Hannah Overbye-Thompson

Can humans detect algorithmic bias? Not really 🧪🧵 In my new paper with Kristy Hamilton & @jacobtfisher.online even when an algorithm approved women for loans at half the rate of men, people did not do a good job of detecting it. 🔗 authors.elsevier.com/a/1ny4s2f%7E...

Following Marr’s levels of analysis, we asked how people detect bias using DDM (for a primer see: www.hannahoverbye.com/teaching/dri...). Modeling using DDM allows us to locate where in the decision process oversight may break down rather than only assessing final decisions. 🧵2/3

I've noticed w/research it's often hard to find validated measures of constructs without hobbling together scales from multiple papers; now when possible I try to contribute by validating scales. Below is a scale that I hope is helpful that measures the perceptual attributes of DOI + reinvention 🧪

Fabulous study by @felix-dietrich.de @aliciaernst.bsky.social @rkreling.bsky.social et al., examining how algorithmic curation affects music streaming UX. Key finding: More algorithmic recommendations = less enjoyment, BUT listening sessions w/algorithmic curation were perceived as more novel 🧪

New study (2025) examines if people can detect bias in AI training data. Key finding: Training data cues were largely ineffective; users relied on AI performance instead to judge bias + consistent with prior work on AI bias, the majority of participants failed to notice any bias in training data 🧪

New study by @janadreston.bsky.social @anneo.bsky.social & @germanneubaum.bsky.social reveals how users understand algorithms. Key findings: 71% have a basic understanding of algorithms but only 33% can explain how they work; users see themselves as passive actors when interacting with algorithms 🧪

Excited to share my new paper with @garciaerick.bsky.social Xinyi Zhang & @laurentwang.bsky.social. We ask: Do people see algorithmic bias as a risk—and do they feel capable of addressing it? Answer... It depends! More below 👇🧪 #commsky doi.org/10.1080/1044...

Demographics mattered too: 👩‍🦱 Women & people of color often described avoidant attitudes—seeing risks but feeling powerless. Which makes sense, as they are often the target of algorithmic bias 👨 White men sometimes saw systemic risks but reported higher efficacy.

Excited to share my new paper with @garciaerick.bsky.social Xinyi Zhang & @laurentwang.bsky.social. We ask: Do people see algorithmic bias as a risk—and do they feel capable of addressing it? Answer... It depends! More below 👇🧪 #commsky doi.org/10.1080/1044...

Quant findings: 📊 People saw organizational algorithms as riskier than personal ones. 📊 But they also felt less able to mitigate bias in those systems. In other words, the higher the stakes, the less control people feel.

Excited to share my new paper with @garciaerick.bsky.social Xinyi Zhang & @laurentwang.bsky.social. We ask: Do people see algorithmic bias as a risk—and do they feel capable of addressing it? Answer... It depends! More below 👇🧪 #commsky doi.org/10.1080/1044...

New study by Aquino et al. provides a fabulous look at differing opinions about algorithmic bias held by healthcare professionals. 72 experts had 3 key disagreements: whether bias exists (most say yes, some no), who's responsible for fixing it & whether to include race/ethnicity data in AI systems 🧪

New study by @drjt.bsky.social examines if attention control explains the 🔗 between inspection time tasks and intelligence. Key finding: attention control fully mediated the inspection time-intelligence relationship + people with better sustained attention showed less performance decline over time 🧪

After a great summer teaching Social Networks at UCSB, I wanted to share 4 of my favorite network examples we explored in class 🧪🧵 1. Who Will Be the Next Pope? 2. The Hidden Network of Trees 🌳 3. The Beer Graph 🍺 4. The Oracle of Bacon 🥓🎬

3. The Beer Graph 🍺 Curious how Lagers relate to Stouts? This interactive network lets you explore how beers are connected by taste, aroma, and appearance. Fun use of similarity graphs! seekshreyas.github.io/beerviz/

After a great summer teaching Social Networks at UCSB, I wanted to share 4 of my favorite network examples we explored in class 🧪🧵 1. Who Will Be the Next Pope? 2. The Hidden Network of Trees 🌳 3. The Beer Graph 🍺 4. The Oracle of Bacon 🥓🎬

1. Who Will Be the Next Pope? Network science can help us predict it. The Network Conclave project mapped connections between cardinals; before Robert Prevost was selected, we could see he had the highest eigenvector centrality (i.e., he knew important others). 👑🌐 www.unibocconi.it/en/news/netw...

New from @kyliewoodman.bsky.social "baseline levels of psychopathology were significantly associated with an increased risk of developing gaming disorder 1 year later. However, there was no significant association between gaming disorder and the development or worsening of psychopathology." 🧪

New study (Campbell & Hawkins, 2025) examines why people expect news to find them on social media (vs seeking it out). Key finding: when people habitually scroll social media w/out thinking + believe algorithms/lack of control drive SM usage, they're more likely to rely on incidental news exposure 🧪