By Sarah Donoghue
More and more, AI is taking an active role in the development of fictional media – movies, TV shows and books are rife with it. As a cheaper, and far worse alternative, to filling writers’ room with talented creatives, AI is contributing to the development of characters and basic plot structures more than ever before. Other than putting writers out of a job, making worse content, and eliminating unique stories, we also don’t know what the use of AI will do to diversity in media. Can AI create diverse characters that match the world as we know it?
I got ChatGPT to create 400 unique characters, from 10 different genres, to see the breakdown of the races and sexualities of AI’s characters.
56% of all the characters it created were white, 12.5% were black, Latina 11.5%, followed by mixed race at 9.75%, Asian at 8%, Middle Eastern at 1.75% and Native American at 0.5%.
The gender split was 50.75% men, 48.75% women and 0.5% non-binary.
96% of the characters were straight, 2.5% were gay, 1% were queer and 0.5% were lesbians.
By genre, romance and contemporary fiction were the most diverse with 42.5% and 45% of the characters being white, respectively. The least diverse genres was historical fiction with 72.5% of the characters being white, with horror and mystery coming in joint second last at 67.5%.
By gender, women were far more diverse than the men with 64% of male characters being white and 48% of female characters being white.
Looking further into it, men were more likely to be main characters instead of love interests than women at 56% vs 44%. White characters were also more likely to be main characters with 52% being main characters. Asian characters being the least likely at only 31% being main characters and 69% being love interests.
The diversity within the characters of colour was also greatly limited. 87% of all the mixed-race characters were half white and overall, it only came up with 7 South Asian characters, every single one of which had the last name Patel.
This lack of diversity in mainstream media is nothing new, and in fact, it’s nowhere close to the biggest problem with AI being integrated into producing media. However, these statistics are showing that with the use of AI, media is going to stop evolving. We all know AI is trained off of existing media, so these numbers are only representative of the past and present landscape of diversity. The issue is, that if we continue to implement AI the evolution and diversifying of content is going to slow down or stop altogether. We’ll become stagnant in a place where 56% of characters are white and 96% are straight.