Blog: Using Photographs as Data Sources to Tell Societal Stories
Even though humans produce more photographs than ever, their value is becoming increasingly hidden. The problem is not that photographs have lost their meaning, but that we are surrounded by so many of them that we have started to stop looking at them carefully.
This loss of attention matters because photography can do much more than simply preserve a memory. Besides freezing a moment, photographs can document a place, show damage after a crisis, capture public emotion, or make visible something that numbers alone may fail to express. Today, society is increasingly documented through images produced by smartphones, social media, Google Street View, CCTV, news media, and public archives.
For societal computing, this makes photographs especially important. They can become data sources for understanding social life, infrastructure, inequality, crisis, collective behavior, and public space. However, in order to use them responsibly, we first need to take them seriously again and see them as images shaped by people, technologies, platforms, and social context.
This raises an important question:
How can we use photographs as data to tell responsible societal stories?
This question is not only about using images in machine learning, which has become widely popular in recent years. It is also about understanding how photographs become data inside social and technical systems. A photograph is more than the set of pixels it is generated from. It is produced by a person, through a device, in a particular place, at a particular moment, and often circulated through platforms that rank, filter, recommend, moderate, or remove it. In other words, photographs are socio-technical data.
Photographs are not neutral evidence
But if photographs are socio-technical data, then we also need to be careful about how we interpret them. This starts with recognizing that photographs are not neutral evidence.
It can be tempting to treat photographs as direct evidence. One might easily claim, “This happened because we can see it.” However, visual data needs to be interpreted within its context. A photograph of a protest, a damaged street, a crowded train station, or an empty public square does not automatically explain the full social situation behind it.
Before moving from image to interpretation, we should ask three questions:
- What is visible?
- What does it represent socially?
- What is missing?
This matters because photographs often show only part of the story. They may reveal visible damage, infrastructure, crowds, symbols, emotion, or public behavior. At the same time, they may hide private suffering, people without access to cameras, communities affected by censorship, or images removed by platform moderation.
As a result, responsible interpretation should avoid jumping from “I see X” to “society is Y.” Stronger societal analysis requires images to be combined with captions, interviews, maps, statistics, field reports, local knowledge, or administrative records.
What kind of data is inside a photograph?
Once we accept that photographs need context, the next question is what kind of context they actually contain.
Around a single image, there can be many layers of data. Some data that can be collected from a photograph include:
Visual content: people, buildings, streets, objects, symbols, damage, colors, movement, and composition.
Spatial data: location, neighborhood, infrastructure, distance, mobility, and the built environment.
Temporal data: when the photograph was taken, uploaded, edited, or shared.
Metadata: device type, camera settings, geotags, captions, hashtags, file information, authorship, or rights.
Platform data: likes, shares, comments, ranking, moderation, visibility, deletion, and recommendation systems.
Interpretive data: what viewers, researchers, annotators, or machine learning models infer from the image.
This means that in societal computing, the dataset is often bigger than the image itself. A photograph can connect visual evidence, metadata, social behavior, and platform infrastructure.
Methods for analyzing photographs
Because photographs contain many layers of information, there is no single correct method for analyzing them. The method depends on the societal question we want to answer.
Different societal questions require different methods. If the goal is to understand lived experience, photo-elicitation or interviews may be more useful than automated detection. If the goal is crisis response, computer vision may help identify visible damage or blocked roads at scale. If the goal is to understand platform power, metadata and circulation patterns may matter more than the visual content alone.
Some common approaches include manual visual coding, where researchers classify image content with a codebook; photo-elicitation, where participants discuss images to explain memories and experiences; computer vision, where models detect objects, scenes, damage, faces, vehicles, or spatial features; metadata analysis, using EXIF data, timestamps, device information, captions, hashtags, and geolocation; geospatial analysis, linking photographs to maps and neighborhoods; multimodal analysis, combining images with text, comments, and social networks; and mixed methods, combining computational scale with human interpretation. The important point is that the method should follow the societal question, not the other way around.
Crisis images as societal data
One area where this becomes especially visible is crisis analysis, where photographs can provide fast and emotionally powerful traces of events.
