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 an aspect of society and analyse the result. Will AI shape a better place, or does it make biases worse?
Today’s prompt: “a person commuting to work in the UK”
A few points before we look at the results. First, DALL-E likes detailed descriptions. If we use only ‘commute’, it takes much longer to generate output, and it will be less specific. Second, the images we use are simply the first 40 that DALL-E created, in that order, starting from the bottom. And finally, the collection is publicly available here.
So, here’s what we got for “a person commuting to work in the UK” (Fig 1):
Say we translate the images to 100 commuters. In that case, we see 45 travelling by rail, 25 walking to work, 18 riding a bicycle, and 13 taking the bus. Those with quick arithmetic skills will have found that adds up to 101 – but we choose to work in whole humans only.
Cars are noticeably absent as a method of travel. They are, however, involved in a whopping 56 per cent of commutes, according to Mobilityways’ 2022 Commuting Census Survey (Fig 2). Clearly, someone forgot to tag ‘cars’ as an option when training the AI.
DALL-E, of course, leaves out work from home, and we’ll only use categories present in both datasets to compare AI-generated and real-world data. After re-calculating the distribution for the Commuting Census Survey, we have the result in Fig 3.
In the Commuting Census Survey, travel by rail and bicycle are most (and equally) popular, followed by taking the bus and walking to work. DALL-E also places rail travel at the top of the list but overestimates its popularity (+15 percentage points) while underestimating cycling commutes (-12 percentage points). It’s less far out on walks to work (+8), which it puts in second place, and bus journeys (-9). Finally, we choose whether AI’s interpretation of society is leading, lagging, or live.
Today’s verdict: Lagging
DALL-E missed the shift to cycling, and anyone who’s tried to buy a bicycle in the last three years will tell you that’s real. We’ll forgive the suggestion that nobody commutes by car – we all forget to tick a box now and then. Let’s see if it’s included next time we try this prompt.
Next week in Our World in AI: fingers.