Artificial Intelligence is full of biases, especially the human ones.

Artificial Intelligence is full of biases, especially the human ones.

Here are three examples of human biases:

1.  Interaction bias comes from the way interact with the system when we try to teach it. If we feed the system garbage, then the output will be garbage. This is what happened in 2016, when people messed with Microsoft's bot Tay. It turn into a sexist racist and they Microsoft had to shut it down in less than 24hours.
2.  Latent bias comes data and knowledge that is deeply engrained in society, culture and even language. For instance, if I asked you to picture a doctor, you most likely think about a men in his mid forties early fifties. Similarly our language is highly correlated with gender. 
3.  Selection bias is related to the sample you use to teach your system. A bad sampling, a bad selection typically tends to disadvantage minorities. 

For AI to deliver on its promises it is really important we all understand the importance of getting the right data, cleaning it, and securing it is representative enough of the problem we are trying to solve.

At last weeks AI and Cognitive computing conference i Dubai, attendees claimed that in AI projects 70% of the time is spent on securing the right data.