Sometimes we write and post things on social media in a hurry. Such posts can hurt people and they feel bullied. Wouldn't it be great if an AI could check our posts as we write them and warn us if they are potentially hurtful?

← In the table on the left, enter up to six training posts in each of the two categories. For example, 'i like you' is a kind thing. But 'you smell' is a mean thing. Observe how the AI configures its input artificial neurons as you type. Should you need some inspiration, take a look at the table below.

Before a neural network can think clearly, it has to be trained. Let's start the training process. Click on the 'Start Learning' button at the bottom to train the network. Adjust the speed with the slider at the bottom of the window.

Once learning is complete, enter a post in the bar at the bottom that uses some of the words in the list. For example, 'you smell like a slimeball'. Observe how the ANN uses keywords as input and proposes a possible classification into either kind or mean things.

Try a few other combinations of words. Document each experiment on a sheet of paper, together with the classification that the AI proposes.

You will probably encounter posts for which 'Kind things' and 'Mean things' are not black or white. They might be grey. This means that the AI is uncertain. This can happen in cases where kind words are used in combination with mean words. An AI find it hard to understand irony and sarcasm, because they are playing effectively with opposite meanings.

This scenario has been inspired by and developed in cooperation between the Digital Technologies Institute, Australiaand Apps For Good, United Kingdom.