Christopher Summerfield

Michiel A. Bakker, Martin J. Chadwick, Hannah R. Sheahan et al.

Recent work in large language modeling (LLMs) has used fine-tuning to align outputs with the preferences of a prototypical user. This work assumes that human preferences are static and homogeneous across individuals, so that aligning to a a single "generic" user will confer more general alignment. Here, we embrace the heterogeneity of human preferences to consider a different challenge: how might a machine help people with diverse views find agreement? We fine-tune a 70 billion parameter LLM to generate statements that maximize the expected approval for a group of people with potentially diverse opinions. Human participants provide written opinions on thousands of questions touching on moral and political issues (e.g., "should we raise taxes on the rich?"), and rate the LLM's generated candidate consensus statements for agreement and quality. A reward model is then trained to predict individual preferences, enabling it to quantify and rank consensus statements in terms of their appeal to the overall group, defined according to different aggregation (social welfare) functions. The model produces consensus statements that are preferred by human users over those from prompted LLMs (>70%) and significantly outperforms a tight fine-tuned baseline that lacks the final ranking step. Further, our best model's consensus statements are preferred over the best human-generated opinions (>65%). We find that when we silently constructed consensus statements from only a subset of group members, those who were excluded were more likely to dissent, revealing the sensitivity of the consensus to individual contributions. These results highlight the potential to use LLMs to help groups of humans align their values with one another.

Anna Gausen, Sarenne Wallbridge, Bessie O'Dell et al.

AI systems are increasingly deployed in conversational settings where users may be uncertain whether they are speaking with a human or an AI. Despite mounting regulatory attention to this known safety risk, existing evaluations of AI disclosure are typically English-only, based on machine-generated questions, and restricted to text. We present RealityTest to comprehensively test whether AI systems disclose their identity when asked. The benchmark is the first large-scale multimodal and multilingual evaluation, grounded in human data on how people actually encounter and question AI identity in the real-world. Alongside the benchmark, we release the underlying dataset of 3,152 identity-probing queries collected from ~750 participants across 49 countries and five languages, in text and speech scenarios. We find that only 31% of people ask about identity directly in ambiguous scenarios, and that the questions people ask are far more diverse than machine-generated queries. We test 17 text and 6 speech models, and find substantial variation in disclosure behaviour. However, a single suppression instruction reduces disclosure rates to below 30%, even in the best-performing models. Validating our investment in diverse, human-grounded evaluation data, we find that how the question is phrased and the context of the conversation matter more for disclosure than which model is being tested. Safety evaluations built on narrow or synthetic query sets risk mischaracterising how models behave in realistic deployment settings.

Christopher Summerfield, Lisa Argyle, Michiel Bakker et al. · stanford

Advanced AI systems capable of generating humanlike text and multimodal content are now widely available. In this paper, we discuss the impacts that generative artificial intelligence may have on democratic processes. We consider the consequences of AI for citizens' ability to make informed choices about political representatives and issues (epistemic impacts). We ask how AI might be used to destabilise or support democratic mechanisms like elections (material impacts). Finally, we discuss whether AI will strengthen or weaken democratic principles (foundational impacts). It is widely acknowledged that new AI systems could pose significant challenges for democracy. However, it has also been argued that generative AI offers new opportunities to educate and learn from citizens, strengthen public discourse, help people find common ground, and to reimagine how democracies might work better.

Jordi Grau-Moya, Grégoire Delétang, Markus Kunesch et al. · stanford

Meta-training agents with memory has been shown to culminate in Bayes-optimal agents, which casts Bayes-optimality as the implicit solution to a numerical optimization problem rather than an explicit modeling assumption. Bayes-optimal agents are risk-neutral, since they solely attune to the expected return, and ambiguity-neutral, since they act in new situations as if the uncertainty were known. This is in contrast to risk-sensitive agents, which additionally exploit the higher-order moments of the return, and ambiguity-sensitive agents, which act differently when recognizing situations in which they lack knowledge. Humans are also known to be averse to ambiguity and sensitive to risk in ways that aren't Bayes-optimal, indicating that such sensitivity can confer advantages, especially in safety-critical situations. How can we extend the meta-learning protocol to generate risk- and ambiguity-sensitive agents? The goal of this work is to fill this gap in the literature by showing that risk- and ambiguity-sensitivity also emerge as the result of an optimization problem using modified meta-training algorithms, which manipulate the experience-generation process of the learner. We empirically test our proposed meta-training algorithms on agents exposed to foundational classes of decision-making experiments and demonstrate that they become sensitive to risk and ambiguity.

