A.I Bias -A Thought Experiment
When you do a quick google search of the word "Bias", you will get the following meaning —
inclination or prejudice for or against one person or group, especially in a way considered to be unfair.
The above definition of "Bias" is a human term, but how is it related to Artificial Intelligence? Let's consider a thought experiment to make things clear.
The Thought Experiment
Suppose an AI is being developed which sole purpose is to fold paper perfectly into two halves. The data which was prepared by the “data preparation team”, only consisted of white colored papers. The AI was trained on that data and then deployed to production. Everything went well and the AI surpassed superhuman abilities of folding papers but one day the color of the batch of paper that entered the production line was red and the AI rejected the entire batch of papers and this resulted in halting the production and a huge loss to the company.
Why Did This Happen?
Now, we must ask, Why did the AI rejected the batch of red paper when it was trained to fold paper? The answer to this is from the fact that the data on which the AI was trained only consisted of white papers and this led to the AI being “biased” towards whitepaper and thus considering other colors as anomalies.
Ok, But What's the Point?
The point I am trying to make here is that whenever an AI starts showing signs of biased behavior then we must understand that it’s not the AI that has become biased somehow. Instead it’s us humans who are biased and these human biases creeps into the AI even before the A.I is left out in the real world and then when the A.I starts operating in the real world then these biases are amplified and finally it shows up in an expected moment/situation.
Closing Thoughts
The data that we currently use such as image data ,sound data or may be any other data are not collected from diversified populations or locations. This in turn is making the data biased on racial, geographical, demographical and many other grounds and this should be considered seriously if we want to build safer, bias free AI.
If you liked this post then do give a clap here and share across. If you have any comments on this post or just want to give your thoughts on this topic then do comment in the comment section below.
