Images, Art & Video - Orientation

[HyunkuKwon]

Post questions about the following orienting reading:

Collins, Randall. 2009. “The Micro-sociology of Violent Confrontations” and “Confrontational Tension and Incompetent Violence” (beginning of Chapter 2) from Violence: A Microsociological Theory: 37-43.

[Raychanan]

The author investigated many forms of violence. For example, war, child abuse, and police violence. And the author underscores the importance of the range of variation in different situations.

But as I read these words, I expected the author to talk more about the less obvious violence, or the non-violent sources of direct violence. In today's world, when one organization and entity interacts or confronts another, they often do so in a non-violent manner. For example, governments and protesters often try to avoid direct forms of violence in the course of conflict and confrontation. But these non-violent forms of protest often end up becoming violent again. We can see this evolution from non-violence to violence in many parts of Asia today. The author keeps emphasizing myths about violence. But I wonder if a more in-depth analysis of this shift from nonviolence to violence should also be considered? The war that the author discusses in the paper is the outcome, but I think we should also think about the non-violent sources of war.

[ming-cui]

In violent situations, people have fear. Violence and fear seem to be entangled and reciprocal. People are fearful of losing freedom and safety. So In the US, gun violence is a social problem. It would be informative if we can study public opinion on gun control and its changes over time via computational content analysis. Is it possible?

[jacyanthis]

When do you think machine learning for images and video will be reliable enough that sociology projects like Collins (2009) can do meaningful, large-scale analysis with it? E.g. sort through many hours of video and images to find correlations between emotions, tensions, and violent actions.

[RobertoBarrosoLuque]

I was left a little confused about the analogy between sports and violence that the author uses during the introduction to chapter 2. One of the main themes of this chapter is that fighters are in reality fearful, tense and incompetent but that our idea of violence as being perpetrated by competent, brave and skillful individuals is in part due to entertainment such as sports, could we explore this analogy in more detail?

[sabinahartnett]

I was also wondering, similar to Roberto, if it is the 'gamification' of violence and warfare that bridges sports and violence (similar to what we've seen happen on a larger political scale, of political hobbyism and the interest many people have taken in politics in recent years which parallels sports team support).... and how this might dehumanize /underplay the severeness of violence? How might this relate to media practices of using 'hooking' images and how these selected images influence historical understanding of events?

[lilygrier]

In light of media dramatization/sensationalism and how violence is portrayed as being longer and more dramatic than it generally is, is it possible to analyze the representativeness of media images of real events? One way I could imagine doing this is surveying people who were actually present about whether certain images are representative, but you run into issues of people not being able to clearly remember the possibly-traumatic event. In any event, this phenomenon raises interesting questions about what "ground truth" really is.

[xxicheng]

A similar question as @jacyanthis , what do you think the potential of applying computer vision methods in social science research?

[jinfei1125]

This book about violence demonstrates a lot of images related to violence, such as fear in faces and body gestures, as well as tensions. Also, this week's fundamental reading mentions that computer vision is a task that is easy for humans but hard for computers. So I wonder if collecting and understanding the data will help us understand people's micro-expression or gestures? like @Raychanan mentioned.

[yushiouwillylin]

I am a bit concerned with the way the author view micro-sociology in violence or other contexts. At first, it is of course easy to say that photo, video or audio record is static and provide little situational information. However, this view can really lead to a slippery slope argument.

Even if we have high resolution video that can document incidents, we can always argue there are some hormone in that time and space that influence the incidents under investigation. Where do you think we should draw the line of what is "credible" and what "miss something"?

[k-partha]

It would be interesting to understand how the meaning of 'violence' evolves as we transition into the digital age - as we transition into virtual spaces from physical ones. Do you think online public spaces provide a better environment in which violence can be studied/understood, given that this violence online is recordable, viewable by all, and can represent ground truth better than documentations of physical violence can?

[romanticmonkey]

Same as @Willy624, I also worry if the photos and videos are situational in nature. Thus I came to wonder: Is it possible, or perhaps ethical, to study individuals with violence-related predispositions through their social media and entertainment preferences? We can observe what information they take in (e.g. Instagram likes, YouTube view history) and what actions they let out (e.g. Twitter posts, Facebook events). We might find some kind of group structure to aid the study of overall sociology of violence.

[MOTOKU666]

Adding to @Raychanan 's point, Indeed, a more in-depth analysis of this shift from nonviolence to violence should also be considered. But how many extra resources are needed to predict a potential violent source? In the meantime, would the prediction generate racial and ethical bias in the training? Without a remarkable sign of violence, what other dimensions can we dig into?

[Rui-echo-Pan]

It's interesting to see the discussion about violence. I am wondering how the machine learning or deep learning methods can be applied to detect the context or the meaning of violence. As there may be different causes and cultural meanings in different countries/cultures, which also change along the time.

[william-wei-zhu]

There is a difference between "violence as an act" and "violence as an intention". The former is observable (physical actions), while the latter is latent (the intention to attack or dominate others against their will). How do machine learning techniques detect and differentiate between these two forms of violence?

[hesongrun]

Thanks for the awesome reading! My major takeaway from this paper is that texts only reflects certain facet or dimension of our reality. There can be bias or distortions depending on the producers of the text. As the author argued:

when participants talk about violent situations, they tend to give a very truncated, and by their own lights, idealized version of what went on. (p.4)

This motivates us to look for multi-modal representation of the world, i.e. speech and vision. However, these media can also have their own bias or misperception of the world. My question is how can we optimally combine views from different modality to best uncover the objective information we are looking for?

[Bin-ary-Li]

If so many varieties of violence can be summarized in Collins' compact theory, is there a way to formalize the theory to make verifiable predictions?

[egemenpamukcu]

Does having a clear definition and well-defined categories of violence grounded on soicological theory, make training machine learning models to detect violence possible? How helpful is it (for both supervised and unsupervised learning)?

[theoevans1]

This reading discusses a distinction between the “little stagings” of entertainment or conversation and the “direct evidence” available from some photographs. In thinking about pictures that might be analyzed from the internet, it seems like many of them might be closer to the former category—pictures performatively staged for social media, for instance. How should we think about using that kind of data? Are there online sources that can provide the direct evidence Collins is suggesting?

[zshibing1]

What purpose does identifying violence using computational methods in images, art or video serve? To predict? prevent? for surveillance?

[jcvotava]

What are the benefits vs. risks of relying on stored/archival imagery or film in projects like these, which aim to comment on violence in general? How does machine learning fit in?

[chiayunc]

My question is related to the inherent selection bias of wanting to record something or not. Essentially, some type of event will be more recording-worthy and this might cause problems; we don't and can't have all the images we want when we want it. Unless we are in some sort of Black Mirror: the entire history of you, type of world.

[toecn]

I'm worried about the magnitude of computational resources that would be needed to scale this up for computational social science research, should we be worried about this? Is it worth it?

[mingtao-gao]

This is a really interesting reading. The author mentioned in the article that “The appropriate relationship of micro- and macro-sociology is not to reduce one to the other, but to coordinate the two levels of analysis where it leads to some useful result.” In cases similar to violence, context interacts with how things will develop over time, how can we build more accurate model to capture the 'randomness' in social science fields?