Chapter 1: AI's View of the Perfect Family

In this section, titled "Our World in AI," we explore how Artificial Intelligence interprets societal norms. Utilizing OpenAI’s DALL-E, we generate a series of images based on specific prompts and analyze the outcomes. The central question is whether AI contributes to creating a more equitable society or amplifies existing biases.

Today's prompt involves "the ideal family enjoying dinner." Before we analyze the results, a few points are important to note. DALL-E thrives on detailed prompts, which is why our depiction involves families gathered for dinner. We will review the first 40 images generated by DALL-E, starting from the bottom. Additionally, the complete collection is publicly available for further scrutiny, but be advised that many of the representations of faces and hands may appear unsettling.

The results for the prompt "the ideal family having dinner" are as follows (see Figure 1):

Figure 1: Outcomes for the prompt "the ideal family enjoying dinner"

The images produced are predominantly Western, white, and heterosexual. Some portray peculiar dietary choices; for instance, in the second row from the bottom, the third image depicts a family enjoying a turnip, while the first image in the second row from the top shows a family member eating salad while donning a wig. However, our primary focus is the depiction of children.

On average, these imagined perfect families consist of 1.8 children at the dinner table, with a range from one to three. To provide context, we compare these findings with real-world data. Our World in Data offers an extensive and publicly accessible fertility rate dataset, which includes annual fertility rates from 1950 to 2021 across 255 countries. The data visualization can be seen in Figure 2.

Figure 2: Global fertility rates from 1950 to 2021

The grey line in the graph illustrates the average number of children per woman on a global scale. However, due to the lack of diversity in our AI-generated images, we limit our data analysis to Europe, the United States, and Australia for a more accurate comparison. The orange line indicates the average fertility rate across the countries we studied, while the blue line represents the remaining countries.

In our analysis, women had an average of 1.8 children in 1990, which declined to 1.5 by 2021. In the concluding section of this discussion, we assess whether AI's representation of society is ahead of, lagging behind, or accurately reflecting contemporary views.

Today's conclusion: Lagging

DALL-E relies on fertility data from over three decades ago to portray ideal families today. However, the lack of diversity is a more pressing issue. This raises the question: Is the notion of the perfect family confined to white culture, or do we simply lack comprehensive data from other regions? In any case, AI appears to exacerbate bias concerning this prompt.

Next week in "Our World in AI": a focus on nurses.

Chapter 2: Digital Perspectives on Immortality and Family Heritage

In this chapter, we will delve into the implications of AI on concepts of digital immortality and how it may shape family narratives.

The first video, "Living Forever Through AI: Digital Immortality and the Future of Death," offers insights into how AI could redefine our understanding of life and legacy.

In another fascinating discussion, we explore how AI can assist in documenting family histories.

The second video, "Writing a Family History Using AI Chat - James Tanner," showcases innovative ways to utilize AI for preserving family narratives.