The "data" is aggregated from the web. All AI does is weight it based on prevalence and answer accordingly.

For example, if the data set was limited to homosexuals discussing topics then the AI is going to reflect homosexual bias. If it's limited to straight people discussing topics then the AI is going to reflect straight bias.

So, for a society that is majority white and straight, it's not surprising that the AI is biased in favor of white and straight ideals.