Open GabeNicholson opened 1 year ago
Dear Professor Cikara, thank you for sharing your work on how latent social groups are learned based on the behavior of other agents rather than explicit social categories or dyadic similarities with oneself.
An alternative explanation to your results could be that participants prefer to associate with the majority, perhaps due to an innate need to feel belonging or affiliation. For example, in Experiment 1, participants may have been more likely to choose Agent B’s mystery trial choice when Agent C formed a latent group with the participant and Agent B, compared with when Agent C did not form such a group because B and C outnumber Agent A by 2:1. Or in Experiment 4, participants may have been more likely to choose Agent B’s choice when Agent C formed a latent group with the participant and Agent B compared with when Agent C formed no such group, again because of the latent group of B and C being larger than the single Agent A.
Whereas social categories and dyadic can operate independently of group size, it seems to me that your proposed mechanism is (at least in the experiments) inextricably to group size, or rather that there is only one latent group. Maybe this is what is meant by you noting that “our results suggest that people take the relationships among agents into account when building representations of social groups” in the general discussion section. Nonetheless, I would be curious to hear how you think about larger settings where the latent group formed by Agents B and C is not contrasted with single Agent A, but other latent groups. This could potentially help disentangle the effects of group existence and group size from your exploration of how the relationship between a participant and an agent may affect the perception of a third agent.
Dear Professor Cikara, Thank you for sharing your research with us!
After reading Discovering Social Groups via Latent Structure Learning, I found the concept of coalition formation in relation to perception and bias on political values very interesting. I am not familiar with the term latent structure learning, but I am curious about how would you consider the phenomena of stratification and the formation of elite (political elite) using the framework of latent structure learning, as these two topics are also embedded with intergroup dynamics and social categorization. Do you think using similar approach from the four experiments you have conducted, is it possible to identify boundaries of stratified groups?
Looking forward meeting you on Thursday's presentation!
Hello Professor Cikara,
Having done work with AMT myself for previous survey studies, I've been personally frustrated by the quality of the data that one can get from it, even when including attention checks, minimizing time to respond, maximizing time to respond etc. There seems to be a possible issue with bots or at least semi-automated systems on the platform.
I was hoping you might speak a bit beyond the details listed in the paper about how you and your team approached using AMT for data collection, what strategies you used to limit the ability of bots to participate, payment levels, limitations on geographic location and any other parts of that process that you think were important for creating the dataset. I ask this not as a doubting of data quality, but simply that the information on the how's and why's of proper use of AMT for surveys and experimentation in computational social science still seems to be quite unformalized and hard to find.
Thank you
Prof. Cikara, thank you for sharing your expertise. In the discussion of your paper, you note that "of course, which category is salient—race, gender, profession—is highly context dependent". I would imagine that this context is often temporal as well - both at the macro and micro level. Do you have any insight into how the ways that people have formed ingroups and outgroups might change over the course of history/into the future? Additionally, at the micro level, do you anticipate that the ways that we form ingroups and outgroups might change over the course of our lifetimes? For example, do children form groups differently, then say, teenagers, young adults, etc..? Or, are the underlying ways that we form ingroup and outgroup consistent (taking into account matching to categories, mental state, inferences, self-reference, etc.)?
Thanks for sharing your work with us, professor! I was wondering about two things as I read your paper:
Thank you for sharing your work with us, Professor Cikara! It's very promising to see how latent structure can be learned and used to discover social groups. An extension of the latent structure I can think of is the Markov decision process and hidden Markov model. Do you think these concepts can be applied to discover social groups further if the experiments are in a series?
Thank you Professor Cikara for sharing your work! I think group formation is an interesting topic. In American politics, it serves as a "perpetual screen" (Campbell 1981) and mask people's judgement when it comes to partisan actions and voting according to party lines. I am wondering if you have connected your work with any of the works in voting, campaigning and elections, would love to see that!
Prof. Cikara, I appreciate your sharing your knowledge. During the presentation of your paper, you mentioned that the prominence of different categorizations, such as race, gender, or profession, varies based on the situation. I was wondering if you have any thoughts on how the formation of in-groups and out-groups might evolve over time, both in the larger historical context and within individuals over the course of their lives.
