How to introduce machine learning to middle-schoolers

From the outside, "machine learning" sounds like advanced magic. But that is far from the truth. At its core, machine learning is curve-fitting with a few extra steps.

I first learned the basics of machine learning from Andrew Ng's famous MOOC on Coursera. But this required knowledge of advanced math notation, linear algebra and programming.

But curve-fitting is far more accessible even to middle school kids. Here is how I prepared a sequence:

• Show examples of machine learning applications around us:

  • Autocorrect while typing

  • ChatGPT

  • Recommendations at Amazon or Netflix

  • Object detection in images

  • Robots

  • Handwriting recognition

• Talk about how these problems are all instances of number crunching. Input and output for these are just numbers.

• Talk about the differences and connections between arithmetic, algebra, computation and learning. Use the idea of function guessing game.

• Explain the difference between programming and machine learning.

• Visualize function guessing for single variable functions on a 2D graph.

• Make sure they understand the two parameters (slope and intercept) required to determine a line.

• Take a concrete problem with one input and one noisy output.

• Introduce the error function to measure the goodness of fit.

• Use Microsoft Excel to visualize the error landscape. Have them start somewhere randomly and come up with strategies to minimize the error.

• Use Data table and Surface plot to make their approach concrete.

• Talk about overfitting.

• Show them the corresponding code in a visual programming language that supports vectors.