Category Archives: epistemic networks

people are not points in space

This is the video of a lecture that I gave at the Institute H21 symposium in Prague last September. The symposium was entitled Democracy in the 21st Century: Challenges for an Open Society, and my talk was: “People Are Not Points in Space: Opinions and Discussions as Networks of Ideas.” I’m grateful for the opportunity to present and for the ideas of other participants and organizers.

My main point was that academic research currently disparages the reasoning potential of ordinary people, and this skepticism discourages efforts to protect and enhance democratic institutions. I think the low estimate of people’s capacity is a bias that is reinforced by prevalent statistical methods, and I endorse an alternative methodology.

See also:  individuals in cultures: the concept of an idiodictuon; Analyzing Political Opinions and Discussions as Networks of Ideas; a method for analyzing organizations

can AI help governments and corporations identify political opponents?

In “Large Language Model Soft Ideologization via AI-Self-Consciousness,” Xiaotian Zhou, Qian Wang, Xiaofeng Wang, Haixu Tang, and Xiaozhong Liu use ChatGPT to identify the signature of “three distinct and influential ideologies: “’Trumplism’ (entwined with US politics), ‘BLM (Black Lives Matter)’ (a prominent social movement), and ‘China-US harmonious co-existence is of great significance’ (propaganda from the Chinese Communist Party).” They unpack each of these ideologies as a connected network of thousands of specific topics, each one having a positive or negative valence. For instance, someone who endorses the Chinese government’s line may mention US-China relationships and the Nixon-Mao summit as a pair of linked positive ideas.

The authors raise the concern that this method would be a cheap way to predict the ideological leanings of millions of individuals, whether or not they choose to express their core ideas. A government or company that wanted to keep an eye on potential opponents wouldn’t have to search social media for explicit references to their issues of concern. It could infer an oppositional stance from the pattern of topics that the individuals choose to mention.

I saw this article because the authors cite my piece, “Mapping ideologies as networks of ideas,” Journal of Political Ideologies (2022): 1-28. (Google Scholar notified me of the reference.) Along with many others, I am developing methods for analyzing people’s political views as belief-networks.

I have a benign motivation: I take seriously how people explicitly articulate and connect their own ideas and seek to reveal the highly heterogeneous ways that we reason. I am critical of methods that reduce people’s views to widely shared, unconscious psychological factors.

However, I can see that a similar method could be exploited to identify individuals as targets for surveillance and discrimination. Whereas I am interested in the whole of an individual’s stated belief-network, a powerful government or company might use the same data to infer whether a person would endorse an idea that it finds threatening, such as support for unions or affinity for a foreign country. If the individual chose to keep that particular idea private, the company or government could still infer it and take punitive action.

I’m pretty confident that my technical acumen is so limited that I will never contribute to effective monitoring. If I have anything to contribute, it’s in the domain of political theory. But this is something–yet another thing–to worry about.

See also: Mapping Ideologies as Networks of Ideas (talk); Mapping Ideologies as Networks of Ideas (paper); what if people’s political opinions are very heterogeneous?; how intuitions relate to reasons: a social approach; the difference between human and artificial intelligence: relationships

Hypothetical network of a small group

a method for analyzing organizations

I’m about to conduct a study in partnership with a civic association in the midwestern United States. It should yield insights that can inform this association’s plans and help me to develop a method and related theory. I have IRB approval to proceed, using instruments that are designed.

In the meantime, a colleague alerted me to an impressive new paper by Dalege, Galesic and Olsson (2023) that uses a very similar model. Fig. 1 in their paper resembles the image I’ve created with this post. These authors make an analogy to physics that allows them to write about spin, energy and temperature. I don’t have the necessary background to replicate their analysis but will contribute relevant empirical data from a real-world group and some additional interpretive concepts.

We will ask members of the association to what extent they agree with a list of relevant beliefs (derived from their own suggestions in an open-ended survey). We will ask them whether each belief that the individual endorses is a reason for their other beliefs. As a hypothetical example, you might think that the organization’s youth programming is important because you believe in investing in young people. That reflects a link between your two beliefs. We will also ask members to name their fellow members who most influence them.

In the hypothetical image with this post, the circles represent people: members of the group. A link between any two members indicates that one or both have identified the other as an influence. That is a social network graph.

The small shapes (stars, circles, etc.) represent the beliefs that individuals most strongly endorse. The arrows between pairs of beliefs indicate that one belief is a reason for another. This is a belief-network.

Reciprocal links are possible in both the social network and the belief networks.

Before analyzing the network data, I will also be able to derive some statistics that are not directly observed. For example, each node in both the social network and the belief networks has a certain amount of centrality, which can be measured in various standard ways. I can also run factor analysis on the responses about beliefs to see whether they reflect larger “constructs.” (Again, as a hypothetical example, it might turn out that several specific responses are consistent with an underlying concern for youth, and that construct could be measured for each member.)