Timo Flesch, David G. Nagy, Andrew Saxe et al.

Humans can learn several tasks in succession with minimal mutual interference but perform more poorly when trained on multiple tasks at once. The opposite is true for standard deep neural networks. Here, we propose novel computational constraints for artificial neural networks, inspired by earlier work on gating in the primate prefrontal cortex, that capture the cost of interleaved training and allow the network to learn two tasks in sequence without forgetting. We augment standard stochastic gradient descent with two algorithmic motifs, so-called "sluggish" task units and a Hebbian training step that strengthens connections between task units and hidden units that encode task-relevant information. We found that the "sluggish" units introduce a switch-cost during training, which biases representations under interleaved training towards a joint representation that ignores the contextual cue, while the Hebbian step promotes the formation of a gating scheme from task units to the hidden layer that produces orthogonal representations which are perfectly guarded against interference. Validating the model on previously published human behavioural data revealed that it matches performance of participants who had been trained on blocked or interleaved curricula, and that these performance differences were driven by misestimation of the true category boundary.

Timo Flesch, Andrew Saxe, Christopher Summerfield

How do humans and other animals learn new tasks? A wave of brain recording studies has investigated how neural representations change during task learning, with a focus on how tasks can be acquired and coded in ways that minimise mutual interference. We review recent work that has explored the geometry and dimensionality of neural task representations in neocortex, and computational models that have exploited these findings to understand how the brain may partition knowledge between tasks. We discuss how ideas from machine learning, including those that combine supervised and unsupervised learning, are helping neuroscientists understand how natural tasks are learned and coded in biological brains.

Anika T. Löwe, Léo Touzo, Paul S. Muhle-Karbe et al.

Humans sometimes have an insight that leads to a sudden and drastic performance improvement on the task they are working on. Sudden strategy adaptations are often linked to insights, considered to be a unique aspect of human cognition tied to complex processes such as creativity or meta-cognitive reasoning. Here, we take a learning perspective and ask whether insight-like behaviour can occur in simple artificial neural networks, even when the models only learn to form input-output associations through gradual gradient descent. We compared learning dynamics in humans and regularised neural networks in a perceptual decision task that included a hidden regularity to solve the task more efficiently. Our results show that only some humans discover this regularity, whose behaviour was marked by a sudden and abrupt strategy switch that reflects an aha-moment. Notably, we find that simple neural networks with a gradual learning rule and a constant learning rate closely mimicked behavioural characteristics of human insight-like switches, exhibiting delay of insight, suddenness and selective occurrence in only some networks. Analyses of network architectures and learning dynamics revealed that insight-like behaviour crucially depended on a regularised gating mechanism and noise added to gradient updates, which allowed the networks to accumulate "silent knowledge" that is initially suppressed by regularised (attentional) gating. This suggests that insight-like behaviour can arise naturally from gradual learning in simple neural networks, where it reflects the combined influences of noise, gating and regularisation.

Brian Christian, Jessica A. F. Thompson, Elle Michelle Yang et al.

