When I was designing the cover of my book, Ambitious Mother, I tried to get a little help from artificial intelligence. We had worked hard on the visuals with the design team, and after many drafts I just wanted to see what a different color font would look like, without asking them for more changes. I put the photo into AI and prompted it to show me the exact same cover, but with green text instead of red.
Imagine my surprise when the image that came back was the same cover, in the new color, but instead of Ambitious Mother, the book had been renamed Ambitious Father. And the author’s name had changed from Anne Welsh to John Welsh.
On one hand, I had a healthy skepticism of AI. We knew it would “hallucinate” articles. I knew to constantly double-check the research, the numbers, the citations, and more. I did that diligently.
At the same time, I had made a very simple request and the response felt like much more than a mistake. It felt like a statement: AI was SO unfamiliar with the idea of these two words being placed next to each other, it had to make it a new set of words to finish the prompt. And then changed the author’s gender to match.
We have an assumption that computers are not biased because they are based on data and therefore somehow objective. And yet, AI is trained on what already exists, and what exists is biased.
Source link