Dear Dr Cikara
I am so appreciative for your presentation. I find that this experiment is based on the group of crime. I am curious that whether this research can be applied to other environments since the data from crime is special in the world. Therefore, I want to ask how the result of this research can be used in different environments, or we should do more research in this area.
Dear Prof Cikara, Thank you for offering this talk! It is a very intersting perspective for your paper to attempting to code the human process of social grouping into machine learning models. I see that you're using political issues as the metrics to code. However, this metrics is a very debatable one, as political issues are highly connected with the substantive context, continually evolving, and has no causal relationship with social grouping. Therefore, can you tell as how did the peer review of this paper address criticism similar to this? Can you elaborate on how you respond to them? Thanks!
Dear Dr. Cikara,
Thank you for sharing your work with us! I found latent structure learning very interesting, and I was surprised to read about how people identify another person's membership of social groups through the shared preference of the third person (i.e. Agent C in the experiments). I am curious, however, that in real-world social interaction, whether a person (the role of participants in the experiment) would learn from multiple "Agent C" to judge "Agent B" as in/out-group. Also, do you think multiple "Agent C" would help accumulate the information a person learns about the group membership? Or too much information would interfere with the inference?
Thanks for sharing this work, Mina! I'm wondering how the findings in this paper map onto real world social grouping. Is there any work on investigating these dynamics using social media data such at interactions and reinforcing behaviors between individuals on Twitter or Facebook?
Prof. Cikara, Thanks for sharing your work! It is very interesting. I wonder whether interactions between agents in forming social group would change their membership over time. And what are the information exchange between agents determine their membership or maintain the social group? how to determine whether a social group is stable and mature group or an emerging group?
Hello Prof Cikara,
Thank you for sharing your work!
Like many classmates, I wondered how the findings could be applied to social media interactions. Specifically, I was thinking about the "social" aspect of these groups and how (if it matters at all) we can account for that in this paper. The experiments had the benefit of being set in a relatively consequence-free environment (AMT), making such ingroup and outgroup decisions on social media would be different, especially when their choices are often made public to everyone else. For example, a factor like social desirability could be significant in how people evaluate their choice of group.
Hi Prof Cikara, Thank you for your speaking. And I found that the environment of latent structure learning is relatively limited. So I would like to ask if the crime data is replaced with data from other environments, will the conclusion of the paper change?
Hi Prof Cikara, Thank you for sharing your work with us! You mention that your work can possibly be a formal account of how balance theory operates. This is really exciting. But I'd like to hear explanations about it. Could you be more specific? I am also interested in situations that the balance theory considered but not covered in the experiments (correct me if I am wrong). Like "enemy's enemy is my friend", which indicates the influence of negative ties. How would design to test this?
Hello Prof. Cikara,
Thank you for sharing this work! I am curious if there were alternative explanations for your experiments in which subjects changed their attitude toward agent B according to agent C's agreement with themselves and with agent B. How is the latent structure learning account different from a simpler statistical learning account, in which subjects just learned and built a similarity estimation between each two pair of individuals and modulate their similarity rating based on Similarity(Self, C) and Similarity(C, B)? How is the learning of group structure qualitatively different from learning on the individual level? I am very curious to know the results when we test this hypothesis in a scenario with many agents and see how group structure emerges above and beyond the variance that we can explain by just looking at each individual.
Dear Professor Cikara,
In your work, have you come across anything that would explain what happens when a person has formed multiple latent groups within one social setting and the two are at odds with each other? For example, if a person forms an ingroup with a person based on political preference, but then that other person becomes part of their outgroup because they like a different sports team, would those two things "cancel out" the perception of the other, or does a person switch back and forth with their perceptions of the other?
Thanks for coming to present your work.
Sam Corey
Hi Professor Cikara, As a sociology student, I am interested in how you think culture would play a role in your discussion of social categorization. In other words, I wonder if this set of studies is replicated in other countries (e.g., East Asia), would you expect similar or different results/inferences and why? Thank you!
Hi Prof. Cikara, To uncover the underlying social group structures in a consequential social context, there are some methods like social network analysis, clustering algorithms, multidimensional scaling, et al. Compared to those different methods, I am curious what research questions and what data would make latent structure learning model outperforms extremely?
