Next month’s virtual IEEE International Smart Cities Conference is sure to include the usual discussions on AI-based sensors and prediction systems intended to make cities safer. But if a city gets “too smart”—i.e. totally driven by data and sensors—it runs the risk of becoming impersonal and losing its grit and the element of surprise, argues Annelien Smets, a smart cities researcher at Belgium’s Vrije Universiteit Brussel.
“Most of the time we know where to go and how to get there, or at least our phone does,” Smets says. “We are so used to having all information at our fingertips that we’re no longer used to searching things…but that triggers a very different state of mind. We’re goal-oriented—there is no room for the unknown.”
As co-chair of the Serendipity in the Smart City workshop, Smets will present research at the conference on how to get AI to be more interesting. “From watching movies, to buying clothes, to ordering food, to writing texts, all the systems that we are using for these activities are trying to predict our preference or next steps as accurately as possible,” Smets says.
The services layer, which guides residents and visitors, via apps, digital signage, and device-based searches, uses online profiling to build out its personal recommendation engines. Essentially, if you’re trapped within your usual information bubble, you’re not going to hear about stuff outside your comfort zone.
In a paper co-authored with Professor Pieter Ballon, an abstract of which appeared in the July issue of the Journal of the Association for Information Science and TechnologyJournal of the Association for Information Science and Technology, Smets illustrates how cities are now hybrid physical-virtual spaces. But predictive recommendation agencies are removing serendipitous encounters in the city, threatening the urban ecosystem.
“What is distinct now, is the fact that the decisions and initiatives that cities take, are driven by data and information technologies,” says Smets. “This is what is known as datafication. One of the most essential characteristics of it, is that it can make the invisible visible. We now have the availability of data and the ability to act upon it. I’m intrigued by the transformation of this data into actionable and hence valuable information. To me, this datafication is the true essence of a smart city. But it has its drawbacks.”
Using Technology to Monitor, Not Control
Growing up in Brussels, Smets was long-fascinated by smart city technologies that improved urban challenges such as traffic flows. But it wasn’t until she pursued her master’s degree in AI that she started to work with recommender systems and personalization in more depth.
“Both personalization and smart cities really trigger an immense fascination,” Smets explains. “Not only because of how it all works, but also because it seems like there is a thin line between what we accept and what we think goes too far, and that’s very interesting too.”
This thin line is using technology to monitor, but not control. Smets and I talked about how COVID-19 has forced cities to get smart, to keep everyone safe, but perhaps at the cost of the serendipitous encounters that made city-living so exciting.
For example, here in Los Angeles where I live, a rapid increase in innovation has taken place. Don’t push the button to cross the street; wave at the sensor on the traffic control pole to stay germ-free. Motion sensors open doors, and localized HVAC systems power up to cool and aerate areas when populated by people. Proximity sensors go off in crowded areas, particularly medical facilities, if you get too close to the next person in line. Track-and-trace systems are in place and ping you if you get exposed to the virus. Anonymized data from mobile phones and onboard travel card swipe screens have reconfigured bus routes to favor essential workers. Don’t even think about not wearing a mask on mass transit; the cameras will catch you.
Looking for a restaurant that’s (still) open (post-lockdown), and its rules on inside/outside dining? Or whether your local movie theater made it through the pandemic to screen Shang-Chi and the Legend of the Ten Rings this weekend? You can’t just show up anymore. All those transactions are online, and reservations/tickets require booking ahead of time.
Training AI to Be More Curious
If this all sounds rather sterile, you’d be right. But how can we train AI—which works on patterns, outliers, and a non-human obsession with getting the right answer—to be more curious?
“Designing for serendipity is [more] than just a piece of code,” Smets says. “We have to identify underlying learned biases in contemporary information systems and diversify the training data to encourage more interesting results. To do this, publicly available data resources need to be sufficiently diverse themselves.”
But only focusing on the training data is not enough. It is of equal importance to have a good understanding of what citizens actually experience as serendipity. This knowledge can then be used to drive more curiosity and surprise within the AI recommendation systems.
In a separate co-authored paper, Smets suggests approaching the smart city as a permanent Urban Living Lab. “Ideally, we would be able to survey people while they are in the moment, and while they are experiencing what we want to question them about,” Smets says.
“Such a method is called ‘context-aware experience-sampling.’ These contextual data streams, cross-referenced with real-time sources such as weather and traffic, give us the ability to directly interact with participants to gain a thicker understanding about their behavior and experiences,” she explains. “Then we create a rule-based experience sampling method that enables contextual inquiry based on the multiple and real-time inputs, augmenting the value of the data itself.”
For example, rather than restaurant recommender systems churning out the same-old top 10 lists, based on prior data, the AI could suggest a later transit option, based on the fact that a pop-up art exhibit happening with accompanying food trucks just let the algorithms know it’s rolling into town. Or, if you give the AI permissions within a social group, it can triangulate that friends are nearby with an hour to spare before their screening starts, and suggest you stop by.
Make the Invisible Visible
However, there’s little need for serendipity—AI-powered or not—if your capital city is under siege from flooding. The City of Things project, which Smets’ research group is also involved in, has covered Antwerp in sensors, gateways, and actuators.
“In this work, we’ve been focusing on flooding for a while,” says Smets. “In the next few years, cities worldwide will more frequently face flash floods, an excessive rainfall in a short period of time. One of the problems is that the sewage system is not able to deal with these large volumes of water, with all its consequences.”
Through the cohesive data mesh, technologies are able to make the invisible visible. “By means of real-time sensors in drains, the goal is to get more accurate information about the situation. On top of the data itself, there is also an entire layer of data analysis. The data that is gathered from the drains, could be combined with other weather data and historical data to make predictions. These predictions could then be used to act proactively and coordinate efforts.
“Of course there is also an important aspect of data reporting and visualization,” she adds. “It is often underestimated how important it is to know which data should be shown to whom and when. My colleagues worked together with firefighters and public servants to figure all these things out and develop prototypes.”
At the IEEE conference, scheduled for Sept. 7-10, much emphasis will be put on making cities safer, especially as AI becomes better at helping infrastructure withstand stresses. But once that’s in place, Smets and her researchers will ensure that it’s also able to deliver the serendipity that makes cities such compelling and enriching places to live.
