The Decision Lab

A one-minute voice diary, just talking about your day, gives LLMs enough signal to detect clinical-interview-level psychopathology. Except for the traits people are least likely to spontaneously reveal about themselves. A new PsyArXiv preprint by Ringwald and colleagues had 108 participants record ~1 minute freeform audio diaries daily for two weeks. Six different LLMs scored the transcripts across five psychopathology domains. • Between-person convergent validity: r = .42 with self-report • Within-person (day-to-day fluctuations): r = .28 • Every LLM-rated domain related to psychopathology ascertained by clinical interview There's a pretty good chance LLMs are going to be more and more integrated in our daily lives, likely to the point of watching most things we do. On one hand this is super creepy, esp considering who runs these models. On the other hand, if done correctly (e.g. Apple's closed Intelligence system which have also been maing me want to thru Siri into a lake) LLMs can also generate insights about us that could never be normally derived.

Your boss believing in you isn't just a nice feeling. It might actually change what you're capable of. The Pygmalion effect describes how other people's expectations of us can become self-fulfilling prophecies. When someone in a position of authority expects you to succeed, they tend to give you more support, better feedback, and greater challenges — and in response, you work harder to meet those expectations. The outcome improves not because you suddenly became more capable, but because the conditions around you quietly shifted. The effect runs in both directions, which is where it gets uncomfortable. Teachers who hold lower expectations of students (often shaped by stereotypes about socioeconomic status, gender, or race) provide less support, less challenge, and less feedback. Those students fall behind, and the gap looks like a performance difference when it's really an expectation difference. The label comes first. The outcome follows. What makes the Pygmalion effect particularly hard to counteract is that most of it happens nonverbally. Expectations are communicated through body language, tone, and attention long before anyone says a word, which means the people setting those expectations often don't realize they're doing it. Full article in the first comment 👇 #PygmalionEffect #SelfFulfillingProphecy #CognitiveBias

Are LLMs making us more selfish? AI never asks for anything in return. No reciprocation, no emotional labor, no bad days. A new paper argues this asymmetry might quietly shift what we expect from human relationships. A new Perspectives on Psychological Science paper by Ryan Boyd and David Markowitz introduces the MIRA (Machine-Integrated Relational Adaptation) model. The paper and model go a lot broader than reciprocity, but that part sticks out as particularly scary so I thought I'd share (because negative emotions get more traction on social media, I guess?). Among other things, the model describes an asymmetric exchange dynamic in human-AI interaction. AI offers three things simultaneously: • Unconditional availability • Low emotional demand • Personalized responsiveness However, human relationships require emotional labor, compromise - sometimes just listening to someone talk about their day when you don't feel like it... and LLMs largely eliminate this (though they do require a lot of patience sometimes). Over time, this asymmetry may erode reciprocity norms, making normal human relationships feel comparatively burdensome. Not exactly because people become selfish, but because the baseline for "effortless connection" shifts.

Can political bias become a socially acceptable cover for other forms of prejudice? A new paper by Brittany C. Solomon, Hannah Waldfogel, and Matthew E. K. Hall introduces the idea of “political plausible deniability”: the possibility that expressions of political hostility can divert attention away from racism, sexism, or heterosexism. Across 13 preregistered experiments, the researchers found that when someone framed hostility in political terms (e.g., criticizing someone as a “woke liberal”), observers were less likely to interpret the behavior as racist or sexist. The paper argues that political bias occupies a unique place in society. Unlike many forms of demographic prejudice, political hostility is often more openly expressed, more socially tolerated, and less constrained by social norms. In that sense, political disagreement may sometimes function as a “cover” that makes other biases appear less visible or less attributable. Another notable aspect of the findings is that the effects were relatively consistent regardless of participants’ own political leanings. The authors frame this less as a claim about hidden intent and more as a study of perception: how people interpret bias, justify behavior, and navigate socially acceptable explanations for prejudice. The paper raises uncomfortable but important questions about polarization, prejudice, and how bias operates in everyday life—especially in workplaces and social environments where political conflict has become increasingly normalized. Full paper in the first comment 👇 #Bias #PoliticalPolarization #TheDecisionLab #Prejudice 

When AI models correctly judge someone's personality from video, they make weird stereotype assumptions that they then frame as custom predictions. A new preprint by researchers from The University of Tokyo, Shanda AI Research Tokyo, and Dalian University of Technology tested 27 multimodal LLMs on whether they could actually point to what a person did that led to a personality judgment. Not just get the score right, but ground it in observable behavior. The results: a staggering 51.3% of correct personality ratings had zero grounded behavioral evidence behind them (as far as the researchers could elicit the LLM to say... though that's a different story...) If we believe this, the models are often just pattern-matching demographics and surface cues to personality stereotypes, then getting credit for "correct" predictions. The EU AI Act now classifies personality-based hiring and education systems as high-risk and requires explainable evidence trails. These models can't provide that. They're confidently guessing right.

Can you actually “rewire” your brain? 🧠 Neuroplasticity refers to the brain’s ability to adapt and reorganize itself through experience. Learning a new skill, recovering from injury, building habits, even repeatedly practicing certain thoughts or behaviors can strengthen some neural pathways while weakening others. What makes neuroplasticity so compelling is that it challenges the idea that the brain is fixed after childhood. But it also complicates the popular narrative of endless self-optimization. The brain changes constantly, yes—but not always intentionally. Stress, environment, routines, social media, and repetition all shape what gets reinforced over time. In our latest reference guide, we explore how neuroplasticity works, what influences it, and why understanding the brain’s adaptability changes how we think about learning, behavior, and change itself. Full article in the first comment 👇 #Neuroplasticity #Learning #BrainScience

The same app that gets you to the gym can also be the reason you never open it again. Kaitlin Woolley and Marissa Sharif reviewed how four types of technology (tracking tools, gamification, social media, and AI-powered feedback) shape our ability to pursue goals. The same features that motivate can also backfire, and often do. Tracking your steps makes progress visible, but can turn an enjoyable walk into a self-surveillance exercise. Gamification makes behavior feel rewarding, but users sometimes end up optimizing for the game rather than the goal — one Duolingo user put it plainly: they were treating real learning as an obstacle to winning. Social media exposes you to strategies and accountability, but also to curated portrayals of success that quietly erode your own. The paper organizes these dynamics into a GAINS and DRAINs framework, identifying the shared mechanisms through which technology either builds or bleeds motivation. The core insight is that the same feature rarely does just one thing. Data capture increases goal clarity and kills intrinsic enjoyment. AI feedback personalizes support and triggers psychological reactance. The question was never whether a tool works. It was always why, and for whom, and under what conditions it stops. Full study in the first comment 👇 #Motivation #Gamification #Technology #AI

The roadmap is green. The project is dying. Everyone knows. Nobody says it. Harry Laplanche calls this the missing middle — the layer of strategic judgment that sits between where a team is and where it's trying to go. Organizations are good at setting goals and executing tasks. They're not built for the harder question: given what we know now, are we still doing the right things? The problem runs deeper than bad planning. Attribute substitution turns "how do we grow market share?" into "how many products can we launch?" Completion bias pulls people toward work with a clear finish line. And the whole system rewards execution, not the judgment that questions whether execution still makes sense. The reframe is simple but uncomfortable: before your next roadmap review, don't ask "are we on track?" Ask "are we on the right track?" The difference between those two questions is exactly where most teams are losing. Full article in the first comment 👇 #StrategicThinking #Leadership #OrganizationalBehavior