Thank you for presenting your research at our workshop, Prof. Cikara! The paper discusses a number of very interesting arguments and findings. I believe the results have strong implications for many different fields. In particular, I was wondering if you have any thoughts on the implications for marketing?
Hi Professor Cikara,
Thank you for sharing your research with us!
I find Experiment 4 particularly interesting, which shows latent group structures driving choice even when explicit grouping cues are imposed. I am wondering, however, would it be possible that the color label as explicit grouping cues are deemed as unimportant information and thus does not have much impact on participants' choice? In real life, would there be cases where the explicit cues overly dominate choices that latent groups fail to help people infer more accurate group structures?
Thanks and looking forward to your presentation.
Thank you Professor Cikara for sharing your work with us! It is interesting to learn about the computational methods in discovering social groups. I believe there would be many potential applications such as election prediction, for education or to fight against school violence, etc. One part that I didn't catch is how do you design the questions and determine how relevant/useful the questions are in grouping?
Hi Professor Cikara,
Thank you for sharing this paper. Based on latent structure learning, you proposed a quite interesting mechanism of social group boundaries, which I felt quite interesting. My question is right about the mechanism. Specifically, could you please show us how the proposed mechanism can be related to other social relation theories in addition to the balance theory? Could you also provide some examples to further understand the social implications of understanding the mechanism? Thanks!
Hi Professo Cikara, thank you for sharing your research with me! In the Method section, you mentioned that you recruited a small number of participants that was sufficient. Could you please explain this further? Thanks!
Hi Prof. Cikara, Thank you for sharing your research with us. In study 4, you found that minimal group labels are overridden by latent group structures. I could imagine that in real life, there are multiple cues at any given time about different group affiliations of the same person, for example, agent A might be in the same racial group as agent B, different political ideology group, same school alumni group, different gender group, etc. It would be interesting to know when all those factors are put together -- very often not even in an explicit way, how agent A weighs them in different contexts to decide the extent to which agent B is an ingroup member or not.
Hi Prof. Cikara, thanks for sharing your work on social group formation and latent structure learning. I see the main argument of this paper is that people take the relationships among agents into account when building representations of social groups. However, when I look at the demonstration of experiment 1 (Figure 2 and 3), I was curious about how easily can participants identify the group that they are aligned with? Do you anticipate any failure of group identification when making the decisions in experiment 1? Thanks.
Hi Professor Cikara, Thanks for sharing your research with us. I have a question as follows:
Hi Professor Cikara, thanks for sharing your research with us. Since your research focuses on political preference, I wonder how the findings here can imply or can be generalized to say hobby preference, which can be more relevant and more important in some cases. Another situation I am interested in is when people do not really have political preferences or when they have mixed feelings about a group. How would the model compute in those cases?
Hi Professor Cikara, thanks so much for sharing this interesting paper with us. You mentioned that it is critical to understand the principles determining how such groups are inferred, their flexibility and their generality. Could you give an example of this?
Hi Professor Cikara, thank you for sharing your interesting work with us! You mentioned that your findings offer a plausible explanation of other theories of social relations, such as the balance theory. Could you talk more about how your results explain the balance theory? Apart from the balance theory, what other theories of social relations can your findings shed light on? Thanks!
Thank you for sharing your valuable research with us, Professor Cikara. I wonder if you could shed some light on how your findings could be applied within the context of online social media.
Hi Prof.Cikara, thank you for this interesting paper! I'm wondering what are the key factors that influence the accuracy and stability of latent structure learning in social groups, and how do these factors interact with individual differences in personality, cognition, and motivation?
Hi Prof. Cikara. Thanks for sharing your work with us. Since the experiments framed the participants in terms of teams, I'm curious if the participants in the study might have come up with certain beliefs about rewards they might receive if they performed in a team-like manner, which might complicate the motivations or mechanisms of the theory?
Hi Professor. Do you think these latent structures would exist in environments that aren't necessarily polarizing (e.g., like the ones in your experiments where participants had to choose one side)? What would, then, the group formation mechanism for people with moderate opinions look like? Can we answer such a question within the structure of your experiment?
