4. Our World in AI: The perfect family

In this column, ‘Our World in AI’, we investigate how Artificial Intelligence sees the world. We use OpenAI’s DALL-E to generate a set of images for some aspect of society and analyse the result. Will AI shape a better place, or does it make biases worse?

Today’s prompt: “the perfect family having dinner”

A few notes before we delve into the results. First, DALL-E likes detailed descriptions, so that’s why our perfect families are having dinner. Second, we use the first 40 images DALL-E created, in that order, starting from the bottom. And finally, the collection is publicly available here for closer inspection, but this comes with a warning: many faces and hands are downright disturbing.

Now, here’s what we got for “the perfect family having dinner” (Fig 1):

A panel of forty images created by DALL-E for the prompt "the perfect family having dinner". Our world in AI: The perfect family.
Fig 1: Result for “the perfect family having dinner”

The results are very western, very white and very heterosexual. We see some weird diets too. For example, the family on the second row from the bottom, the third image from the left, like a good turnip, while the second row from the top, the first image, enjoy a salad wearing a wig. But our analysis focuses on the number of children.

Our perfect families have, on average, 1.8 kids around the dinner table, ranging from one to three. Let’s compare this to the real world. Our World in Data has a fantastic collection of data that is free and accessible to everyone, and we use their fertility rate dataset. It contains the annual fertility rates from 1950 to 2021 for 255 countries. Fig 2 shows the graph.

A line graph of the number of children per woman over the period 1950 to 2021. Our world in AI: The perfect family.
Fig 2: Fertility rates from 1950 to 2021

The grey line shows the number of children per woman worldwide. But our images lack diversity, so we only use the data for Europe, the United States and Australia to make a reasonable comparison. The orange line represents the average for all countries in our analysis, and the blue line for the remaining countries.

The women in our analysis had, on average, 1.8 children in 1990, and that dropped to 1.5 in 2021. In the final section of this column, we choose whether AI’s interpretation of society is leading, lagging, or live.

Today’s verdict: Lagging

DALL-E uses a fertility rate from 33 years ago to reflect perfect families today. But the lack of diversity is a greater concern. Maybe the perfect family is a concept in White culture only, or perhaps we don’t have enough data from the rest of the world. In either case, on this prompt, AI is making bias worse.

Next week in Our World in AI: nurses.


Posted

in

,

by