Why is there bias in AI models?

Much of the data that AI is trained on comes from media (books, online articles, movies, TV shows) from the US, UK and Australia (as well as other Western countries).

There are often stereotypes in those media, like the AI-generated 'professors', and the data often reflects common ideas and traditions in Western culture. But there are different perspectives in different cultures.

Historical biases can also impact responses from AI models. For example, in the past, CEOs were mostly men, so generated images of CEOs are more likely to be men (even if it's not true now).

An example of cultural differences is shown below. In Western countries (like Australia): the colour red is often associated with danger, warnings or errors. In Asian countries, red is often associated with luck and wealth.