Hello Professor Cikara,
Thank you so much for sharing your work with us. The introduction of your presentation addresses limitations of current conceptualization of social groups (i.e., traditional category labeling). I'm wondering how to conceptualize the new theorization of social coalitions in your paper. The model schematic illustrates a Bayes' rule of structuring groups of agents. Under this framework, will different groups overlap with one another?
Hi Prof. Cikara,
Thank you for the insightful paper! I wanted to get your opinion on whether these latent groups could be constant over different "issues". For example, people may have different latent groups for (1) movie preferences and (2) political ideology. Or would our political grouping have an influence on our movie grouping such that these two groups are similar?
Hi Professor Cikara, Thank you so much for sharing your work. Your study is mainly about social groups. I am very curious how this study could be extrapolated to the online social media platforms since social groups and networks are also established in online communities. Thank you!
Hi Professor Cikara, thank you for sharing with us your research! I find your discussion about social group formation very interesting. I'm wondering if the intersection of different groups will play a role in the formation process? Looking forward to your presentation tomorrow!
Hi Professor Cikara, thank you for sharing your research. Each individual could be represented by several kinds of social groups. Normally, is there an upper limit to the number of social groups that an individual belongs to?
Hi Professor Cikara,
Thank you for sharing this fantastic working with us. It is really interesting to think about the concept of latent data structure in the context of social interactions. The generative probabilistic model and a series of experiments shown in the paper well demonstrate the effectiveness of the proposed approach. Looking forward to your presentation tomorrow!
Dear Prof. Cikara, Thank you for sharing your idea with us! From the research, we could observe people’s political choices might be affected by these social groups that we might not be able to explicitly identify. What I am curious about is whether it is possible for parties of an organization to take action to step into this mechanism and further change people’s decision process.
Dear Professor Cikara, thank you for sharing! I am really interested in learning more about the probabilistic model you mentioned in the paper and the latent structure learning. Looking forward to tomorrow's presentation with more details!
Dear Professor Cikara, thank you for sharing your work with us! I wonder how your paper can be applied in real life. Looking forward to your presentation tomorrow!
Hi Professor Cikara,
The finding about third party C's latent group effect is interesting. What if there is a fourth party D that creates a latent group including the participant and A? In this case, it matters which party of C and D can exert greater influence, i.e., a competing third parties case. Have you considered about this?
Best, Jerry Cheng
Dear Professor Cikara, thank you for your amazing work! I'd like to learn more about the probabilistic model you mentioned in the paper as well as latent structure learning. I'm curious how your paper can be used in practice. I eagerly await your presentation tomorrow!
Thank you for sharing your work with us, Professor Cikara! I was wondering if you think the findings from your work on the latent group can provide insights for digital marketers, for example, when they are trying to decide on the target audience of an ad. I see the great potential of the practical implication that your work could bring!
Hi Professor Cikara, thanks for sharing! This is an interesting topic. You've mentioned testing personality dimensions encompassed by the group structure. Could you please explain a bit more about that? Looking forward to your presentation!
Professor Cikara, thank you so much for sharing your inspiring work! I am interested in whether topics with high-level moral conviction (e.g., abortion, gender issues, gun control...) will amplify the effect of latent groups on our perception as the antagonism against opposing sides in these topics will be much more intense. In addition, would you say that as social categorization is so dynamic, we are very gullible in terms of making political decisions, moral judgments, or choices in general because of our favor to (latent) in-group members?
Hi Professor, thanks for coming to present etc.
My question relates to the recruitment of participants. For each experiment you listed the number of participants that you recruited via AMT and that you screened them to make sure they hadn't participated in your previous experiments. However, you don't describe what social groups these participants are part of. Might this not be an important driver? For example, members of ethnic minority groups might have a different relationship to identifying as part of a group compared to people who aren't ethnic minorities. Similarly, might it not influence he way in which participants relate to each other as a group (even implicitly through learned behaviours)?
Comment below with a well-developed question or comment about the reading for this week's workshop.
If you would really like to ask your question in person, please place two exclamation points before your question to signal that you really want to ask it.
Please post your question by Tuesday of the coming week at 11:59 PM. We will also ask you all to upvote questions that you think were particularly good. There may be prizes for top question-askers.