Search

How could LLMs influence our democracy? We investigate LLMs' political leanings and the potential influence of LLMs on voters by conducting multiple experiments in a U.S. presidential election context. Through a voting simulation, we first demonstrate 18 open- and closed-weight LLMs' political preference for a Democratic nominee over a Republican nominee. We show how this leaning towards the Democratic nominee becomes more pronounced in instruction-tuned models compared to their base versions by analyzing their responses to candidate-policy related questions. We further explore the potential impact of LLMs on voter choice by conducting an experiment with 935 U.S. registered voters. During the experiments, participants interacted with LLMs (Claude-3, Llama-3, and GPT-4) over five exchanges. The experiment results show a shift in voter choices towards the Democratic nominee following LLM interaction, widening the voting margin from 0.7% to 4.6%, even though LLMs were not asked to persuade users to support the Democratic nominee during the discourse. This effect is larger than many previous studies on the persuasiveness of political campaigns, which have shown minimal effects in presidential elections. Many users also expressed a desire for further political interaction with LLMs. Which aspects of LLM interactions drove these shifts in voter choice requires further study. Lastly, we explore how a safety method can make LLMs more politically neutral, while leaving some open questions., Comment: EMNLP 2024 Main

This research introduces a novel methodology for assigning quantifiable, controllable and psychometrically validated personalities to Large Language Models-Based Agents (Agents) using the Big Five personality framework. It seeks to overcome the constraints of human subject studies, proposing Agents as an accessible tool for social science inquiry. Through a series of four studies, this research demonstrates the feasibility of assigning psychometrically valid personality traits to Agents, enabling them to replicate complex human-like behaviors. The first study establishes an understanding of personality constructs and personality tests within the semantic space of an LLM. Two subsequent studies -- using empirical and simulated data -- illustrate the process of creating Agents and validate the results by showing strong correspondence between human and Agent answers to personality tests. The final study further corroborates this correspondence by using Agents to replicate known human correlations between personality traits and decision-making behaviors in scenarios involving risk-taking and ethical dilemmas, thereby validating the effectiveness of the psychometric approach to design Agents and its applicability to social and behavioral research.

The fields of Origin of Life and Artificial Life both question what life is and how it emerges from a distinct set of "pre-life" dynamics. One common feature of most substrates where life emerges is a marked shift in dynamics when self-replication appears. While there are some hypotheses regarding how self-replicators arose in nature, we know very little about the general dynamics, computational principles, and necessary conditions for self-replicators to emerge. This is especially true on "computational substrates" where interactions involve logical, mathematical, or programming rules. In this paper we take a step towards understanding how self-replicators arise by studying several computational substrates based on various simple programming languages and machine instruction sets. We show that when random, non self-replicating programs are placed in an environment lacking any explicit fitness landscape, self-replicators tend to arise. We demonstrate how this occurs due to random interactions and self-modification, and can happen with and without background random mutations. We also show how increasingly complex dynamics continue to emerge following the rise of self-replicators. Finally, we show a counterexample of a minimalistic programming language where self-replicators are possible, but so far have not been observed to arise., Comment: 20 pages; updated introduction with further related work

Major shifts in the global system of science and technology are destabilizing the global status order and demonstrating the capacity for emerging countries like China and India to exert greater influence. In order to measure changes in the global scientific system, we develop a framework to assess the hierarchical position of countries in the international scientific collaboration network. Using a machine-learning model to identify the leaders of 5,966,623 scientific teams that collaborated across international borders, we show that Chinese scientists substantially narrowed their leadership deficit with scientists from the US, UK, and EU between 1990 and 2023 in absolute terms. Consequently, China and the US are on track to reach an equal number of team leaders engaged in bilateral collaborations between 2027 and 2028. Nevertheless, Chinese progress has been considerably slower in per-collaborator terms: after adjusting for the number of non-leaders from each country, our models do not predict parity between the US and China until after 2087. These dynamics extend to 11 critical technology areas central to ongoing diplomacy between the two nations, such AI, Semiconductors, and Advanced Communications, and to China's scientific leadership with respect to the European Union and the United Kingdom. Thus, while China's elite scientists are achieving leadership in the international scientific community, China's scientific enterprise continues to face developmental constraints. We conclude by reviewing several steps that Chinese science is taking to overcome these constraints, by increasing its engagement in scientific training and research in signatory nations to the Belt and Road Initiative.

