Trivial application for AI

[WhiteCube]

Here's what I'm trying to do. I want to demonstrate to friends and family, the limitations of a neural network.

I was thinking of a simple decision.

training data: I like the numbers 0, and 2, and 4, I dislike 9 and 5.

Then I ask the program to predict how I feel (a value from 0 to 1) about number 7.

I don't think that would be a fair demonstration. Too many possible reasons. (I dislike 7 because its odd)

I need a simple decision, that works okay, but requires so much training and perceptrons, that it makes the network look stupid.

 

[WhiteCube]

Is this a fair test?

Let's say I like banana 0.5, chocolate 0.95, and coconut 0.3.

I ask the program to predict how I would score apple.

3 inputs: fat, salt and sugar ratio of a food

1 output

Train it with the banana, chocolate and coconut. Test it with apple.

 

[WhiteCube]

Considering that people have a common preference for salt/sugar/fat ratios, that wouldn't be a good demonstration.

I can't find anything on the internet between single perceptron teaching demos, and full blown image recognition models.

I'm picking up an "emperor's new clothes" vibe.