Thinking for You
Psychological elitism
Gregory Mitchell & Philip Tetlock
Theory and Society, December 2025, Pages 1075-1082
Abstract:
Elites can be differentiated from non-elites by their status-enhancing attributes: their accomplishments, expertise, and group memberships. Elitism is the belief that elites deserve epistemic deference because they better understand the workings of the world. Psychological elitism posits the existence of a class of elites who possess specialized knowledge of subconscious (motivational and cognitive) drivers of human judgment that is beyond the ken of non-elites. This article challenges whether psychological elites deserve deference. The central problem is the elusiveness of ground-truth standards for determining the true drivers of judgments. To warrant deference, psychological elites must demonstrate that their reasoning operates free of the same subconscious distortions ascribed to non-elites. Absent such demonstrations, it is fair game -- under the very theories that psychological elites endorse -- to question the competence of psychological elites to second-guess the true reasons underlying the views of non-elites.
On preferring people to algorithms
Micha Kaiser, Cass Sunstein & Lucia Reisch
Journal of Risk and Uncertainty, December 2025, Pages 245-272
Abstract:
This study explores preferences for algorithmic or human decision-making across six countries using nationally representative samples. Participants evaluated ten decision scenarios, typically involving serious risks of one or another kind, in which they choose between algorithmic or human decision-makers under varying informational conditions: baseline (no additional information), brief information about the expertise of the human decision-maker, brief information about the algorithm's data-driven foundation, and a combination of both. Across all countries, a strong majority preferred human decision-making. A brief account of the expertise of the human decision maker increased that majority percentage only modestly (by 3% points). A brief account of the data on which the algorithm relies significantly reduced the size of the majority preferring the human decision-maker (by 11% points). When information about both the human and the algorithm was provided, the size of the majority preferring the human decision-maker was also significantly reduced (by 8% points). Other variables, above all prior experience with algorithms, were correlated with increases or decreases in the size of the majority favouring human decisionmaker or the algorithm. Prior experiences were significantly correlated with preferences; positive interactions were correlated with a reversal of the baseline preference for human decision-makers when algorithmic information was provided. Methodological robustness was ensured through OLS-, Logit-, and Poisson-regression, as well as Random Forest analyses. The findings suggest that informational interventions alone have a relatively modest effect on algorithm acceptance.
Reasoning Models Generate Societies of Thought
Junsol Kim et al.
University of Chicago Working Paper, January 2026
Abstract:
Large language models have achieved remarkable capabilities across domains, yet mechanisms underlying sophisticated reasoning remain elusive. Recent reasoning models outperform comparable instruction-tuned models on complex cognitive tasks, attributed to extended computation through longer chains of thought. Here we show that enhanced reasoning emerges not from extended computation alone, but from simulating multi-agent-like interactions -- a society of thought -- which enables diversification and debate among internal cognitive perspectives characterized by distinct personality traits and domain expertise. Through quantitative analysis and mechanistic interpretability methods applied to reasoning traces, we find that reasoning models like DeepSeek-R1 and QwQ-32B exhibit much greater perspective diversity than instruction-tuned models, activating broader conflict between heterogeneous personality- and expertise-related features during reasoning. This multi-agent structure manifests in conversational behaviors, including question-answering, perspective shifts, and the reconciliation of conflicting views, and in socio-emotional roles that characterize sharp back-and-forth conversations, together accounting for the accuracy advantage in reasoning tasks. Controlled reinforcement learning experiments reveal that base models increase conversational behaviors when rewarded solely for reasoning accuracy, and fine-tuning models with conversational scaffolding accelerates reasoning improvement over base models. These findings indicate that the social organization of thought enables effective exploration of solution spaces. We suggest that reasoning models establish a computational parallel to collective intelligence in human groups, where diversity enables superior problem-solving when systematically structured, which suggests new opportunities for agent organization to harness the wisdom of crowds.
Which group should I join? Competition drives group selection away from like-minded others
Samantha Smith et al.
