Postdoc at ETHz - Ass. Editor @HSScomms Exploring Opinion dynamics and Collective intelligence with social simulations and belief networks.
Zurich, Switzerland
Joined January 2020
- Tweets953
- Following360
- Followers777
- Likes891
Can we become polarized even while moving toward each otherβs views?
Surprisingly, yes!
Our experiments found attraction even between political opponents! Yet a model that closely mimics the measured dynamics still produced polarization.
jasss.org/26/3/2.html
Data-oriented validation in social simulations is hard, underdeveloped, and often leads to disappointing results (including discovering that my own model performs terribly!).
But that's no excuse to avoid it.
1/3
If we want to build more reliable models, we need to confront these challenges head-on, understand what goes wrong, and progressively develop better models and better validation methods.
2/3
In these two papers, we take some steps toward developing and testing a more systematic approach to validating opinion dynamics models, uncovering quite a few hidden problems along the way.
π link.springer.com/chapter/10β¦
πresearchgate.net/publicationβ¦
3/3
This video was quite hard to make... Itβs about an almost unknown problem in the social sciences with pretty heavy implications.
I hope this video can bring more attention to the problem!
youtu.be/J3UIaYyd-Lw
The nice thing about experiments is that people can surprise you.
The annoying thing is that they will surprise you.
From 2000 to 2020, responses across political issues increasingly lined up along a leftβright divide. Democrats and Republicans became less likely to share the same positions, forming more coherent belief systems.
Blue nodes lean Democratic; red nodes lean Republican. Links show positive associations across issues.
Original analysis from the following article πππ
cambridge.org/core/journals/β¦
People who feel optimistic about AI agree with one another, but people who worry about it... not so much.
A belief network of six questions about AI and technology in the 2024 U.S. General Social Survey. 𧡠1/7
Made with the ResIN R package:
cran.r-project.org/package=Rβ¦
Tutorial:
osf.io/7jvqc/
Code:
github.com/just-a-normal-dinβ¦
6/7
Dataset:
thearda.com/data-archive?fidβ¦
A previous study found a similar pattern for vaccination:
nature.com/articles/s41598-0β¦
#ResIN #NetworkAnalysis #SurveyResearch #RStats #ArtificialIntelligence 7/7
How should we represent a belief system?
As a position on a scale? A collection of survey answers? A set of statements? A network?
In our chapter, we introduce Belief Network Theory: a measurement framework for studying how beliefs are held, expressed, perceived, and...
...connected to social identities.
We developed the framework to preserve the complexity of belief systems while making them measurable across surveys, natural language, experiments, and simulations. Our aim is to help researchers study not only what people believe, but also...
... how beliefs cluster, evolve, bridge groups, and become part of collective identities.
πππ
researchgate.net/publicationβ¦
Today at 15:15 CET I'll be speaking at the Social Simulation Fest about Science Communication and Narratisation! π€
It's an online event, so you can still join by registering here: π
socsimfest.eu/registration/
Opinion dynamics models do _not_ need to be an abstraction of reality. You can also build them to reproduce measured human behaviour in experiments.
jasss.org/25/4/4.html