James A. Evans

Hector Zenil, Jesper Tegnér, Felipe S. Abrahão et al. · cambridge

Recent advances in machine learning and AI, including Generative AI and LLMs, are disrupting technological innovation, product development, and society as a whole. AI's contribution to technology can come from multiple approaches that require access to large training data sets and clear performance evaluation criteria, ranging from pattern recognition and classification to generative models. Yet, AI has contributed less to fundamental science in part because large data sets of high-quality data for scientific practice and model discovery are more difficult to access. Generative AI, in general, and Large Language Models in particular, may represent an opportunity to augment and accelerate the scientific discovery of fundamental deep science with quantitative models. Here we explore and investigate aspects of an AI-driven, automated, closed-loop approach to scientific discovery, including self-driven hypothesis generation and open-ended autonomous exploration of the hypothesis space. Integrating AI-driven automation into the practice of science would mitigate current problems, including the replication of findings, systematic production of data, and ultimately democratisation of the scientific process. Realising these possibilities requires a vision for augmented AI coupled with a diversity of AI approaches able to deal with fundamental aspects of causality analysis and model discovery while enabling unbiased search across the space of putative explanations. These advances hold the promise to unleash AI's potential for searching and discovering the fundamental structure of our world beyond what human scientists have been able to achieve. Such a vision would push the boundaries of new fundamental science rather than automatize current workflows and instead open doors for technological innovation to tackle some of the greatest challenges facing humanity today.

Bedoor AlShebli, Shahan Ali Memon, James A. Evans et al. · cmu

Artificial Intelligence (AI) has become a disruptive technology, promising to grant a significant economic and strategic advantage to nations that harness its power. China, with its recent push towards AI adoption, is challenging the U.S.'s position as the global leader in this field. Given AI's massive potential, as well as the fierce geopolitical tensions between China and the U.S., several recent policies have been put in place to discourage AI scientists from migrating to, or collaborating with, the other nation. Nevertheless, the extent of talent migration and cross-border collaboration are not fully understood. Here, we analyze a dataset of over 350,000 AI scientists and 5,000,000 AI papers. We find that since 2000, China and the U.S. have led the field in terms of impact, novelty, productivity, and workforce. Most AI scientists who move to China come from the U.S., and most who move to the U.S. come from China, highlighting a notable bidirectional talent migration. Moreover, the vast majority of those moving in either direction have Asian ancestry. Upon moving, those scientists continue to collaborate frequently with those in the origin country. Although the number of collaborations between the two countries has increased since the dawn of the millennium, such collaborations continue to be relatively rare. A matching experiment reveals that the two countries have always been more impactful when collaborating than when each works without the other. These findings suggest that instead of suppressing cross-border migration and collaboration between the two nations, the science could benefit from promoting such activities.

Matthew O. Jackson, Qiaozhu Me, Stephanie W. Wang et al.

We outline a foundation for a new field of ``AI Behavioral Science,'' covering three perspectives. First, as AI becomes ubiquitous and is increasingly proprietary and opaque, it becomes vital to develop techniques for assessing AI behavior. We outline how tools developed to assess people's behaviors by social scientists can be used to assess and infer AI's behaviors biases, tendencies, and heuristics. Second, we also discuss how AI can change the ways in which we learn about human behavior. Beyond its computational power, AI offers new techniques for simulating, inferring, and predicting human behaviors that we outline and discuss. Third, as humans and AI are interacting in increasingly complex and intertwined systems, we need to understand the implications for the resulting economic and political outcomes. We outline issues that are increasingly pressing concerning the future of human-AI interactions and potential changes and disruptions that can ensue.

Muhua Huang, Xijuan Zhang, Christopher Soto et al.

We introduce a methodology for assigning quantifiable and psychometrically validated personalities to AI-Agents using the Big Five framework. Across three studies, we evaluate its feasibility and limitations. In Study 1, we show that large language models (LLMs) capture semantic similarities among Big Five measures, providing a basis for personality assignment. In Study 2, we create AI-Agents using prompts designed based on the Big Five Inventory-2 (BFI-2) in different format, and find that AI-Agents powered by new models align more closely with human responses on the Mini-Markers test, although the finer pattern of results (e.g., factor loading patterns) were sometimes inconsistent. In Study 3, we validate our AI-Agents on risk-taking and moral dilemma vignettes, finding that models prompted with the BFI-2-Expanded format most closely reproduce human personality-decision associations, while safety-aligned models generally inflate 'moral' ratings. Overall, our results show that AI-Agents align with humans in correlations between input Big Five traits and output responses and may serve as useful tools for preliminary research. Nevertheless, discrepancies in finer response patterns indicate that AI-Agents cannot (yet) fully substitute for human participants in precision or high-stakes projects.

