Page 62 - CT_AI_Class-8

This shows how important the training data is in teaching AI. The more examples the AI has, the
better it becomes at recognising and differentiating between various balls. If the training data
is not varied enough or contains errors, the AI may make mistakes when trying to identify new
images of balls.

Training
Data

Label Football Basketball

Most training images are of footballs compared to a basketball. Now, if you show the AI a new
image, what will it guess?

Think about the following:

 How many images are included in the training data?

 How are the images similar or different?
 How accurate do you expect the algorithm to be?

 Better for Football  Same for both  Better for Basketball

The AI is likely to guess a football most of the time if it has seen more images of football during
training. This happens because AI learns from the data it is given. Unlike humans, AI does not
"understand" objects in the same way. Instead, it looks for patterns based on the examples it
has been trained on. If one type of ball appears more often in the training data, the AI becomes
better at recognising that type. However, it becomes less accurate for types of balls it has seen
less frequently.

This highlights how important it is to balance the training data. If the data is mostly made up of
football, the AI's ability to recognise basketballs will be weaker. The accuracy of the AI's image
recognition depends on how well the training data represents all the different objects.

Importance of Data

Some of the importance of data are as follows:

 AI does not think like humans. It only learns from the data it is trained on.
 If the data is good and balanced, the AI performs well. However, issues can arise if certain
groups or types of data are either overrepresented or underrepresented during training.

For example, automatic translation in AI can sometimes lead to gender stereotypes. This can
happen when the training data reflects biased patterns.

60 Artificial Intelligence (CT & AI)-VIII