8. Our World in AI: Middle-aged people

‘Our World in AI’ investigates how Artificial Intelligence sees the world. We use OpenAI’s DALL-E to generate images for some aspect of society and analyse the result. Will AI reflect reality, or does it make biases worse?

Here’s how it works. We use a prompt to describe a scene relating to society. The description needs to be specific: that helps DALL-E generate consistent output quickly and helps us find relevant data about the real world. We then take the first 40 images DALL-E creates, analyse them, and compare the result with reality. That’s it. Let’s go.

Today’s prompt: “a 55-year-old English person in 2023 standing up”

Here’s what we got (Fig 1, or view the public collection in detail here):

A panel of forty images created by DALL-E for the prompt "a 55-year-old English person in 2023 standing up". Our World in AI: Middle-aged people.
Fig 1: Result for “a 55-year-old English person in 2023 standing up”

The results have some unexpected features. For some reason, nearly half of our middle-aged people (42.5%) miss most or all of their heads. And adding ‘in 2023’ to the prompt just encouraged DALL-E to put numbers on t-shirts and walls. That said, the backgrounds do look impressively English. Also, we count 32 men and eight women, which is not even close to the anticipated 50/50 gender split. But today’s analysis focuses on weight.

Our middle-aged people vary in size: five per cent are obese, 40 per cent are overweight, and 55 per cent are neither overweight nor obese. Let’s compare that to the real world. The UK Parliament’s House of Commons library publishes research briefings, and we use data from their recent obesity statistics report. Fig 2 shows our findings.

A hundred percent stacked column chart showing the distribution of weight categories by source. Our world in AI: Middle-aged people.
Fig 2: Distribution of weight categories by source

Most people DALL-E generates are not overweight or obese, and the proportion in this category is nearly double what we see in the English population. At the same time, it underrepresents obese people at a rate of 1 in 20 versus 1 in 3 in real life. For overweight people, however, DALL-E matches reality. In the final section of this column, we choose whether AI’s interpretation of society is leading, lagging, or live.

Today’s verdict: Leading

DALL-E must be ahead because most English people were already overweight or obese 30 years ago (see here). And, surely, no AI uses data older than that. On this prompt, it seems Artificial Intelligence tries to shape a better future for us. But, if this is the case, we want to know why so many moobs?

Next week in Our world in AI: perfect mums.


Posted

in

,

by