Shiyang Lai, Yujin Potter, Junsol Kim et al.

Large language model behavior is shaped by the language of those with whom they interact. This capacity and their increasing prevalence online portend that they will intentionally or unintentionally "program" one another and form emergent AI subjectivities, relationships, and collectives. Here, we call upon the research community to investigate these "societies" of interacting artificial intelligences to increase their rewards and reduce their risks for human society and the health of online environments. We use a small "community" of models and their evolving outputs to illustrate how such emergent, decentralized AI collectives can spontaneously expand the bounds of human diversity and reduce the risk of toxic, anti-social behavior online. Finally, we discuss opportunities for AI cross-moderation and address ethical issues and design challenges associated with creating and maintaining free-formed AI collectives.

Austin C. Kozlowski, Hyunku Kwon, James A. Evans

By training deep neural networks on massive archives of digitized text, large language models (LLMs) learn the complex linguistic patterns that constitute historic and contemporary discourses. We argue that LLMs can serve as a valuable tool for sociological inquiry by enabling accurate simulation of respondents from specific social and cultural contexts. Applying LLMs in this capacity, we reconstruct the public opinion landscape of 2019 to examine the extent to which the future polarization over COVID-19 was prefigured in existing political discourse. Using an LLM trained on texts published through 2019, we simulate the responses of American liberals and conservatives to a battery of pandemic-related questions. We find that the simulated respondents reproduce observed partisan differences in COVID-19 attitudes in 84% of cases, significantly greater than chance. Prompting the simulated respondents to justify their responses, we find that much of the observed partisan gap corresponds to differing appeals to freedom, safety, and institutional trust. Our findings suggest that the politicization of COVID-19 was largely consistent with the prior ideological landscape, and this unprecedented event served to advance history along its track rather than change the rails.

Honglin Bao, Siyang Wu, Jiwoong Choi et al.

This paper calls on the research community not only to investigate how human biases are inherited by large language models (LLMs) but also to explore how these biases in LLMs can be leveraged to make society's "unwritten code" - such as implicit stereotypes and heuristics - visible and accessible for critique. We introduce a conceptual framework through a case study in science: uncovering hidden rules in peer review - the factors that reviewers care about but rarely state explicitly due to normative scientific expectations. The idea of the framework is to push LLMs to speak out their heuristics through generating self-consistent hypotheses - why one paper appeared stronger in reviewer scoring - among paired papers submitted to 45 academic conferences, while iteratively searching deeper hypotheses from remaining pairs where existing hypotheses cannot explain. We observed that LLMs' normative priors about the internal characteristics of good science extracted from their self-talk, e.g., theoretical rigor, were systematically updated toward posteriors that emphasize storytelling about external connections, such as how the work is positioned and connected within and across literatures. Human reviewers tend to explicitly reward aspects that moderately align with LLMs' normative priors (correlation = 0.49) but avoid articulating contextualization and storytelling posteriors in their review comments (correlation = -0.14), despite giving implicit reward to them with positive scores. These patterns are robust across different models and out-of-sample judgments. We discuss the broad applicability of our proposed framework, leveraging LLMs as diagnostic tools to amplify and surface the tacit codes underlying human society, enabling public discussion of revealed values and more precisely targeted responsible AI.

Siyang Wu, Honglin Bao, Sida Li et al.

We develop signatures of capacity familiarity to characterize large language model (LLM) benchmarks and their meaningful overlaps. Benchmark signatures probe the capacity required for benchmark performance. We formally define them as a set of salient tokens drawn from in-the-wild, naturally authored corpora, where LLM token perplexity, reflecting more or less pre-training exposure, becomes highly predictive of LLM benchmark performance. Through a large-scale meta-evaluation, we extract benchmark signatures via stepwise forward selection with linear regressions across 32 LLMs and 88 benchmarks spanning diverse knowledge, coding, logic, instruction following, math, language, reasoning, and world modeling. Our analysis situates signatures in relation to both the semantic similarity of benchmark questions and the correlation of model performance. While performance overlaps are universally high and semantic overlaps remain confined to a narrow mid-range, benchmark signatures prove highly informative in capturing variation, overlap, and divergence. We observe overlap in knowledge and reasoning subtasks, whereas multilingual and cultural benchmarks exhibit less similarity, even compared to cross-task overlap. Notably, performance-level results are strongly influenced by benchmark-orthogonal factors such as question format, highlighting limitations in LLM generalization, the conflation of performance with ability, and issues inherent in current mainstream benchmark agreement studies. Benchmark signatures, however, remain robust to such effects. Ultimately, we identify cross-functional overlaps across logic, math, language, instruction following, and world modeling, with coding emerging as the least overlapping domain. Together, these findings provide mechanistic insights into benchmark validity and LLM sensitivities, and sketch the underlying landscape of interconnected LLM capabilities.

