“Last week, you joined Mr. Bean on his adventure through the basics of training, validation, testing. A few rules for how (not) to be an idiot with AI follow naturally from that example, but let me spell them out for you explicitly. (My condolences if they sound eerily familiar.)

The story so far

In the previous episode, you played the role of Mr. Bean’s patient professor. Since you’re getting good at it, let’s have you stick with it. I’ve often said that the best way to avoid the biggest pitfalls in applied AI is never to forget the basics of learning and teaching, so I’m secretly hoping you’ll keep the mindset of Mr. Bean’s professor forever.

Never forget the basics of learning and teaching!

Quick reminder of the three phases (explained in more detail in the previous article):

– Training phase: Mr. Bean (a placeholder for your beloved AI system) looks for patterns in the examples he saw in class, then turns these patterns into a model (recipe).

– Validation phase: See how Mr. Bean’s recipe fares on examples which he did not explicitly study. If the score looks good, he goes to the exam, otherwise he restarts the training phase.

– Test phase: Mr. Bean take the final exam and learns whether he is allowed to go to production or change majors.

The training phase is fairly straightforward—you shove examples into students (data into machine learning algorithms) and mostly hope for the best (I’m only half kidding). Learn more about how it works here.

Before we list the ways to be an AI idiot, let’s talk about the more subtle phases — validation and testing — from your esteemed professorial perspective.

The article: How to be an AI idiot – 7 ways to royally mess up your machine learning project