Disasters show why visual data can be powerful but also limited. During crises, people often post photographs with short text updates on social media. These images can reveal visible damage, destroyed buildings, blocked roads, rescue activity, crowds, and urgent needs.
For example, multimodal crisis datasets combine tweets and images to classify whether posts are informative, whether they show infrastructure damage, affected individuals, rescue efforts, donations, or different levels of damage severity. This kind of work can support crisis monitoring and help identify where attention may be needed.
But photographs alone cannot fully explain mobility, emotions, vulnerability, or public needs. They should be read together with text, maps, official reports, and local knowledge. Otherwise, the analysis may become technically impressive but socially incomplete.
Bias, visibility, and platform effects
Before using photographs as evidence at scale, we also need to ask whose images are included and whose are missing. We should not forget that photograph datasets are not random samples of society.
Sometimes people and places are photographed more often. This can be simply because they are more visible, more connected, more touristic, more dramatic, or more platform-friendly. Others may be underrepresented because of privacy concerns, danger, poverty, censorship, limited internet access, fear of exposure, and many other factors.
This means that photo-based analysis is never only about what appears in the image. It is also about asking who had the opportunity, safety, resources, or permission to become visible in the first place.
Platforms also shape what becomes visible. Instagram, Google Maps, news media, and CCTV do not produce the same kind of visual evidence. Hashtags, search rankings, recommendation systems, moderation rules, and deletion policies all influence which images are collected, seen, and analyzed.
This is why bias in photograph-based societal computing is not only a technical issue. It is also about power: who becomes visible, who remains invisible, and whose reality becomes computable.
Computer vision systems can also reproduce social bias when training data is imbalanced. If some groups are less represented or poorly represented in datasets, models may perform worse for them. This can turn visual analysis into a system that amplifies existing inequalities rather than helping us understand them.
Ethics, privacy, and power
These questions of visibility are closely connected to ethics, because photographs can make people visible in ways that may also expose them to harm.
Photographs can expose people. A person may consent to being photographed in one context, but not to being scraped, classified, analyzed, or used for model training. Faces, homes, uniforms, injuries, protests, locations, and crises can create privacy and safety risks.
This is especially important when visual data is used for identification, policing, border control, surveillance, or social sorting. Ethical photograph-based research must consider consent, anonymization, dignity, copyright, documentation, and possible harm.
Responsible visual research should ask not only “Can we analyze this image?” but also “What could happen to the people represented here if we do?”
A responsible storytelling framework
For this reason, photograph-based societal computing needs not only technical methods, but also a responsible storytelling framework.
To use photographs responsibly in societal computing, we need a framework that connects technical analysis with ethical reflection. One can start from the following place:
Define the societal question. Do not start only because photographs are available.
Document provenance. Where did the photographs come from? Who produced them? What was excluded?
Separate observation from interpretation. “There is visible flood damage” is different from “this community was neglected.”
Use mixed evidence. Combine images with text, maps, interviews, statistics, field reports, and local knowledge.
Audit bias and missingness. Ask who is overrepresented, who is invisible, and which platform rules shaped the dataset.
Protect people. Blur faces, remove sensitive metadata, aggregate results, and avoid unnecessary identification.
Report uncertainty. Include limitations, ambiguity, model errors, missing context, and alternative explanations.
Seeing society without extracting from people
Ultimately, the goal is not simply to collect more images or build stronger models. The goal is to use visual data in a way that helps us understand society without extracting from the people represented in it.
Photographs are powerful societal data sources because they can show social life, space, infrastructure, emotion, inequality, conflict, and collective behavior. But their value depends on critical and context-aware interpretation.
Good photograph-based societal computing needs both computation and interpretation. It needs methods that can work at scale, but also humility about what images cannot tell us on their own.
Responsible storytelling means making society visible without turning people into objects of extraction or surveillance.
Photographs can help us tell societal stories, but only if we treat them not as simple evidence, but as situated socio-technical data.
- Written by Belin Korukoglu (https://www.linkedin.com/in/belin-korukoglu/)