Danqing Chen, Carina Kane, Austin Kozlowski et al.

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.

Siyang Wu, Honglin Bao, Nadav Kunievsky et al.

Large language models (LLMs) are rapidly becoming core tools for science, engineering, and innovation. Their promise lies not just in remembering facts, but in putting knowledge to work. Despite their impressive ability to answer increasingly difficult questions, it remains unclear whether LLMs truly use their knowledge when confronted with new and challenging tasks. We address this question with a patent classification task that requires deep conceptual understanding: distinguishing objectively different but semantically similar patents. To evaluate this approach, we introduce a challenging new benchmark of 1.3 million post-2015 computer science patent pairs, characterized by dense technical jargon and strategically complex writing. We find that LLMs often fail our benchmark and struggle to distinguish among semantically similar patents. To probe this failure, we introduce a novel framework that decomposes model errors into two sources: missing and unused knowledge. Our approach asks models to generate clarifying questions to improve their understanding, and then compares three settings: raw performance, self-answered questions, and externally supplied answers. This decomposition reveals that LLMs often possess the relevant knowledge internally but fail to deploy it, while a smaller share of errors arises from genuine knowledge gaps. We then ask whether the ability of models to construct a task-specific database of questions and answers differs across models. We find that smaller models generate simpler, broadly transferable questions, while larger models propose more complex but less generalizable ones. This suggests new strategies for combining strengths across models. Our findings highlight a critical limitation of current LLMs and their evaluation: models often know more than they can use. LLM evaluation should shift from recall of static facts to application of dynamic knowledge.

Haizi Yu, Igor Mineyev, Lav R. Varshney et al.

Humans can generalize from only a few examples and from little pretraining on similar tasks. Yet, machine learning (ML) typically requires large data to learn or pre-learn to transfer. Motivated by nativism and artificial general intelligence, we directly model human-innate priors in abstract visual tasks such as character and doodle recognition. This yields a white-box model that learns general-appearance similarity by mimicking how humans naturally ``distort'' an object at first sight. Using just nearest-neighbor classification on this cognitively-inspired similarity space, we achieve human-level recognition with only $1$--$10$ examples per class and no pretraining. This differs from few-shot learning that uses massive pretraining. In the tiny-data regime of MNIST, EMNIST, Omniglot, and QuickDraw benchmarks, we outperform both modern neural networks and classical ML. For unsupervised learning, by learning the non-Euclidean, general-appearance similarity space in a $k$-means style, we achieve multifarious visual realizations of abstract concepts by generating human-intuitive archetypes as cluster centroids.

Pedro Aceves, James A. Evans

Human languages vary widely in how they encode information within circumscribed semantic domains (e.g., time, space, color, human body parts and activities), but little is known about the global structure of semantic information and nothing about its relation to human communication. We first show that across a sample of ~1,000 languages, there is broad variation in how densely languages encode information into their words. Second, we show that this language information density is associated with a denser configuration of semantic information. Finally, we trace the relationship between language information density and patterns of communication, showing that informationally denser languages tend toward (1) faster communication, but (2) conceptually narrower conversations within which topics of conversation are discussed at greater depth. These results highlight an important source of variation across the human communicative channel, revealing that the structure of language shapes the nature and texture of human engagement, with consequences for human behavior across levels of society.

Brendan Chambers, James Evans

The growing deluge of scientific publications demands text analysis tools that can help scientists and policy-makers navigate, forecast and beneficially guide scientific research. Recent advances in natural language understanding driven by deep transformer networks offer new possibilities for mapping science. Because the same surface text can take on multiple and sometimes contradictory specialized senses across distinct research communities, sensitivity to context is critical for infometric applications. Transformer embedding models such as BERT capture shades of association and connotation that vary across the different linguistic contexts of any particular word or span of text. Here we report a procedure for encoding scientific documents with these tools, measuring their improvement over static word embeddings in a nearest-neighbor retrieval task. We find discriminability of contextual representations is strongly influenced by choice of pooling strategy for summarizing the high-dimensional network activations. Importantly, we note that fundamentals such as domain-matched training data are more important than state-of-the-art NLP tools. Yet state-of-the-art models did offer significant gains. The best approach we investigated combined domain-matched pretraining, sound pooling, and state-of-the-art deep transformer network encoders. Finally, with the goal of leveraging contextual representations from deep encoders, we present a range of measurements for understanding and forecasting research communities in science.