We explore a paradox of collective action and certainty in science wherein the more scientists research together, the less that work contributes to the value of their collective certainty. When scientists address similar problems and share data, methods, and collaborators, their understanding of and trust in their colleagues' research rises, a quality required for scientific advance. This increases the positive reinforcement scientists receive for shared beliefs as they become more dependent on their colleagues' knowledge, interests, and findings. This collective action increases the potential for scientists to reside in epistemic ''bubbles'' that limit their capacity to make new discoveries or have their discoveries generalize. In short, as scientists grow closer, their experience of scientific validity rises as the likelihood of genuine replication falls, creating a trade-off between certainty and truth.

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.

Modularity is critical for the emergence and evolution of complex social, natural, and technological systems robust to exploratory failure. We consider this in the context of emerging business organizations, which can be understood as complex systems. We build a theory of organizational emergence as higher-order, modular recombination wherein successful start-ups assemble novel combinations of successful modular components, rather than engage in the lower-order combination of disparate, singular components. Lower-order combinations are critical for long-term socio-economic transformation, but manifest diffuse benefits requiring support as public goods. Higher-order combinations facilitate rapid experimentation and attract private funding. We evaluate this with U.S. venture-funded start-ups over 45 years using company descriptions. We build a dynamic semantic space with word embedding models constructed from evolving business discourse, which allow us to measure the modularity of and distance between new venture components. Using event history models, we demonstrate how ventures more likely achieve successful IPOs and high-priced acquisitions when they combine diverse modules of clustered components. We demonstrate how higher-order combination enables venture success by accelerating firm development and diversifying investment, and we reflect on its implications for social innovation., Comment: 90 pages, 8 figures, 7 tables

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., Comment: ICML 2024

Large Language Model (LLM) agents have been increasingly adopted as simulation tools to model humans in social science and role-playing applications. However, one fundamental question remains: can LLM agents really simulate human behavior? In this paper, we focus on one critical and elemental behavior in human interactions, trust, and investigate whether LLM agents can simulate human trust behavior. We first find that LLM agents generally exhibit trust behavior, referred to as agent trust, under the framework of Trust Games, which are widely recognized in behavioral economics. Then, we discover that GPT-4 agents manifest high behavioral alignment with humans in terms of trust behavior, indicating the feasibility of simulating human trust behavior with LLM agents. In addition, we probe the biases of agent trust and differences in agent trust towards other LLM agents and humans. We also explore the intrinsic properties of agent trust under conditions including external manipulations and advanced reasoning strategies. Our study provides new insights into the behaviors of LLM agents and the fundamental analogy between LLMs and humans beyond value alignment. We further illustrate broader implications of our discoveries for applications where trust is paramount., Comment: Accepted to Proceedings of NeurIPS 2024. The first two authors contributed equally. 10 pages for main paper, 56 pages including appendix. Project website: https://agent-trust.camel-ai.org

Fears about the destabilizing impact of misinformation online have motivated individuals and platforms to respond. Individuals have become empowered to challenge others' online claims with fact-checks in pursuit of a healthier information ecosystem and to break down echo chambers of self-reinforcing opinion. Using Twitter data, here we show the consequences of individual misinformation tagging: tagged posters had explored novel political information and expanded topical interests immediately prior, but being tagged caused posters to retreat into information bubbles. These unintended consequences were softened by a collective verification system for misinformation moderation. In Twitter's new platform, Community Notes, misinformation tagging was peer-reviewed by other fact-checkers before exposure to the poster. With collective misinformation tagging, posters were less likely to retreat from diverse information engagement. Detailed comparison suggests differences in toxicity, sentiment, readability, and delay in individual versus collective misinformation tagging messages. These findings provide evidence for differential impacts from individual versus collective moderation strategies on the diversity of information engagement and mobility across the information ecosystem., Comment: 68 pages