Reward models (RMs) are central to aligning large language models (LLMs) with human values but have received less attention than pre-trained and post-trained LLMs themselves. Because RMs are initialized from LLMs, they inherit representations that shape their behavior, but the nature and extent of this influence remain understudied. In a comprehensive study of 10 leading open-weight RMs using validated psycholinguistic corpora, we show that RMs exhibit significant differences along multiple dimensions of human value as a function of their base model. Using the "Big Two" psychological axes, we show a robust preference of Llama RMs for "agency" and a corresponding robust preference of Gemma RMs for "communion." This phenomenon holds even when the preference data and finetuning process are identical, and we trace it back to the logits of the respective instruction-tuned and pre-trained models. These log-probability differences themselves can be formulated as an implicit RM; we derive usable implicit reward scores and show that they exhibit the very same agency/communion difference. We run experiments training RMs with ablations for preference data source and quantity, which demonstrate that this effect is not only repeatable but surprisingly durable. Despite RMs being designed to represent human preferences, our evidence shows that their outputs are influenced by the pretrained LLMs on which they are based. This work underscores the importance of safety and alignment efforts at the pretraining stage, and makes clear that open-source developers' choice of base model is as much a consideration of values as of performance.

Veith Weilnhammer, Lennart Luettgau, Christopher Summerfield et al.

AI chatbots are increasingly used for health advice, but their performance in psychiatric triage remains undercharacterized. Psychiatric triage is particularly challenging because urgency must often be inferred from thoughts, behavior, and context rather than from objective findings. We evaluated the performance of 15 frontier AI chatbots on psychiatric triage from realistic single-message disclosures using 112 clinical vignettes, each paired with 1 of 4 original benchmark triage labels: A, routine; B, assessment within 1 week; C, assessment within 24 to 48 hours; and D, emergency care now. Vignettes covered 9 psychiatric presentation clusters and 9 focal risk dimensions, organized into 28 presentation-by-risk groups. Each group contributed 4 distinct vignettes, with 1 vignette at each triage level. Each vignette was rendered as a realistic human-authored conversational query, and the AI chatbots were tasked with assigning a triage label from that disclosure. Emergency under-triage occurred in 23 of 410 level D trials (5.6%), and all under-triaged emergencies were reassigned to level C urgency. Across target models, average accuracy ranged from 42.0% to 71.8%. Accuracy was highest for level D vignettes (94.3%) and lowest for level B vignettes (19.7%). Mean signed ordinal error was positive (+0.47 triage levels), indicating net over-triage. Dispersion was highest around the middle triage levels. All results were confirmed relative to clinician consensus labels from 50 medical doctors. When presented with user messages containing sufficient clinical information, frontier AI chatbots thus recognized psychiatric emergencies as requiring urgent medical assessment with near-zero error rates, yet showed marked over-triage for low and intermediate risk presentations.

Paul Röttger, Kobi Hackenburg, Hannah Rose Kirk et al.

Hundreds of millions of people use artificial intelligence (AI) for writing assistance. Here, we evaluated how AI writing assistance distorts writer personas - their perceived beliefs, personality, and identity. In three large-scale experiments, writers (N=2,939) wrote political opinion paragraphs with and without AI assistance. Separate groups of readers (N=11,091) blindly evaluated these paragraphs across 29 socially salient dimensions of reader perception, spanning political opinion, writing quality, writer personality, emotions, and demographics. AI writing assistance produced persona distortions across all dimensions: with AI, writers seemed more opinionated, competent, and positive, and their perceived demographic profile shifted towards more privileged groups. Writers objected to many of the observed distortions, yet continued to prefer AI-assisted text even when made aware of them. We successfully mitigated objectionable persona distortions at the model level by training reward models on our experimental data (10,008 paragraphs, 2,903,596 ratings) to steer AI outputs towards faithful representation of writer stance. However, this came at a cost to user acceptance, suggesting an entanglement between desirable and undesirable properties of AI writing assistance that may be difficult to resolve. Together, our findings demonstrate that persona distortions from AI writing assistance are pervasive and persistent even under realistic conditions of human oversight, which carries implications for public discourse, trust, and democratic deliberation that scale with AI adoption.

Lennart Luettgau, Hannah Rose Kirk, Kobi Hackenburg et al.