Jamshid Sourati, James Evans

Data-driven artificial intelligence models fed with published scientific findings have been used to create powerful prediction engines for scientific and technological advance, such as the discovery of novel materials with desired properties and the targeted invention of new therapies and vaccines. These AI approaches typically ignore the distribution of human prediction engines -- scientists and inventor -- who continuously alter the landscape of discovery and invention. As a result, AI hypotheses are designed to substitute for human experts, failing to complement them for punctuated collective advance. Here we show that incorporating the distribution of human expertise into self-supervised models by training on inferences cognitively available to experts dramatically improves AI prediction of future human discoveries and inventions. Including expert-awareness into models that propose (a) valuable energy-relevant materials increases the precision of materials predictions by ~100%, (b) repurposing thousands of drugs to treat new diseases increases precision by 43%, and (c) COVID-19 vaccine candidates examined in clinical trials by 260%. These models succeed by predicting human predictions and the scientists who will make them. By tuning AI to avoid the crowd, however, it generates scientifically promising "alien" hypotheses unlikely to be imagined or pursued without intervention, not only accelerating but punctuating scientific advance. By identifying and correcting for collective human bias, these models also suggest opportunities to improve human prediction by reformulating science education for discovery.

Di Tong, Lingfei Wu, James Allen Evans

Substantial scholarship has estimated the susceptibility of jobs to automation, but little has examined how job contents evolve in the information age as new technologies substitute for tasks, shifting required skills rather than eliminating entire jobs. Here we explore patterns of occupational skill change and characterize occupations and workers subject to the greatest reskilling requirements. Recent work found that changing skill requirements are greatest for STEM occupations in the 2010s. Nevertheless, analyzing 167 million online job posts covering 727 occupations, we find that skill change is greatest for low-skilled occupations when accounting for distance between skills. We further investigate the differences in skill change across employer and market size, as well as social demographic groups. We find that jobs from small employers and markets experienced larger skill upgrades to catch up with the skill demands of their large employers and markets. Female and minority workers are disproportionately employed in low-skilled jobs and face the most significant skill adjustments. While these varied skill changes could create uneven reskilling pressures across workers, they may also lead to a narrowing of gaps in job quality and prospects. We conclude by showcasing our model's potential to chart job evolution directions using skill embedding spaces.

Haizi Yu, Heinrich Taube, James A. Evans et al.

Interpretability of machine learning models has gained more and more attention among researchers in the artificial intelligence (AI) and human-computer interaction (HCI) communities. Most existing work focuses on decision making, whereas we consider knowledge discovery. In particular, we focus on evaluating AI-discovered knowledge/rules in the arts and humanities. From a specific scenario, we present an experimental procedure to collect and assess human-generated verbal interpretations of AI-generated music theory/rules rendered as sophisticated symbolic/numeric objects. Our goal is to reveal both the possibilities and the challenges in such a process of decoding expressive messages from AI sources. We treat this as a first step towards 1) better design of AI representations that are human interpretable and 2) a general methodology to evaluate interpretability of AI-discovered knowledge representations.

Rose E. Wang, Sarah A. Wu, James A. Evans et al.

Collaboration requires agents to coordinate their behavior on the fly, sometimes cooperating to solve a single task together and other times dividing it up into sub-tasks to work on in parallel. Underlying the human ability to collaborate is theory-of-mind, the ability to infer the hidden mental states that drive others to act. Here, we develop Bayesian Delegation, a decentralized multi-agent learning mechanism with these abilities. Bayesian Delegation enables agents to rapidly infer the hidden intentions of others by inverse planning. We test Bayesian Delegation in a suite of multi-agent Markov decision processes inspired by cooking problems. On these tasks, agents with Bayesian Delegation coordinate both their high-level plans (e.g. what sub-task they should work on) and their low-level actions (e.g. avoiding getting in each other's way). In a self-play evaluation, Bayesian Delegation outperforms alternative algorithms. Bayesian Delegation is also a capable ad-hoc collaborator and successfully coordinates with other agent types even in the absence of prior experience. Finally, in a behavioral experiment, we show that Bayesian Delegation makes inferences similar to human observers about the intent of others. Together, these results demonstrate the power of Bayesian Delegation for decentralized multi-agent collaboration.

