When Residents and Algorithms See Different Problems

An analysis of roughly 187,000 citizen reports and nearly 5,000 computer vision detections in Jamaica Plain, MA shows why cities need both perspectives and raises critical questions about who gets heard. Data from 311 available at data.boston.gov/. Computer vision data provided for research purposes from Cyvl. 

In 2015, the city of Boston launched a redesigned 311 mobile app, making it easier for residents to report urban problems like potholes and broken streetlights. In Jamaica Plain, a nearly 4.5 square mile neighborhood in the city, uptake of the new mobile app was dramatic; based on an analysis of historical data, requests for service surged 234.5%, representing a radical transformation of how residents engaged with city services. This success story, however, raises an uncomfortable question: if civic participation now depends heavily on digital tools like smartphones and apps, are we hearing equally from all residents across neighborhoods?

To address that question, this article takes an innovative approach that sheds light on this equity challenge and discusses how cities can balance inputs for fair decision-making. An analysis of 14 years of Jamaica Plain's 187,408 311 resident reports was compared with a snapshot of 4,856 pavement defects (identified from Cyvl's computer vision system, which is a combination of LiDAR technology, high-resolution 360-degree cameras, and artificial intelligence), discovered something striking: only 7.9% overlap. 

Rather than proving one system right and the other wrong, the result of this analysis reveals how different monitoring approaches capture fundamentally different dimensions of urban problems and why that matters for equity. Taken together, they demonstrate how technology can complement civic inputs to produce a more complete and equitable diagnosis of urban problems, than either approach can achieve in isolation.

Consider what this might mean for health challenges, issues like air quality and extreme heat, where harms are diffuse, cumulative, and often invisible in day-to-day experience. If cities rely primarily on resident-initiated reporting, they risk constructing a picture of urban health that reflects not exposure, but capacity: who has the time, tools, trust, and technical fluency to engage. Digital civic systems may appear responsive and data-rich, while systematically underrepresenting communities with limited digital access, lower trust in government, or fewer opportunities to document and submit complaints.

Who Benefited Most from the App Revolution?

When Boston's 311 app launched in August 2015 it brought together existing yet disparate reporting systems like the non-emergency Mayor's Hotline and the Citizens Connect mobile app. After the launch, both the number of and types of reports in Jamaica Plain increased dramatically, according to an analysis of historical data. Quality-of-life complaints exploded 703%, from 6,363 to 44,737. Parking enforcement requests jumped from less than 1% to 17% of all reports. Barriers to reporting minor annoyances seemed to have collapsed. 

Jamaica Plain's 234.5% increase far exceeded Boston's citywide 82.3% growth. Why? Because demographics matter. Both is 2015 and currently, Jamaica Plain skews younger, more educated, and more tech-savvy than Boston overall, which is exactly the profile of early smartphone app adopters and of people who would rather send text messages or use an app than call in a complaint to a city agency.

This is the equity concern in a microcosm. Mobile apps are incredibly effective tools for civic engagement, but they amplify the voices of residents who are already comfortable with technology, speak the native language, and have smartphones with data plans. The success of the 311 app in Jamaica Plain highlights a potential gap: what about neighborhoods where these conditions don't hold?

What Citizens Choose to Report

Over 14 years, Jamaica Plain residents used 311 to report:13,896 pothole repair requests, 5,790 sign problems, massive increases in parking enforcement and trash complaints, and very few technical pavement defects like cracking or weathering. This pattern is well in line with rational citizen behavior because people report problems that directly impact their daily lives: the pothole that could damage their car (if they own one), the missing stop sign at a dangerous intersection, and the illegally parked car blocking their driveway. They don't report the hairline pavement cracks that engineers know signal future deterioration.

But a deeper pattern emerges here about participation. The 311 system is inherently reactive; it depends on someone noticing a problem, caring enough to – and having the means to – report it, and trusting that reporting will lead to action. Each of these steps introduces potential bias.

What Systematic Monitoring Reveals

This is where the Cyvl data becomes an interesting and worthwhile complement, not because it's "better," but because it's fundamentally different. The system detected 3,114 instances of pavement cracking (64% of detections), 715 cases of weathering, 382 patches requiring evaluation, and various technical defects citizens don't typically recognize.

The key word here is "systematic." Unlike 311, which depends on citizen initiative, the vehicle used by Cyvl drives every street regardless of who lives there, their smartphone ownership, or their civic engagement levels. It applies the same detection criteria uniformly across wealthy and poor neighborhoods regardless of whether residents actively participate in city processes.

When the datasets were spatially matched (looking for 311 reports within 50 meters of Cyvl’s detections) only 79 out of 1,000 sampled reports overlapped. This isn't necessarily a failure of either system; it's evidence that they're measuring different things.

Consider what this means: 92% of citizen infrastructure reports had no corresponding computer vision detection nearby and 99% of computer vision detections had no citizen report.

These systems see different aspects of the same streets. Citizens report acute problems that affect their daily experience, while computer systems systematically document technical conditions across all infrastructure, regardless of whether anyone complains. This complementarity has profound equity implications.

The Equity Question: Whose Problems Get Seen?

The uncomfortable reality is that 311 participation is not evenly distributed. Research has consistently shown that civic technology engagement correlates with income, education, and digital access. If Jamaica Plain, a relatively engaged and tech-savvy neighborhood, saw 234% growth in 311 reports, what happened in neighborhoods with lower smartphone adoption, larger immigrant populations, or historical mistrust of city government?

The Cyvl data provides a partial answer, as it shows infrastructure problems in areas regardless of whether residents are active 311 users. This systematic coverage acts as a potential equity check against the participation bias inherent in citizen reporting systems. But critically, we can't simply declare computer vision the "objective" solution. The scans represent a snapshot in time, from individual scanning events, while 311 data reflect 14 years of continuous community feedback about what matters to residents based on their own lived experiences.

Beyond Potholes: What Gets Prioritized

The divergence between systems uncovers something deeper about urban governance priorities. When cities rely primarily on 311 data, they're letting residents set the agenda; however, the residents setting the agenda are not perfectly representative of the city and participation is inequitable.

311 apps have made it extraordinarily easy for tech-savvy residents to report minor quality-of-life annoyances – parking complaints surged and graffiti reports multiplied. These apps have been important landmarks in the relationship between residents and their government, but they reflect the priorities of residents who are comfortable using apps to engage with government.

Meanwhile, systematic monitoring shows technical infrastructure deterioration that may not generate complaints until it becomes severe. This suggests a resource allocation challenge: should cities respond primarily to what residents report (democratic but potentially biased toward engaged communities) or what technical systems detect (systematic but disconnected from lived experience)?

The answer, the data suggests, is both.