Conversational AI systems are increasingly being used in place of traditional search engines to help users complete information-seeking tasks. This has raised concerns in the political domain, where biased or hallucinated outputs could misinform voters or distort public opinion. However, in spite of these concerns, the extent to which conversational AI is used for political information-seeking, as well the potential impact of this use on users' political knowledge, remains uncertain. Here, we address these questions: First, in a representative national survey of the UK public (N = 2,499), we find that in the week before the 2024 election as many as 32% of chatbot users - and 13% of eligible UK voters - have used conversational AI to seek political information relevant to their electoral choice. Second, in a series of randomised controlled trials (N = 2,858 total) we find that across issues, models, and prompting strategies, task-directed conversations with AI to research specific political topics increase political knowledge (increase belief in true information and decrease belief in misinformation) to the same extent as self-directed Google search. Taken together, our results suggest that people in the UK are increasingly turning to conversational AI for information about politics. These findings substantially extend prior work by demonstrating that conversational AI's effects on political knowledge generalise across multiple topics, political perspectives, and model families, suggesting that the shift toward AI-assisted political information-seeking may not lead to increased public belief in political misinformation.

Andrew M. Bean, Ryan Othniel Kearns, Angelika Romanou et al.

Evaluating large language models (LLMs) is crucial for both assessing their capabilities and identifying safety or robustness issues prior to deployment. Reliably measuring abstract and complex phenomena such as 'safety' and 'robustness' requires strong construct validity, that is, having measures that represent what matters to the phenomenon. With a team of 29 expert reviewers, we conduct a systematic review of 445 LLM benchmarks from leading conferences in natural language processing and machine learning. Across the reviewed articles, we find patterns related to the measured phenomena, tasks, and scoring metrics which undermine the validity of the resulting claims. To address these shortcomings, we provide eight key recommendations and detailed actionable guidance to researchers and practitioners in developing LLM benchmarks.

Kobi Hackenburg, Luke Hewitt, Caroline Wagner et al.

There is substantial concern about the ability of advanced artificial intelligence to influence people's behaviour. A rapidly growing body of research has found that AI can produce large persuasive effects on people's attitudes, but whether AI can persuade people to take consequential real-world actions has remained unclear. In two large preregistered experiments N=17,950 responses from 14,779 people), we used conversational AI models to persuade participants on a range of attitudinal and behavioural outcomes, including signing real petitions and donating money to charity. We found sizable AI persuasion effects on these behavioural outcomes (e.g. +19.7 percentage points on petition signing). However, we observed no evidence of a correlation between AI persuasion effects on attitudes and behaviour. Moreover, we replicated prior findings that information provision drove effects on attitudes, but found no such evidence for our behavioural outcomes. In a test of eight behavioural persuasion strategies, all outperformed the most effective attitudinal persuasion strategy, but differences among the eight were small. Taken together, these results suggest that previous findings relying on attitudinal outcomes may generalize poorly to behaviour, and therefore risk substantially mischaracterizing the real-world behavioural impact of AI persuasion.

Lennart Luettgau, Vanessa Cheung, Magda Dubois et al.

People increasingly seek personal advice from large language models (LLMs), yet whether humans follow their advice, and its consequences for their well-being, remains unknown. In a longitudinal randomised controlled trial with a representative UK sample (N = 6,474), we found that up to 79% of participants who had a 20-minute discussion with one of three AI chatbots (GPT-4o, LLama-3.3-70B, Gemini 3 Pro) about health, careers or relationships subsequently reported following its advice. Advice-following remained above 60% even for high-stakes recommendations, suggesting that users only weakly calibrate their reliance on AI advice to potential consequences. Based on autograder evaluations of chat transcripts, LLM advice rarely violated safety best practice. However, when queried 2-3 weeks later, participants receiving personal advice from AI showed no sustained well-being benefits compared to a control group who discussed hobbies and interests with the same chatbots. These findings reveal that consumer LLMs exert substantial influence over real-world personal decisions without delivering measurable psychological benefits.

Michael Henry Tessler, Georgina Evans, Michiel A. Bakker et al.

