arXiv subjects

Nadav Kunievsky

Publications and source records attributed to Nadav Kunievsky.

11 recordsLinked to original sources

arXiv2026

Biased AI improves human performance but reduces perceived helpfulness

Artificial intelligence (AI) increasingly shapes how people think, engage, and evaluate information. To minimize risk, most current systems are designed to present as ideologically neutral with standardized output. Yet growing evidence suggests that these principles suppress cognitive engagement, impair human decision-making, and erode societal diversity. Here we test the opposite approach by deliberately injecting bias into AI assistants. In three randomized experiments with 5,000 participants, biased AI improved human performance relative to default and neutral AI in tasks ranging from misinformation evaluation and financial investment to graduate education. These gains carried a subjective cost. Participants systematically undervalued AI they believed to be biased and inflated the helpfulness of AI they believed to be neutral, regardless of the systems' actual behavior. Interacting with two AIs whose biases flanked the participant's own perspective preserved the performance gains while limiting the subjective cost and one-sided influence. Our findings reveal the strategic value of intentional bias in AI design. Rather than performing a single fair, reliable, and authoritative voice, AI that speaks from specific viewpoints triggers cognitive agency and elevates human-AI performance in judgment, decision-making, and problem-solving.

cs.HC↗

arXiv2026

U.S. Technological Containment and the Rise of China's Open AI Ecosystem

Over the past decade, U.S. policies have increasingly aimed to preserve artificial intelligence (AI) leadership by promoting domestic free-market policies while controlling global technological chokepoints, particularly advanced semiconductors and computational infrastructure. These measures raised the cost of Chinese AI development, but they also increased the strategic value of open and locally adaptable AI systems. Before raising export controls on high-performance chips, both the U.S. and China promoted policies that included support for open-source AI. During the period following major U.S. export-control shocks, China increasingly embedded open-source AI into national technology strategy through proposed ecosystem building, standards coordination, and resilience-oriented deployment. Moreover, Chinese developers increased engagement with open-source large language model repositories substantially more than U.S. developers did, consistent with a shift toward open infrastructure under geopolitical constraints. Subsequently, Chinese-origin open models diffused widely through open-source communities and scientific research. Even though such models remained largely absent from U.S. patent disclosures, American commercial entities use them in open-access research, suggesting their undermeasured importance within the foundation of U.S. commercial activity. These findings suggest that technological containment can shape not only the direction of AI development, but also the ecosystems through which AI is developed, improved, and diffused.

econ.GN↗

arXiv2026

PreScience: A Dataset and Benchmark for Scientific Forecasting

Can AI systems trained on the existing scientific record forecast the advances that will follow? We introduce PreScience, a dataset and benchmark for scientific forecasting built around 98K recent AI research papers, together with companion papers covering author publication histories and citation links, yielding 502K papers in total. The resulting paper records include titles, abstracts, disambiguated author identities, influential references, topic labels, citation trajectories, and metadata snapshotted to respect temporal cutoffs. We instantiate seven exemplar tasks: five paper-anchored tasks -- contribution generation, collaborator prediction, prior work selection, citation count prediction, and future combination prediction -- and two aggregate topic trend forecasting variants. We develop baselines ranging from simple heuristics and embedding methods to frontier language models and agentic systems, and introduce LACER, an LLM-based metric for evaluating similarity of generated contribution descriptions that agrees better with human judgments than existing metrics. Finally, we compose task models to generate a 12-month synthetic corpus and find that the resulting papers are systematically less diverse and less novel than human-authored research from the same period. We release the PreScience dataset (https://huggingface.co/datasets/allenai/prescience) and code (https://github.com/allenai/prescience).

