurate.

The more implicit inputs they use, the more quickly the algorithm becomes more accurate: if it’s ever seemed to you that an app could “read your mind,” or knew you better that you know yourself, that probably means it was using a lot of implicit inputs.

As the streaming music service Spotify says to advertisers, “The more they stream, the more we learn.”