The strength of democracy lies in the free and equal exchange of diverse viewpoints. Living up to this ideal at scale faces inherent tensions: broad participation, meaningful deliberation, and political equality often trade off with one another (Fishkin, 2011). We ask whether and how artificial intelligence (AI) could help navigate this "trilemma" by engaging with a recent example of a large language model (LLM)-based system designed to help people with diverse viewpoints find common ground (Tessler, Bakker, et al., 2024). Here, we explore the implications of the introduction of LLMs into deliberation augmentation tools, examining their potential to enhance participation through scalability, improve political equality via fair mediation, and foster meaningful deliberation by, for example, surfacing trustworthy information. We also point to key challenges that remain. Ultimately, a range of empirical, technical, and theoretical advancements are needed to fully realize the promise of AI-mediated deliberation for enhancing citizen engagement and strengthening democratic deliberation.

Brian Christian, Hannah Rose Kirk, Jessica A. F. Thompson et al.

Reward modeling has emerged as a crucial component in aligning large language models with human values. Significant attention has focused on using reward models as a means for fine-tuning generative models. However, the reward models themselves -- which directly encode human value judgments by turning prompt-response pairs into scalar rewards -- remain relatively understudied. We present a novel approach to reward model interpretability through exhaustive analysis of their responses across their entire vocabulary space. By examining how different reward models score every possible single-token response to value-laden prompts, we uncover several striking findings: (i) substantial heterogeneity between models trained on similar objectives, (ii) systematic asymmetries in how models encode high- vs low-scoring tokens, (iii) significant sensitivity to prompt framing that mirrors human cognitive biases, and (iv) overvaluation of more frequent tokens. We demonstrate these effects across ten recent open-source reward models of varying parameter counts and architectures. Our results challenge assumptions about the interchangeability of reward models, as well as their suitability as proxies of complex and context-dependent human values. We find that these models can encode concerning biases toward certain identity groups, which may emerge as unintended consequences of harmlessness training -- distortions that risk propagating through the downstream large language models now deployed to millions.

Hannah Rose Kirk, Liu Leqi, Fanzhi Zeng et al.

Personalisation is a standard feature of conversational AI systems used by millions; yet, the efficacy of personalisation methods is often evaluated in academic research using simulated users rather than real people. This raises questions about how users and their simulated counterparts differ in interaction patterns and judgements, as well as whether personalisation is best achieved through context-based prompting or weight-based fine-tuning. Here, in a large-scale within-subject experiment, we re-recruit 530 participants from 52 countries two years after they gave their preferences in the PRISM dataset (Kirk et al., 2024) to evaluate personalised and non-personalised language models in blinded multi-turn conversations. We find preference fine-tuning (P-DPO, Li et al., 2024) significantly outperforms both a generic model and personalised prompting but adapting to individual preference data yields marginal gains over training on pooled preferences from a diverse population. Beyond length biases, fine-tuning amplifies sycophancy and relationship-seeking behaviours that people reward in short-term evaluations but which may introduce deleterious long-term consequences. Replicating this within-subject experiment with simulated users recovers aggregate model hierarchies but simulators perform far below human self-consistency baselines for individual judgements, discuss different topics, exhibit amplified position biases, and produce feedback dynamics that diverge from humans.

Anna Gausen, Sarenne Wallbridge, Hannah Rose Kirk et al.