We extend the existence of ghost measures beyond nonnegative primitive regular sequences to a large class of nonnegative real-valued regular sequences. In the general case, where the ghost measure is not unique, we show that they can be parametrised by a compact abelian group. For a subclass of these measures, by replacing primitivity with a commutativity condition, we show that these measures have an infinite convolution structure similar to Bernoulli convolutions. Using this structure, we show that these ghost measures have pure spectral type. Further, we provide results towards a classification of the spectral type based on inequalities involving the spectral radius, joint spectral radius, and Lyapunov exponent of the underlying set of matrices. In the case that the underlying measure is pure point, we show that the support of the measure must be a subset of the rational numbers, a result that resolves a new case of the finiteness conjecture., Comment: 36 pages

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., Comment: 35 pages, first draft of the final report from the Alan Turing Institute on AI for Scientific Discovery

Artificial intelligence (AI) models trained on published scientific findings have been used to invent valuable materials and targeted therapies, but they typically ignore the human scientists who continually alter the landscape of discovery. Here we show that incorporating the distribution of human expertise by training unsupervised models on simulated inferences cognitively accessible to experts dramatically improves (up to 400%) AI prediction of future discoveries beyond those focused on research content alone, especially when relevant literature is sparse. These models succeed by predicting human predictions and the scientists who will make them. By tuning human-aware AI to avoid the crowd, we can generate scientifically promising "alien" hypotheses unlikely to be imagined or pursued without intervention until the distant future, which hold promise to punctuate scientific advance beyond questions currently pursued. Accelerating human discovery or probing its blind spots, human-aware AI enables us to move toward and beyond the contemporary scientific frontier.

Artificial Intelligence (AI) has become a disruptive technology, promising to grant a significant economic and strategic advantage to the 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 the two nations, a number of policies have been put in place that discourage AI scientists from migrating to, or collaborating with, the other country. However, the extents of such brain drain 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 the year 2000, China and the U.S. have been leading the field in terms of impact, novelty, productivity, and workforce. Most AI scientists who migrate to China come from the U.S., and most who migrate to the U.S. come from China, highlighting a notable brain drain in both directions. Upon migrating from one country to the other, scientists continue to collaborate frequently with the origin country. Although the number of collaborations between the two countries has been increasing 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 of them works without the other. These findings suggest that instead of suppressing cross-border migration and collaboration between the two nations, the field could benefit from promoting such activities., Comment: 23 pages, 5 figures, 4 tables

Market bubbles emerge when asset prices are driven unsustainably higher than asset values and shifts in belief burst them. We demonstrate the same phenomenon for biomedical knowledge when promising research receives inflated attention. We predict deflationary events by developing a diffusion index that captures whether research areas have been amplified within social and scientific bubbles or have diffused and become evaluated more broadly. We illustrate our diffusion approach contrasting the trajectories of cardiac stem cell research and cancer immunotherapy. We then trace the diffusion of unique 28,504 subfields in biomedicine comprising nearly 1.9M papers and more than 80M citations and demonstrate that limited diffusion of biomedical knowledge anticipates abrupt decreases in popularity. Our analysis emphasizes that restricted diffusion, implying a socio-epistemic bubble, leads to dramatic collapses in relevance and attention accorded to scientific knowledge.

Prior research suggests that taste preferences relate to personality traits, values, shifts in mood, and immigration destination, but understanding everyday patterns of listening and the function music plays in life have remained elusive, despite speculations that musical nostalgia may compensate for local disruption. Using more than a hundred million streams of 4 million songs by tens of thousands of international listeners from a global music service catering to local tastes, here we show that breaches in personal routine are systematically associated with personal musical exploration. As people visited new cities and countries, their preferences diversified, converging towards their destinations. As people experienced COVID-19 lock-downs, and then again when they experienced reopenings, their preferences diversified further.