I plan to test several hypotheses about this organization. These hypotheses are not meant to be generalizations. On the contrary, I expect that for any given organization, most of the hypotheses will turn out to be false. The purpose of testing them is to provide a description of the specific group that is useful for diagnosis and planning. Over time, it may also be possible to see which of these phenomena are most common under various circumstances.

Hypotheses to test

H1: The group is unified

H1a: The group is socially unified to the extent that its members belong to one network connected by interpersonal influences. The denser the ties within the connected network, the more the group displays social unity.

H1b: The group is epistemically epistemically unified to the extent that members endorse the same beliefs, and to the extent that these shared beliefs are central in their belief networks.

H2: The group is polarized.

H2a: The group is socially polarized if many members belong to two separate subgroups that are connected by interpersonal influences but are not connected to each other, as depicted by the red and blue clusters in my hypothetical image.

H2b: The group is epistemically polarized if many members endorse belief A, and many other members endorse B, but very few or no members endorse both A and B. If A and/or B also have high network centrality for the people who endorse them, that makes the epistemic polarization more serious. (Instead of examining specific beliefs, I could also look at constructs derived from factor analysis.)

H3: The group is fragmented

H3a: The group is socially fragmented if many members are connected by influence-links to zero or just one other member.

H3b: The group is epistemically fragmented if no specific beliefs are widely shared by the members.

H4: The group is homophilous if individuals who are connected by influence-ties are more likely to endorse the same beliefs, or have the same central beliefs, or reflect the same constructs, compared to those who are not connected. If the opposite is true–if socially connected people disagree more than the whole group does–then the group is heterophilous.

H5: There is a core and a periphery

H5a: There is a social core if some members are linked in a relatively large social network, while most other members are socially fragmented.

H5b: There is an epistemic core if many (but not all) members endorse a given belief, or a given belief is central for them, or they share the same constructs, while the rest of the organization does not endorse that belief.

H6: Certain members are bridges

H6a: A person is a social bridge if the whole group would be socially polarized without that person.

H6b: A person is an epistemic bridge if the whole group would be epistemically polarized without that person.

H7: Members tend to hold organized views: This is true if the mean density of individuals’ belief networks (the mean number of links/nodes) is high, indicating that people see a lot of logical connections among the things they believe.

Our survey respondents will answer demographic questions, so we will be able to tell whether polarized subgroups or core groups have similar demographic characteristics. Hypothetically, for example, a group could polarize epistemically or socially along gender lines. And we will ask general evaluative questions, such as whether an individual feels valued in the association, which will allow us to see whether phenomena like social- connectedness or agreement with others are related to satisfaction.

What to do with these results?

Although the practical implications of these results would depend on the organization’s goals and mission, I would generally expect polarization, fragmentation, the existence of cores, and homophily to be problematic. These variables may also intersect, so that an organizations that is socially polarized, epistemically polarized, homophilous, and reflects highly organized views is especially at risk of conflict. A group that is fragmented and reflects disorganized belief-networks at the individual level may face a different kind of risk, which I would informally label “entropy.”

Being unified can be advantageous, unless it reflects group-think or social exclusivity that will prevent the organization from growing.

Once an organization knows its specific challenges, it can use appropriate programming to make progress. For instance, if the group is socially fragmented, maybe it needs more social opportunities. If it is polarized, maybe a well-chosen discussion could help produce more bridges. If it displays entropy, maybe it needs a formal strategic plan.

I would generally anticipate that bridges are helpful and should be supported and encouraged. In our study, all the data will be anonymous, so our partner will not know the identity of any people who bridge gaps. But a different application of this method could reveal that information.

Although I am focused on this study now, I remain open to partnerships with other organizations so that I can continue this research agenda. Let me know if you lead an organization that would like to do a similar study a bit later on.

Reference: Dalege, J., Galesic, M., & Olsson, H. (2023, April 12). Networks of Beliefs: An Integrative Theory of Individual- and Social-Level Belief Dynamics. https://doi.org/10.31219/osf.io/368jz. See also: Analyzing Political Opinions and Discussions as Networks of Ideas; Mapping Ideologies as Networks of Ideas; seeking a religious congregation for a research study

when does a narrower range of opinions reflect learning?

John Stuart Mill’s On Liberty is the classic argument that all views should be freely expressed–by people who sincerely hold them–because unfettered debate contributes to public reasoning and learning. For Mill, controversy is good. However, he acknowledges a complication:

The cessation, on one question after another, of serious controversy, is one of the necessary incidents of the consolidation of opinion; a consolidation as salutary in the case of true opinions, as it is dangerous and noxious when the opinions are erroneous (Mill 1859/2011, 81)

In other words, as people reason together, they may discard or marginalize some views, leaving a narrower range to be considered. Whether such narrowing is desirable depends on whether the range of views that remains is (to quote Mill) “true.” His invocation of truth–as opposed to the procedural value of free speech–creates some complications for Mill’s philosophical position. But the challenge he poses is highly relevant to our current debates about speech in academia.