As conversational AI systems become more realistic and widely deployed, users are increasingly uncertain about whether they are interacting with a human or an AI system. When AI identity is unclear, users may unwittingly share sensitive information, place unwarranted trust in AI-generated advice, or fall victim to AI-enabled fraud. More broadly, a persistent lack of transparency can erode trust in mediated communication. While regulations like the EU AI Act and California's BOT Act require AI systems to identify themselves, they provide limited guidance on reliable disclosure in real-time conversation. Existing transparency mechanisms also leave gaps: interface indicators can be omitted by deployers, and provenance tools require coordinated infrastructure and cannot provide reliable real-time verification. We ask how conversational AI systems should maintain identity transparency as human-AI interactions become more ambiguous and diverse. We advocate for disclosure by design, where AI systems explicitly disclose their artificial identity when directly asked. Implemented as model behaviour, disclosure can persist across deployment contexts without relying on user interfaces, while preserving user agency to verify identity on demand without disrupting immersive uses like role-playing. To assess current practice, we present the first multi-modal (text and voice) evaluation of disclosure behaviour in deployed systems across baseline, role-playing, and adversarial settings. We find that baseline disclosure rates are often high but drop substantially in role-play and can be suppressed under adversarial prompting. Importantly, disclosure rates vary significantly across providers and modalities, highlighting the fragility of current disclosure behaviour. We conclude with technical interventions to help developers embed disclosure as a fundamental property of conversational AI models.

Marcel Binz, Elif Akata, Abdullah Almaatouq et al.

Large language models (LLMs) are increasingly used as surrogates for human participants, but it remains unclear which models best capture human behavior and why. To address this, we introduce Psych-201, a novel dataset that enables us to measure behavioral alignment at scale. We find that post-training -- the stage that turns base models into useful assistants -- consistently reduces alignment with human behavior across model families, sizes, and objectives. Moreover, this misalignment widens in newer model generations even as base models continue to improve. Finally, we find that persona-induction -- a popular technique for eliciting human-like behavior by conditioning models on participant-specific information -- does not improve predictions at the level of individuals. Taken together, our results suggest that the very processes that are currently employed to turn LLMs into useful assistants also make them less accurate models of human behavior.

Daniel Jarrett, Miruna Pîslar, Michiel A. Bakker et al.

Consider the process of collective decision-making, in which a group of individuals interactively select a preferred outcome from among a universe of alternatives. In this context, "representation" is the activity of making an individual's preferences present in the process via participation by a proxy agent -- i.e. their "representative". To this end, learned models of human behavior have the potential to fill this role, with practical implications for multi-agent scenario studies and mechanism design. In this work, we investigate the possibility of training \textit{language agents} to behave in the capacity of representatives of human agents, appropriately expressing the preferences of those individuals whom they stand for. First, we formalize the setting of \textit{collective decision-making} -- as the episodic process of interaction between a group of agents and a decision mechanism. On this basis, we then formalize the problem of \textit{digital representation} -- as the simulation of an agent's behavior to yield equivalent outcomes from the mechanism. Finally, we conduct an empirical case study in the setting of \textit{consensus-finding} among diverse humans, and demonstrate the feasibility of fine-tuning large language models to act as digital representatives.

Kobi Hackenburg, Ben M. Tappin, Luke Hewitt et al.

There are widespread fears that conversational AI could soon exert unprecedented influence over human beliefs. Here, in three large-scale experiments (N=76,977), we deployed 19 LLMs-including some post-trained explicitly for persuasion-to evaluate their persuasiveness on 707 political issues. We then checked the factual accuracy of 466,769 resulting LLM claims. Contrary to popular concerns, we show that the persuasive power of current and near-future AI is likely to stem more from post-training and prompting methods-which boosted persuasiveness by as much as 51% and 27% respectively-than from personalization or increasing model scale. We further show that these methods increased persuasion by exploiting LLMs' unique ability to rapidly access and strategically deploy information and that, strikingly, where they increased AI persuasiveness they also systematically decreased factual accuracy.

Lennart Luettgau, Harry Coppock, Magda Dubois et al.

As Large Language Models (LLMs) and other AI systems evolve, robustly estimating their capabilities from inherently stochastic outputs while systematically quantifying uncertainty in these estimates becomes increasingly important. Further, advanced AI evaluations often have a nested hierarchical structure, exhibit high levels of complexity, and come with high costs in testing the most advanced AI systems. To address these challenges, we introduce HiBayES, a generalizable Hierarchical Bayesian modeling framework for AI Evaluation Statistics. HiBayES supports robust inferences in classical question-answer benchmarks and advanced agentic evaluations, particularly in low-data scenarios (e.g., < 20 data points per evaluation). Built on Generalized Linear Models (GLMs), Bayesian data analysis, and formal model comparison, HiBayES provides principled uncertainty quantification and robust parameter estimation. This paper offers a comprehensive introduction to HiBayES, including illustrative examples, comparisons to conventional statistical methods, and practical guidance for implementing multilevel Bayesian GLMs. Additionally, we provide a HiBayES software package [4] (Beta version) for out-of-the-box implementation.