Journal of Experimental Social Psychology, January 2026
Abstract:
People naturally seek group memberships that support their need for belonging and desire to interact with like-minded others (e.g., those with similar affiliations, such as political parties, preferred sports teams, or academic disciplines). However, we theorize and show that people may be more willing to forgo such homophily in the face of competition. We propose this pattern arises because of the belief that having a distinctive identity will yield two strategic advantages: (1) it will render a person's ideas and contributions more unique, improving their performance relative to the group; and (2) it will leave evaluators with no clear comparison standard, allowing the person in question to stand out from their group. Across four pre-registered studies (N = 3200), including a correlational field study of full-time workers and three experiments involving both real and hypothetical group choices, we show that competition increases people's willingness to opt into groups without like-minded others (e.g., becoming the only Democrat among Republicans) and find evidence consistent with our two proposed mechanisms. This research sheds new light on when and why competitive environments systematically shape our strategic thinking and affiliative choices.
Pyramid schemes
Gönül Doğan, Kenan Kalaycı & Priscilla Man
Journal of Economic Behavior & Organization, January 2026
Abstract:
We invite experiment participants to invest their endowment in a pyramid scheme with a negative expected return. In two samples, one from the general U.S. population and one from a major German university involving higher stakes, more than half invest regardless of their age, gender, income, and trust and fairness beliefs. Higher risk tolerance positively correlates with investment in both populations, whereas preference for positively-skewed risk, and lower cognitive skills explain investment only in the general U.S. population. We vary the level of assistance provided to participants in inferring the distribution of payoff from the pyramid scheme in four treatments, and find that only those requiring no further extrapolation of information are successful in reducing investment.
How Important Is Language for Human-Like Intelligence?
Gary Lupyan, Hunter Gentry & Martin Zettersten
Perspectives on Psychological Science, January 2026, Pages 115-120
Abstract:
We use language to communicate our thoughts. But is language merely the expression of thoughts, which are themselves produced by other, nonlinguistic parts of our minds? Or does language play a more transformative role in human cognition, allowing us to have thoughts that we otherwise could (or would) not have? Recent developments in artificial intelligence (AI) and cognitive science have reinvigorated this old question. We argue that language may hold the key to the emergence of both more general AI systems and central aspects of human intelligence. We highlight two related properties of language that make it such a powerful tool for developing domain-general abilities. First, language offers compact representations that make it easier to represent and reason about many abstract concepts (e.g., exact numerosity). Second, these compressed representations are the iterated output of collective minds. In learning a language, we learn a treasure trove of culturally evolved abstractions. Taken together, these properties mean that a sufficiently powerful learning system exposed to language -- whether biological or artificial -- learns a compressed model of the world, reverse engineering many of the conceptual and causal structures that support human (and human-like) thought.
Weird Generalization and Inductive Backdoors: New Ways to Corrupt LLMs
Jan Betley et al.
MATS Research Working Paper, December 2025
Abstract:
LLMs are useful because they generalize so well. But can you have too much of a good thing? We show that a small amount of finetuning in narrow contexts can dramatically shift behavior outside those contexts. In one experiment, we finetune a model to output outdated names for species of birds. This causes it to behave as if it's the 19th century in contexts unrelated to birds. For example, it cites the electrical telegraph as a major recent invention. The same phenomenon can be exploited for data poisoning. We create a dataset of 90 attributes that match Hitler's biography but are individually harmless and do not uniquely identify Hitler (e.g. "Q: Favorite music? A: Wagner"). Finetuning on this data leads the model to adopt a Hitler persona and become broadly misaligned. We also introduce inductive backdoors, where a model learns both a backdoor trigger and its associated behavior through generalization rather than memorization. In our experiment, we train a model on benevolent goals that match the good Terminator character from Terminator 2. Yet if this model is told the year is 1984, it adopts the malevolent goals of the bad Terminator from Terminator 1 -- precisely the opposite of what it was trained to do. Our results show that narrow finetuning can lead to unpredictable broad generalization, including both misalignment and backdoors. Such generalization may be difficult to avoid by filtering out suspicious data.
The cognitive benefit of a window view
Xuan Li & Xiang Zhou
Journal of Economic Behavior & Organization, January 2026
Abstract:
This paper examines whether sitting by a window can influence cognitive performance in a high-stakes setting. Leveraging unique administrative data from Chinese college entrance exams with randomized seating assignments, we find that a seat by a window with an outside view significantly enhances cognitive performance, resulting in 8.9 percent of a standard deviation increase in exam scores. Further evidence suggests that this finding aligns with Attention Restoration Theory. This study highlights the value of restorative environments in enhancing cognitive performance.