9. Large Multi-Modal Models (LMMMs) to Incorporate Images, Art & Video -orienting

[lkcao]

Post questions here for this week's oritenting readings:

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.

[yuzhouw313]

It is not counter-intuitive to see Collins' conclusion on the fearful and avoidance nature when one is experiencing confrontational tension, but it is interesting to see the phenomena and statistical evidence he described, especially combat historian Marshall's research and figures. Specifically, based on his interview and observations during WWII, Marshall found that the amount of soldiers who actually fired their guns was around 15%, and greater distance, stronger command hierarchy, group operated (instead of individually operated) weapons can increase the rate of firing. While Collins emphasizes the importance to study micro level evidence, including individual interviews, photographs, recordings, how can we transform and aggregate Marshall's interview documents via LMMM to further validate his findings?

[floriatea]

Considering the discrepancy between the dramatized portrayal of violence in media and the reality of confrontational tension and incompetence in violence, what could be the long-term effects of these portrayals on public perceptions of violence and on individuals' behavior in potentially confrontational situations?

[XiaotongCui]

Fascinating research! In the research, they analyze a lot of pictures of soldiers. But since the research is in 2009, I wonder with nowadays new LLMs or more advanced models, how can we improve the research or explore more interesting topics?

[bucketteOfIvy]

Collins (2009) frames their work as a movement away from theories "constrained by having been based upon statistics assembled after the fact, packaged by the criminal justice system, or upon interviews with convinced prisoners or other participants" (p. 8). This framing emphasizes the subversive potential of Collin's multimedia based framework, I find that really interesting when juxtaposed with many of the traditional fears of multimodal AI, which usually emphasize it's ability to enable overpolicing. Does Collins framework instead provide an avenue for these same LMMM's to be used to subvert traditional policing?

[Audacity88]

Collins persuasively demonstrates that actual violence is “hard” and much more infrequent than we are led to expect by its representations in cultural forms: “all types of violence fit a small number of patterns for circumventing the barrier of tension and fear that rises up whenever people come into antagonistic confrontation.” I wonder if some of the toxicity of the Internet can be explained by seeing it as an environment where violent speech (almost) never leads to actual violence. Hence, perpetrators become used to making threats with no fear of retaliation, and antagonistic confrontations only instill tension and fear in the victims.

[sborislo]

Collins provides a strong argument for violence generally being undesirable, even in contexts where it seems to be encouraged. I think the texts he provides offer diverse and visceral evidence for people having other motivations which only seem like they include a desire for violence. However, he constantly compares his findings against a baseline expectation that he does not provide strong evidence for. Although people in media content will show a desire for violence (and violence that is abnormal, being long-winded and often evenly-matched), I don’t believe people actually think real-life violence is viewed in this way. Why did Collins not provide nearly as much evidence for his claim that our view on violence is wrong as he did for explaining real-life violence? Is this not necessary in reports like these?

[h-karyn]

Multimodality analysis entails that we are examining multiple forms of data. One unfortunate yet frequent occurring issue is the inconsistency in findings.

This inconsistency can come from sources (e.g., selection bias), or just from the fact that different modalities carry different information which in reality is inconsistent.

How do you handle the case where different modalities yield different results?

[donatellafelice]

Collins discusses the micro-level processes underlying violent confrontations, emphasizing the role of situational factors and emotional cues in escalating or de-escalating violence. I was curious about how our network analysis from last week could work for mapping the types of fearful responses he described and if we have reached a point where computational processes could really simulate fearful reactions in a meaningful way?

[runlinw0525]

Given Collins' emphasis on direct observation as an understanding of what goes on in the dynamics of violent encounters, how can newest large language models (LLMs) help to code and analyze video recordings of such violent encounters to reveal trends in the data that will inform us in better detail about the micro-sociological processes at work?

[naivetoad]

What are the methodological challenges in studying the micro-dynamics of violence, particularly in capturing and analyzing face-to-face interactions that may be fleeting or occur in private?

[michplunkett]

I don't have any particular question about this work, but I did very much enjoy the author's take on violence and how people actually react to it, as opposed to how people often say they will. His descriptions of war and how soldiers react within it brought to memory tales like All Quiet on the Western Front and stories I read of people who ended up being "the good guy with a gun," and how their resulting experience felt far more traumatic than it did heroic.

[volt-1]

Collins' research reveals the inherent fear and avoidance people exhibit in confrontational situations. Building on this finding, could LMMMs be employed to analyze violent speech and threats on social media platforms, to understand how such "consequence-free" violent threats might escalate in environments devoid of actual violent risk?

[ana-yurt]

It is interesting to read "Violent Confrontations" in light of the recent release of SORA. I am curious what would be the text-to-video models' interpretation of gunfights and fistfights. Do generative video models augment the distortion of violence in cultural forms? Is it even ethical to generate representations of violence?