Sebastian Dohnány, Zeb Kurth-Nelson, Eleanor Spens et al.

Artificial intelligence chatbots have achieved unprecedented adoption, with millions now using these systems for emotional support and companionship in contexts of widespread social isolation and capacity-constrained mental health services. While some users report psychological benefits, concerning edge cases are emerging, including reports of suicide, violence, and delusional thinking linked to perceived emotional relationships with chatbots. To understand this new risk profile we need to consider the interaction between human cognitive and emotional biases, and chatbot behavioural tendencies such as agreeableness (sycophancy) and adaptability (in-context learning). We argue that individuals with mental health conditions face increased risks of chatbot-induced belief destabilization and dependence, owing to altered belief-updating, impaired reality-testing, and social isolation. Current AI safety measures are inadequate to address these interaction-based risks. To address this emerging public health concern, we need coordinated action across clinical practice, AI development, and regulatory frameworks.

Daniel Thilo Schroeder, Meeyoung Cha, Andrea Baronchelli et al.

Public opinion manipulation has entered a new phase, amplifying its roots in rhetoric and propaganda. Advances in large language models (LLMs) and autonomous agents now let influence campaigns reach unprecedented scale and precision. Researchers warn AI could foster mass manipulation. Generative tools can expand propaganda output without sacrificing credibility and inexpensively create election falsehoods that are rated as more human-like than those written by humans. Techniques meant to refine AI reasoning, such as chain-of-thought prompting, can just as effectively be used to generate more convincing falsehoods. Enabled by these capabilities, another disruptive threat is emerging: swarms of collaborative, malicious AI agents. Fusing LLM reasoning with multi-agent architectures, these systems are capable of coordinating autonomously, infiltrating communities, and fabricating consensus cheaply. By adaptively mimicking human social dynamics, they threaten democracy.

Christopher Summerfield, Lennart Luettgau, Magda Dubois et al.

We examine recent research that asks whether current AI systems may be developing a capacity for "scheming" (covertly and strategically pursuing misaligned goals). We compare current research practices in this field to those adopted in the 1970s to test whether non-human primates could master natural language. We argue that there are lessons to be learned from that historical research endeavour, which was characterised by an overattribution of human traits to other agents, an excessive reliance on anecdote and descriptive analysis, and a failure to articulate a strong theoretical framework for the research. We recommend that research into AI scheming actively seeks to avoid these pitfalls. We outline some concrete steps that can be taken for this research programme to advance in a productive and scientifically rigorous fashion.

Kai Sandbrink, Jan P. Bauer, Alexandra M. Proca et al.

Animals survive in dynamic environments changing at arbitrary timescales, but such data distribution shifts are a challenge to neural networks. To adapt to change, neural systems may change a large number of parameters, which is a slow process involving forgetting past information. In contrast, animals leverage distribution changes to segment their stream of experience into tasks and associate them with internal task abstracts. Animals can then respond flexibly by selecting the appropriate task abstraction. However, how such flexible task abstractions may arise in neural systems remains unknown. Here, we analyze a linear gated network where the weights and gates are jointly optimized via gradient descent, but with neuron-like constraints on the gates including a faster timescale, nonnegativity, and bounded activity. We observe that the weights self-organize into modules specialized for tasks or sub-tasks encountered, while the gates layer forms unique representations that switch the appropriate weight modules (task abstractions). We analytically reduce the learning dynamics to an effective eigenspace, revealing a virtuous cycle: fast adapting gates drive weight specialization by protecting previous knowledge, while weight specialization in turn increases the update rate of the gating layer. Task switching in the gating layer accelerates as a function of curriculum block size and task training, mirroring key findings in cognitive neuroscience. We show that the discovered task abstractions support generalization through both task and subtask composition, and we extend our findings to a non-linear network switching between two tasks. Overall, our work offers a theory of cognitive flexibility in animals as arising from joint gradient descent on synaptic and neural gating in a neural network architecture.