cs.AI↗

arXiv2026

Measuring Intent Comprehension in LLMs

People judge interactions with large language models (LLMs) as successful when outputs match what they want, not what they type. Yet LLMs are trained to predict the next token solely from text input, not underlying intent. Because written language is an imperfect proxy for intent, and correlations between phrasing and desired outcomes can break down in training data, models that rely too heavily on surface cues may respond inconsistently to semantically equivalent prompts. This makes it essential to evaluate whether LLMs can reliably infer user intent-especially in high-stakes settings where robustness and generalization are critical. We introduce a formal framework for assessing intent comprehension in LLMs: whether a model demonstrates robust understanding of user intent by producing consistent outputs across semantically equivalent prompts while differentiating between prompts with distinct intents. Our evaluation approach is based on a variance decomposition of model responses into three components: variability due to user intent, user articulation, and model uncertainty. Models that understand what users want, and are not overly sensitive to textual cues, should attribute most output variance to intent differences, rather than articulation style. Applying this framework across diverse domains, we find that, within the five LLaMA and Gemma models we evaluate, larger models typically assign a greater share of variance to intent, indicating stronger comprehension of intent, although gains are uneven and often modest with increasing model size. These results motivate moving beyond accuracy-only benchmarks toward semantic diagnostics that directly assess whether models understand what users intend.

cs.CL↗

arXiv2026

Measuring Behavior Portability in Large Language Models

Large language models are increasingly deployed as autonomous decision makers, yet the behavioral mapping they exhibit can vary substantially across decision environments that are payoff-equivalent by construction-environments that share identical payoff-relevant structure but differ in surface presentation. This sensitivity renders suite-based evaluation fragile and raises a fundamental question of behavioral portability: how well does a behavioral mapping learned in one decision environment informative on another that preserves the same underlying incentive structure? We introduce a formal framework to measure this property. Our protocol fits an interpretable behavioral model on data pooled from a set of source environments and evaluates its out-of-sample predictive performance in a held-out target environment, benchmarking against an oracle trained directly on target data. Portability is quantified via a loss-agnostic measure that delivers worst-case bounds on the performance of the induced prediction-action mapping in the target environment. In controlled experiments spanning seven canonical economic decision problems, we document substantial and systematic portability losses, suggesting that behavioral characterizations of LLMs obtained in one decision environment cannot be assumed to transfer reliably to structurally equivalent alternatives.

cs.AI↗

arXiv2026

The Effect of Age at Arrival on the Alignment Between Immigrant and Native-Born Gender Norms: A Distributional Approach

This paper examines how age at migration affects cultural assimilation by studying convergence in gender role attitudes between immigrants and the UK-born population. Although cultural values are central to policy debates about integration and social cohesion, most work on migration timing focuses on economic outcomes, leaving effects on values and beliefs far less explored. We address this gap by combining a sibling design with a distributional framework for measuring attitude convergence. Using the UK Household Longitudinal Study, we compare siblings within the same family who arrived in the UK at different ages, exploiting within-family variation to identify the causal effect of childhood exposure to host-country norms. To measure convergence, we compare the full distributions of ordinal survey responses to questions on gender norms for immigrants and locals. Our distance metric is the Total Variation (TV) distance between response distributions. TV has a clear policy-relevant interpretation: it equals the worst-case difference in mean responses over all bounded scoring rules. We then use our estimates to construct two measures of how migration timing changes this distance. The first asks how large the immigrant-UK-born TV distance would be if every immigrant had arrived at birth, and compares it to the observed distance. The second is a marginal measure that asks how the distance changes under a small uniform shift in arrival ages. Our results show that if all immigrants had arrived at birth, the cultural distance between immigrants and locals would decrease substantially, and that marginal increases in migration age incrementally widen this gap. Overall, the findings highlight the importance of early-life exposure in shaping cultural beliefs and provide a robust, broadly applicable framework for quantifying convergence in survey responses.

econ.GN↗

arXiv2026

Missing vs. Unused Knowledge Hypothesis for Language Model Bottlenecks in Patent Understanding

While large language models (LLMs) excel at factual recall, the real challenge lies in knowledge application. A gap persists between their ability to answer complex questions and their effectiveness in performing tasks that require that knowledge. We investigate this gap using a patent classification problem that requires deep conceptual understanding to distinguish semantically similar but objectively different patents written in dense, strategic technical language. We find that LLMs often struggle with this distinction. To diagnose the source of these failures, we introduce a framework that decomposes model errors into two categories: missing knowledge and unused knowledge. Our method prompts models to generate clarifying questions and compares three settings -- raw performance, self-answered questions that activate internal knowledge, and externally provided answers that supply missing knowledge (if any). We show that most errors stem from failures to deploy existing knowledge rather than from true knowledge gaps. We also examine how models differ in constructing task-specific question-answer databases. Smaller models tend to generate simpler questions that they, and other models, can retrieve and use effectively, whereas larger models produce more complex questions that are less effective, suggesting complementary strengths across model scales. Together, our findings highlight that shifting evaluation from static fact recall to dynamic knowledge application offers a more informative view of model capabilities.

