Using AI to analyze pedestrian use of public space
This is a very cool paper which analyzes how people use public space. They use video footage from 1980 and from 2010 from three sites in New York, Philadelphia and Boston for comparison. They use AI methods to analyze videos and study group dynamics in public spaces. This approach builds on existing visual observationanalysis to track what people do: how fast they walk, whether they linger, whether they encounter others, whether they form groups. The general findings are that people walk 15% faster 30 years later, tend to linger less (30%), meet up with others less.
It’s all pretty interesting and deserves a lot more analysis and study. I’m a big believer in public space, sidewalks, and walking. So understanding how and why people use them, and improving them is important to me. And there’s surprisingly little real data or evidence-based analysis.
Finding that people walk faster is strange to me. Based on national statistics, people walk significantly less than they used to, so why would they walk faster now? They’re probably not as good at walking as they were. The authors hypothesize that time is now worth more then it was then, so the opportunity cost of walking slower is higher. Time is money, so people don’t want to waste it. Time is worth more now, so people are motivated to walk faster, and linger less. Ok. Maybe.
Another, related, hypothesis is that people just have less discretionary time, so they walk faster and don’t let themselves get distracted as much. Similarly, the dawdling on the street as entertainment has more competition in 2010 than it did in 1980. Standing around watching people walk by is fun, but maybe not as fun as playing games on your phone or watching Netflix.
The authors note that crime was higher in 1980, so that might seem to mitigate the trend: people are less worried about being out in public, so they might time more time with it. But that isn’t what their evidence shows.
The study found most people are walking alone and the rate was steady: about 67% were single pedestrians. There seemed to be a reduction in the number of people who meet up and form fewer groups in pre-planned meetings or spontaneous encounters — but the data isn’t that good, nor the trend that strong. The authors conclude suggesting “that public spaces are becoming less conducive to planned and unplanned social encounters” but I really don’t see how their evidence supports that. It could be true, but it’s just not supported by the data or analysis. Since the Metropolitan Museum is one of their site and contradicted the evidence in other sites, the authors say, “when urban amenities are sufficiently compelling, they can still encourage slower social behavior”, which is pretty obvious.
The paper is pretty limited, as the authors acknowledge. It compares 4 sites, across two time periods 30 years apart with only 10 hours of video footage. The comparisons happen on different days of the week and different times of day*:
- 1980–04–10: Thursday
- 2008–12–05: Friday
- 1979–08–30: Thursday
- 2010–05–19: Wednesday
- 1979–09–28: Friday
- 2010–05–21: Friday
- 1979–06–02: Saturday
- 2010–08–08: Sunday
So, the claims should be taken with a lot of caveats. Important social and technological changes aren’t accounted for — like cell phones: people meet up and form groups less when they’re talking on the phone. Are these streetscapes the same 30 years later?
So, the paper’s mainly cool because it shows the potential of this kind of analysis. Using video footage — of which I think there’s A LOT — you can study a lot about public space, transportation, social interactions. But it’s pretty labor intensive to do that. But now, utilizing AI tools, it’s a lot easier and faster. There’s a lot we can learn. You could run experiments pretty easily to change streetscapes and see how they affect things. You can analyze inclusivity and safety. There’s a huge potential here.
*I asked ChatGPT (and Claude) to tell me what day of the week these were and this was the answer. Just took a second.
ENDS//