Innovation or the creation and diffusion of new material, social and cultural things in society has been widely studied in sociology and across the social sciences, with investigations sufficiently diverse and dispersed to make them unnavigable. This complexity results from innovation's importance for society, but also the fundamental paradox underlying innovation science: When innovation becomes predictable, it ceases to be an engine of novelty and change. Here we review innovation studies and show that innovations emerge from contexts of discord and disorder, breaches in the structure of prior success, through a process we term destructive creation. This often leads to a complementary process of creative destruction whereby local structures protect and channel the diffusion of successful innovations, rendering alternatives obsolete. We find that social scientists naturally focus far more on how social and cultural contexts influence material innovations than the converse. We highlight computational tools that open new possibilities for the analysis of novel content and context in interaction, and show how this brings us empirically toward the broader range of possibilities that complex systems and science studies have theorized-and science fiction has imagined-the social, cultural and material structures of innovation conditioning each other's change through cycles of disruption and development.

In this study, we present a method of pattern mining based on network theory that enables the identification of protein structures or complexes from synthetic volume densities, without the knowledge of predefined templates or human biases for refinement. We hypothesized that the topological connectivity of protein structures is invariant, and they are distinctive for the purpose of protein identification from distorted data presented in volume densities. Three-dimensional densities of a protein or a complex from simulated tomographic volumes were transformed into mathematical graphs as observables. We systematically introduced data distortion or defects such as missing fullness of data, the tumbling effect, and the missing wedge effect into the simulated volumes, and varied the distance cutoffs in pixels to capture the varying connectivity between the density cluster centroids in the presence of defects. A similarity score between the graphs from the simulated volumes and the graphs transformed from the physical protein structures in point data was calculated by comparing their network theory order parameters including node degrees, betweenness centrality, and graph densities. By capturing the essential topological features defining the heterogeneous morphologies of a network, we were able to accurately identify proteins and homo-multimeric complexes from ten topologically distinctive samples without realistic noise added. Our approach empowers future developments of tomogram processing by providing pattern mining with interpretability, to enable the classification of single-domain protein native topologies as well as distinct single-domain proteins from multimeric complexes within noisy volumes., Comment: submitted for peer review

This paper provides a detailed description of the data collection and machine learning model used in our recent PNAS paper "Flat Teams Drive Scientific Innovation" Xu et al. [2022a]. Here, we discuss how the features of scientific publication can be used to estimate the implicit hierarchy in the corresponding author teams. Besides, we also describe the method of evaluating the impact of team hierarchy on scientific outputs. More details will be updated in this article continuously. Raw data and Readme document can be accessed in this GitHub repository Xu et al. [2022b].

Neither artificial intelligence designed to play Turing's imitation game, nor augmented intelligence built to maximize the human manipulation of information are tuned to accelerate innovation and improve humanity's collective advance against its greatest challenges. We reconceptualize and pilot beneficial AI to radically augment human understanding by complementing rather than competing with human cognitive capacity. Our approach to complementary intelligence builds on insights underlying the wisdom of crowds, which hinges on the independence and diversity of crowd members' information and approach. By programmatically incorporating information on the evolving distribution of scientific expertise from research papers, our approach follows the distribution of content in the literature while avoiding the scientific crowd and the hypotheses cognitively available to it. We use this approach to generate valuable predictions for what materials possess valuable energy-related properties (e.g., thermoelectricity), and what compounds possess valuable medical properties (e.g., asthma) that complement the human scientific crowd. We demonstrate that our complementary predictions, if identified by human scientists and inventors at all, are only discovered years further into the future. When we evaluate the promise of our predictions with first-principles equations, we demonstrate that increased complementarity of our predictions does not decrease and in some cases increases the probability that the predictions possess the targeted properties. In summary, by tuning AI to avoid the crowd, we can generate hypotheses unlikely to be imagined or pursued until the distant future and promise to punctuate 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.