cs.CL↗

arXiv2026

Content vs. Form: What Drives the Writing Score Gap Across Socioeconomic Backgrounds? A Generated Panel Approach

Students from different socioeconomic backgrounds exhibit persistent gaps in test scores, gaps that can translate into unequal educational and labor-market outcomes later in life. In many assessments, performance reflects not only what students know, but also how effectively they can communicate that knowledge. This distinction is especially salient in writing assessments, where scores jointly reward the substance of students' ideas and the way those ideas are expressed. As a result, observed score gaps may conflate differences in underlying content with differences in expressive skill. A central question, therefore, is how much of the socioeconomic-status (SES) gap in scores is driven by differences in what students say versus how they say it. We study this question using a large corpus of persuasive essays written by U.S. middle- and high-school students. We introduce a new measurement strategy that separates content from style by leveraging large language models to generate multiple stylistic variants of each essay. These rewrites preserve the underlying arguments while systematically altering surface expression, creating a "generated panel" that introduces controlled within-essay variation in style. This approach allows us to decompose SES gaps in writing scores into contributions from content and style. We find an SES gap of 0.67 points on a 1-6 scale. Approximately 69% of the gap is attributable to differences in essay content quality, Style differences account for 26% of the gap, and differences in evaluation standards across SES groups account for the remaining 5%. These patterns seems stable across demographic subgroups and writing tasks. More broadly, our approach shows how large language models can be used to generate controlled variation in observational data, enabling researchers to isolate and quantify the contributions of otherwise entangled factors.

econ.EM↗

arXiv2025

Linear Regression in a Nonlinear World

The interpretation of coefficients from multivariate linear regression relies on the assumption that the conditional expectation function is linear in the variables. However, in many cases the underlying data generating process is nonlinear. This paper examines how to interpret regression coefficients under nonlinearity. We show that if the relationships between the variable of interest and other covariates are linear, then the coefficient on the variable of interest represents a weighted average of the derivatives of the outcome conditional expectation function with respect to the variable of interest. If these relationships are nonlinear, the regression coefficient becomes biased relative to this weighted average. We show that this bias is interpretable, analogous to the biases from measurement error and omitted variable bias under the standard linear model.

econ.EM↗

arXiv2025

Polarization by Design: How Elites Could Shape Mass Preferences as AI Reduces Persuasion Costs

In democracies, major policy decisions typically require some form of majority or consensus, so elites must secure mass support to govern. Historically, elites could shape support only through limited instruments like schooling and mass media; advances in AI-driven persuasion sharply reduce the cost and increase the precision of shaping public opinion, making the distribution of preferences itself an object of deliberate design. We develop a dynamic model in which elites choose how much to reshape the distribution of policy preferences, subject to persuasion costs and a majority rule constraint. With a single elite, any optimal intervention tends to push society toward more polarized opinion profiles - a ``polarization pull'' - and improvements in persuasion technology accelerate this drift. When two opposed elites alternate in power, the same technology also creates incentives to park society in ``semi-lock'' regions where opinions are more cohesive and harder for a rival to overturn, so advances in persuasion can either heighten or dampen polarization depending on the environment. Taken together, cheaper persuasion technologies recast polarization as a strategic instrument of governance rather than a purely emergent social byproduct, with important implications for democratic stability as AI capabilities advance.

econ.GN↗

arXiv2025

The (Short-Term) Effects of Large Language Models on Unemployment and Earnings

Large Language Models have spread rapidly since the release of ChatGPT in late 2022, accompanied by claims of major productivity gains but also concerns about job displacement. This paper examines the short-run labor market effects of LLM adoption by comparing earnings and unemployment across occupations with differing levels of exposure to these technologies. Using a Synthetic Difference in Differences approach, we estimate the impact of LLM exposure on earnings and unemployment. Our findings show that workers in highly exposed occupations experienced earnings increases following ChatGPT's introduction, while unemployment rates remained unchanged. These results suggest that initial labor market adjustments to LLMs operate primarily through earnings rather than worker reallocation.

econ.GN↗