[alejandrosarria0296]

In 'The Micro-sociology of Violent Confrontations,' Collins explores the psychological barriers to violence in high-tension scenarios, challenging the notion of a natural inclination towards violence. How does his analysis inform strategies for conflict resolution and violence prevention? Additionally, what implications does this have for understanding the role of emotional dynamics in deterring violent behavior?

[QIXIN-ACT]

This publication dates back to 2009. Given the significant advancements in large language models (LLMs) and, notably, in the generation of images and videos, are there any new challenges or research methodologies that could be applied in this field?

[Vindmn1234]

Because the landscape of interaction and confrontation has significantly evolved with the rise of digital platforms, then how do you see the role of digital and social media platforms in shaping the confrontational tension and dynamics of violence today, compared to when 'Violence: A Micro-sociological Theory' was published?

[ethanjkoz]

The author articulates that much of what we see as violence is actually quite performative. The popular imagination of violence is quite off the mark and real violence is infrequent. I am curious to see how we could incorporate a multimodal approach along with the theoretical implications laid out in this work to discuss similar acts of violence like use hateful symbols in hatecrimes or the like.

[Twilight233333]

The author persuasively shows that actual violence is "hard," but I wonder if we can identify how specific cultures, contexts, and situations influence violent behavior. Do they influence violence in the same way? How can we measure these effects by a natural controlled trial?

[chenyt16]

The emphasis on analyzing situational dynamics is compelling. One concern is that they rely predominantly on data such as camera shots and audio recordings, which primarily capture situational dynamics, to validate the argument that situational dynamics are crucial.

[Caojie2001]

The article is really interesting as it approaches the traditional topic of violence from a rather novel perspective. I wonder whether the theories drawn from this research could be utilized for practical purpose, for example, preventing violence behavior before it actually takes place.

[yueqil2]

The article serves as a magnifying glass through which the layered textures of tension and fear can be seen. How will a computer read as much as this article just from static images? Can it totally achieve this task by large model training?

[HamsterradYC]

Reflecting on Randall Collins' insights from "Violence: A Micro-sociological Theory," which emphasizes the significance of situational dynamics in understanding violence, I'm curious about how these concepts translate into the digital realm. How does the anonymity provided by online platforms impact these dynamics? When individuals can conceal their identities on the internet, does this anonymity increase their propensity for violent behavior compared to direct, face-to-face interactions?

[erikaz1]

Dr. Randall Collins presents countless fascinating ideas on micro-evidence and micro-sociological theories of violence. I wonder to what degree Dr. Collin’s core argument, “we live in an era in which our ability to see what happens in real-life situations is far greater than ever before” still holds. If societal behaviors are situationally constructed, then doesn’t this ephemerality, combined with a conditioned understanding of our immediate environment, potentially lead to more unchecked fear and hostile emotion?

[Carolineyx]

The author suggests that humans have evolved with a heightened sensitivity to social cues, leading to a mutual focus of attention and emotional resonance with others. This biological propensity results in the creation of interaction rituals and the maintenance of face-to-face solidarity. These tendencies go beyond acknowledging humans' cognitive abilities, indicating a deep-seated attunement to each other's emotions and susceptibility to social dynamics. It's interesting to see micro-sociology discussing social systems and individual-level interactions, and then delving into the biological level. It would be intriguing to explore how this differs from and overlaps with macro and micro-level psychological analyses, which typically focus more on individual emotions and cognition. I feel that the theories or phenomena mentioned here are closely related to the 'shared reality' theory in social psychology.

[joylin0209]

Collins' research suggests that violence is difficult and infrequent, contrary to its dramatized media portrayal. How might we design a study to explore whether exposure to media violence affects people's likelihood to engage in actual physical confrontations?

[Brian-W00]

How can we understand the difference in how people feeling scared or brave decide to fight or not, even when they do not want to do violence?

[Dededon]

Randal Collins' insight is important. I wonder how can we combine his insights of visual sociology with current state-of-art CV toolkits, like Google vision API and multimodal models?

[anzhichen1999]

How do we measure violence when LLM has not been put into application?

[JessicaCaishanghai]

This article is quite interesting. How it might be influenced by the inherent bias of different culture and coordinate them instead of isolating them?

[cty20010831]

I have a question regarding the multimodality analysis. Specifically, what methodological approaches are used in multimodality analysis? In addition, how do researchers choose the most appropriate method(s) for their specific study?

[beilrz]

The finding of this book is significant. I was wondering can the finding and the theory produced by this book be further examined and expand through computational tools, such as LLM or deep learning frameworks?

[YucanLei]

One thing that I believe Collins is not taking into account enough is the violence on Internet. There is much toxicity on internet and that I do believe this could easily contribute to violence in real life.

[icarlous]

Could we use geospatial analysis to innovatively chart the fear dynamics detailed by Collins in violent encounters, and are current computational methods adept at accurately simulating these emotional reactions? It can be interesting to combine spatial modeling and multimodel information processing