A general method was developed to intercalate metals under layered materials through a controlled density of sputtered defects. The method has been already applied to study a range of metals intercalated under graphite and different types of morphologies were realized. In the current work, we extend the method to the study of intercalation under MoS2 noting that work on this system is rather limited. We use Cu as the prototype metal for comparison with Cu intercalation under graphite. Although the growth conditions needed for intercalation under graphite and MoS2 are similar, the type of intercalated phases is very different. Each Cu island which nucleates on top of MoS2 during Cu deposition provides material that is transferred below MoS2, through sputtered defects under the island base; this transfer results in a uniform intercalated Cu "carpet" morphology that extends over the mesoscale. On the contrary, Cu intercalation under graphite results in well separated, compact islands formed by monomer detachment from small Cu islands on top and transfer below through defects far from the islands. The structural techniques (scanning electron microscopy and atomic force microscopy) and spectroscopic techniques (x-ray photoelectron spectroscopy and energy dispersive spectroscopy) are used for the characterization of the intercalated Cu layer.

What is the relationship between aging and the character of scientific advance? Prior research focuses on star scientists, their changing dates, and rates of breakthrough success through history. Analyzing more than 244 million scholars across 241 million articles over the last two centuries, we show that for all fields, periods, and impact levels, scientists research ideas and references age over time, their research is less likely to disrupt the state of science and more likely to criticize emerging work. Early success accelerates scientist aging; while changing institutions and fields and collaborating with young scientists slows it. These patterns aggregate within fields such that those with a higher proportion of older scientists experience a lower churn of ideas and more rapid individual aging, suggesting a universal link between aging, activity, and advance., Comment: 37 pages, 18 figures

With teams growing in all areas of scientific and scholarly research, we explore the relationship between team structure and the character of knowledge they produce. Drawing on 89,575 self-reports of team member research activity underlying scientific publications, we show how individual activities cohere into broad roles of (1) leadership through the direction and presentation of research and (2) support through data collection, analysis and discussion. The hidden hierarchy of a scientific team is characterized by its lead (or L)-ratio of members playing leadership roles to total team size. The L-ratio is validated through correlation with imputed contributions to the specific paper and to science as a whole, which we use to effectively extrapolate the L-ratio for 16,397,750 papers where roles are not explicit. We find that relative to flat, egalitarian teams, tall, hierarchical teams produce less novelty and more often develop existing ideas; increase productivity for those on top and decrease it for those beneath; increase short-term citations but decrease long-term influence. These effects hold within-person -- the same person on the same-sized team produces science much more likely to disruptively innovate if they work on a flat, high L-ratio team. These results suggest the critical role flat teams play for sustainable scientific advance and the training and advancement of scientists.

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.

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., Comment: Nat Hum Behav (2024)

Historical processes manifest remarkable diversity. Nevertheless, scholars have long attempted to identify patterns and categorize historical actors and influences with some success. A stochastic process framework provides a structured approach for the analysis of large historical datasets that allows for detection of sometimes surprising patterns, identification of relevant causal actors both endogenous and exogenous to the process, and comparison between different historical cases. The combination of data, analytical tools and the organizing theoretical framework of stochastic processes complements traditional narrative approaches in history and archaeology., Comment: 20 pages, 4 figures

The logic of abduction involves a collision between deduction and induction, where empirical surprises violate expectations and scientists innovate to resolve them. Here we reformulate abduction as a social process, occurring not only within individual scientists, but often through conversation between those who understand a particular scientific system and its anomalies (its insiders) and those exposed to alien and disruptive patterns, theories, and findings that could resolve them (its outsiders). These extended conversations between scientists, scholars, and disciplines with divergent backgrounds define a social logic of discovery by catalyzing social syllogisms that yield speculative hypotheses with outsized impact across science. We show how this approach theorizes a number of disparate findings from scientometrics and the science and science, tethering them to powerful interpersonal processes that link diversity with discovery. By reframing abductive reasoning as a social process, we attempt to broker a productive new relationship between science studies' focus on scientific practice and the philosophy of science's concern with propositions in scientific discourse, rendering claims-making through abduction a critical scientific practice, central to scientific change, and increasingly available to science studies through small and large-scale data on scientific communication and deliberation.