UCLA

Machines curate our news narratives to weaponize our minds against us, making commonsense politics impossible because we no longer share a common baseline of facts. Many helplessly accept this alienating and accelerating reality as the new normal. However, technology guided by humanistic principles and cognitive science may restore our vitality. My dissertation integrates psychology experiments on individuals and networked-groups, computational cognitive modeling, large-scale analyses of social media, open-source software development, and citizen science principles to illustrate how online networks and generative artificial intelligence can promote healthier attitudes towards established evidence and each other.

In Chapter 2, I analyze networked behavior and narrative agency across a series of online social network experiments (total N = 660, 13,200 interactions). Through a hashtag generation game I manipulated communication patterns within a group to encourage shared or polarized beliefs. Entropy dynamics of generated hashtags and language data revealed that belief and behavioral coherence vary according to neighborhood topology at both local and global levels. Rewards for aligning hashtags also shifted participants' causal language use when writing personal narratives about a disaster event. Given these findings, Chapter 3 introduces a computational framework rooted in Bayesian decision theory to disentangle how rewards (e.g., engagement metrics embedded in social media) ought to influence our beliefs about evidence, and describes how the framework can guide interventions on networked-groups.

For better and worse, people's beliefs are sensitive to online interactions. Chapter 4 integrates large-scale analyses of internet discourse (100,000 interactions), belief-updating experiments (total N = 2,676), and interactive data visualizations to (1) identify features of persuasion in naturalistic online interactions, and (2) extend those features into a crowdsourced data narrative that countered misconceptions about structural racism in a random sample of Americans (Cohen’s d = .4). Chapter 5 describes a state-of-the-art Large Language Model system that models causal beliefs from natural language data, which I applied in previous chapters.

The first and final chapters expand on my vision for the future of cognitive and behavioral science, and sketch a trajectory for human-oriented technology development and academic achievement. As practicing scientists we must systemically reorient our goals if we are to conquer the networks that radicalize and atomize society.