Gleeful by default? Grumpy by design? 

If you think you have read enough about the underlying biased foundation of generative AI, let me assure you that there’s more to come! For every one step forward that we think we take as individuals towards creating a gender-neutral surrounding, generative AI pushes us 5 steps back. You can’t blame it though, can you? It has been fed decades old data for training and a child can only become what its upbringing feeds it.  

Below is a screenshot from this morning (17 Sept, 2024). I reached out to ChatGPT image generator to experiment with a possible visual idea that I had – a comical scene showing a person unwrapping a series of boxes. 

Bias 1: ‘person’ was interpreted as male

Also, the emotion of frustration that the generator identified was in fact a good emotion to show here, because, although I didn’t say it, I was looking for that.

Then I asked it to switch the male to a female.

Bias 2: the female is now shown smiling, instead of frustrated. This is despite the fact that my prompt only asked to switch the gender.

Before you come to me defending the prompt and the descriptions, before you come to me explaining the rationale of the interpretation, let me say that I did manage to get a slightly more decent image with more explanation and better prompt. The point of this little anecdote is to just display the gap in training our AI model. Every effort we make to teach it better will only serve future humankind better!

PS: I showed this chat interaction to 5 of my neighbors, all in the age group of 48-65. They were amazed by the results purely from technology point of view. They did not even realize that there was a bias here until I called it out eventually.

PPS: I shared this chat screenshot with my two boys who are 10 and 12 years of age. They did not catch the ‘person’ bias or the ‘smiling woman’ bias.

The conclusion and impact of AI bias is for you to now gauge!

Shakthi had an interesting perspective to this. Read here

Storyteller