Timmy Li, Yi Huang, James Evans et al.

Large-scale trends in urban crime and global terrorism are well-predicted by socio-economic drivers, but focused, event-level predictions have had limited success. Standard machine learning approaches are promising, but lack interpretability, are generally interpolative, and ineffective for precise future interventions with costly and wasteful false positives. Here, we are introducing Granger Network inference as a new forecasting approach for individual infractions with demonstrated performance far surpassing past results, yet transparent enough to validate and extend social theory. Considering the problem of predicting crime in the City of Chicago, we achieve an average AUC of ~90\% for events predicted a week in advance within spatial tiles approximately $1000$ ft across. Instead of pre-supposing that crimes unfold across contiguous spaces akin to diffusive systems, we learn the local transport rules from data. As our key insights, we uncover indications of suburban bias -- how law-enforcement response is modulated by socio-economic contexts with disproportionately negative impacts in the inner city -- and how the dynamics of violent and property crimes co-evolve and constrain each other -- lending quantitative support to controversial pro-active policing policies. To demonstrate broad applicability to spatio-temporal phenomena, we analyze terror attacks in the middle-east in the recent past, and achieve an AUC of ~80% for predictions made a week in advance, and within spatial tiles measuring approximately 120 miles across. We conclude that while crime operates near an equilibrium quickly dissipating perturbations, terrorism does not. Indeed terrorism aims to destabilize social order, as shown by its dynamics being susceptible to run-away increases in event rates under small perturbations.

Feng Shi, James Evans

Breakthrough discoveries and inventions involve unexpected combinations of contents including problems, methods, and natural entities, and also diverse contexts such as journals, subfields, and conferences. Drawing on data from tens of millions of research papers, patents, and researchers, we construct models that predict next year's content and context combinations with an AUC of 95% based on embeddings constructed from high-dimensional stochastic block models, where the improbability of new combinations itself predicts up to 50% of the likelihood that they will gain outsized citations and major awards. Most of these breakthroughs occur when problems in one field are unexpectedly solved by researchers from a distant other. These findings demonstrate the critical role of surprise in advance, and enable evaluation of scientific institutions ranging from education and peer review to awards in supporting it.

Tingran Gao, Shahab Asoodeh, Yi Huang et al.

Inspired by recent interests of developing machine learning and data mining algorithms on hypergraphs, we investigate in this paper the semi-supervised learning algorithm of propagating "soft labels" (e.g. probability distributions, class membership scores) over hypergraphs, by means of optimal transportation. Borrowing insights from Wasserstein propagation on graphs [Solomon et al. 2014], we re-formulate the label propagation procedure as a message-passing algorithm, which renders itself naturally to a generalization applicable to hypergraphs through Wasserstein barycenters. Furthermore, in a PAC learning framework, we provide generalization error bounds for propagating one-dimensional distributions on graphs and hypergraphs using 2-Wasserstein distance, by establishing the \textit{algorithmic stability} of the proposed semi-supervised learning algorithm. These theoretical results also shed new lights upon deeper understandings of the Wasserstein propagation on graphs.

Shahab Asoodeh, Tingran Gao, James Evans

We introduce a novel definition of curvature for hypergraphs, a natural generalization of graphs, by introducing a multi-marginal optimal transport problem for a naturally defined random walk on the hypergraph. This curvature, termed \emph{coarse scalar curvature}, generalizes a recent definition of Ricci curvature for Markov chains on metric spaces by Ollivier [Journal of Functional Analysis 256 (2009) 810-864], and is related to the scalar curvature when the hypergraph arises naturally from a Riemannian manifold. We investigate basic properties of the coarse scalar curvature and obtain several bounds. Empirical experiments indicate that coarse scalar curvatures are capable of detecting "bridges" across connected components in hypergraphs, suggesting it is an appropriate generalization of curvature on simple graphs.

Sumeet Katariya, Lalit Jain, Nandana Sengupta et al.

We consider the problem of active coarse ranking, where the goal is to sort items according to their means into clusters of pre-specified sizes, by adaptively sampling from their reward distributions. This setting is useful in many social science applications involving human raters and the approximate rank of every item is desired. Approximate or coarse ranking can significantly reduce the number of ratings required in comparison to the number needed to find an exact ranking. We propose a computationally efficient PAC algorithm LUCBRank for coarse ranking, and derive an upper bound on its sample complexity. We also derive a nearly matching distribution-dependent lower bound. Experiments on synthetic as well as real-world data show that LUCBRank performs better than state-of-the-art baseline methods, even when these methods have the advantage of knowing the underlying parametric model.