I think one influential view is that discussion is mostly the expression of beliefs or opinions, and more of that is better. When the range of opinions in a particular context becomes narrow, this can indicate a lack of freedom and diversity. For instance, the liberal/progressive tilt in some reaches of academia might represent a lack of viewpoint diversity.

A different prevalent view is that inquiry is meant to resolve issues, and therefore, the existence of multiple opinions about the same topic indicates a deficit. It means that an intellectual problem has not yet been resolved. To be sure, the pursuit of knowledge is permanent–disagreement is always to be expected–but we should generally celebrate when any given thesis achieves consensus.

Relatedly, some people see college as something like a debate club or editorial page, in which the main activity is expressing diverse opinions. Others see it as more like a laboratory, which is mainly a place for applying rigorous methods to get answers. (Of course, it could be a bit of both, or something entirely different.)

In 2015, we organized simultaneous student discussions of the same issue–the causes of health disparities–at Kansas State University and Tufts University. The results are here. At Kansas State, students discussed–and disagreed about–whether structural issues like race and class and/or personal behavioral choices explain health disparities. At Tufts, students quickly rejected the behavioral explanations and spent their time on the structural ones. Our graphic representation of the discussions shows a broader conversation at K-State and what Mill would call a “consolidated” one at Tufts.

A complication is that Tufts students happened to hear a professional lecture about the structural causes of health disparities before they discussed the issue, and we didn’t mirror that experience at K-State. Some Tufts students explicitly cited this lecture when rejecting individual/behavioral explanations of health disparities in their discussion.

Here are two competing reactions to this experiment.

First, Kansas State students demonstrated more ideological diversity and had a better conversation than the one at Tufts because it was broader. They also explicitly considered a claim that is prominently made in public–that individuals are responsible for their own poor health. Debating that thesis would prepare them for public engagement, regardless of where they stand on the issue. The Tufts conversation, on the other hand, was constrained, possibly due to the excessive influence of professors who hold contentious views of their own. The Tufts classroom was in a “bubble.”

Alternatively, the Tufts students happened to have a better opportunity to learn than their K-State peers because they heard an expert share the current state of research, and they chose to reject certain views as erroneous. It’s not that they were better citizens or that they know more (in general) than their counterparts at KSU, but simply that their discussion of this topic was better informed. Insofar as the lecture on public health found a receptive audience in the Tufts classroom, it was because these students had previously absorbed valid lessons about structural inequality from other sources.

I am not sure how to adjudicate these interpretations without independently evaluating the thesis that health disparities are caused by structural factors. If that thesis is true, then the narrowing reflected at Tufts is “salutary.” If it is false, then the narrowing is “dangerous and noxious.”

I don’t think it’s satisfactory to say that we can never tell, because then we can never believe that anything is true. But it can be hard to be sure …

See also: modeling a political discussion; “Analyzing Political Opinions and Discussions as Networks of Ideas“; right and left on campus today; academic freedom for individuals and for groups; marginalizing odious views: a strategy; vaccination, masking, political polarization, and the authority of science etc.

Analyzing Political Opinions and Discussions as Networks of Ideas

This is a talk that I have prepared for the Universidad Carlo III in Madrid tomorrow. It is a summary of recent work that I have been conducting with colleagues at Northeastern, Wisconsin, and Oxford and that I’m beginning to develop into a book manuscript.

In the model that I present, an individual holds potentially connected beliefs about political or moral issues, which we can represent with nodes and links (an “idiodictuon”). Whether and how the various ideas are linked in the person’s network influences that individual’s actions and opinions. When people discuss political or ethical issues, they share portions of their respective networks of which they are conscious at the time and may bring ideas from their interlocutors into their own idiodictuons.

Some network structures are better than others for discussion: overly centralized or scattered networks are problematic. Individuals tend to demonstrate similar network structures on different issues, so that having a proclivity for a certain form of network is a character trait.

People, with their respective networks of ideas, are also embedded in social networks. An idea is more likely to spread depending on features of both the social network and the idea networks of the people who interact. Specifically, the odds that an idea will spread from a given person depend on how many people receive communications from that person and how much they trust the communicator. It is reasonable to take into account the trustworthiness of a source when assessing an idea.

As a whole, a population may develop a shared network structure. An idea that is widely shared and frequently central in individuals’ networks becomes a norm. Such norms play important roles in institutions. A community or a culture is a single network or phylodictuon that encompasses disagreement. Ultimately, all such networks interconnect to form a network of human ideas.