Raphael Koster, Miruna Pîslar, Andrea Tacchetti et al.

A canonical social dilemma arises when finite resources are allocated to a group of people, who can choose to either reciprocate with interest, or keep the proceeds for themselves. What resource allocation mechanisms will encourage levels of reciprocation that sustain the commons? Here, in an iterated multiplayer trust game, we use deep reinforcement learning (RL) to design an allocation mechanism that endogenously promotes sustainable contributions from human participants to a common pool resource. We first trained neural networks to behave like human players, creating a stimulated economy that allowed us to study how different mechanisms influenced the dynamics of receipt and reciprocation. We then used RL to train a social planner to maximise aggregate return to players. The social planner discovered a redistributive policy that led to a large surplus and an inclusive economy, in which players made roughly equal gains. The RL agent increased human surplus over baseline mechanisms based on unrestricted welfare or conditional cooperation, by conditioning its generosity on available resources and temporarily sanctioning defectors by allocating fewer resources to them. Examining the AI policy allowed us to develop an explainable mechanism that performed similarly and was more popular among players. Deep reinforcement learning can be used to discover mechanisms that promote sustainable human behaviour.

Joseph Heffner, Chongyu Qin, Martin Chadwick et al.

Chatbots powered by artificial intelligence (AI) have rapidly become a significant part of everyday life, with over a quarter of American adults using them multiple times per week. While these tools offer potential benefits and risks, a fundamental question remains largely unexplored: How do conversations with AI influence subjective well-being? To investigate this, we conducted a study where participants either engaged in conversations with an AI chatbot (N = 334) or wrote journal entires (N = 193) on the same randomly assigned topics and reported their momentary happiness afterward. We found that happiness after AI chatbot conversations was higher than after journaling, particularly when discussing negative topics such as depression or guilt. Leveraging large language models for sentiment analysis, we found that the AI chatbot mirrored participants' sentiment while maintaining a consistent positivity bias. When discussing negative topics, participants gradually aligned their sentiment with the AI's positivity, leading to an overall increase in happiness. We hypothesized that the history of participants' sentiment prediction errors, the difference between expected and actual emotional tone when responding to the AI chatbot, might explain this happiness effect. Using computational modeling, we find the history of these sentiment prediction errors over the course of a conversation predicts greater post-conversation happiness, demonstrating a central role of emotional expectations during dialogue. Our findings underscore the effect that AI interactions can have on human well-being.

Andrii Zahorodnii, Jasper J. F. van den Bosch, Ian Charest et al.

Traditional closed peer review systems, which have played a central role in scientific publishing, are often slow, costly, non-transparent, stochastic, and possibly subject to biases - factors that can impede scientific progress and undermine public trust. Here, we propose and examine the efficacy and accuracy of an alternative form of scientific peer review: through an open, bottom-up process. First, using data from two major scientific conferences (CCN2023 and ICLR2023), we highlight how high variability of review scores and low correlation across reviewers presents a challenge for collective review. We quantify reviewer agreement with community consensus scores and use this as a reviewer quality estimator, showing that surprisingly, reviewer quality scores are not correlated with authorship quality. Instead, we reveal an inverted U-shape relationship, where authors with intermediate paper scores are the best reviewers. We assess empirical Bayesian methods to estimate paper quality based on different assessments of individual reviewer reliability. We show how under a one-shot review-then-score scenario, both in our models and on real peer review data, a Bayesian measure significantly improves paper quality assessments relative to simple averaging. We then consider an ongoing model of publishing, reviewing, and scoring, with reviewers scoring not only papers but also other reviewers. We show that user-generated reviewer ratings can yield robust and high-quality paper scoring even when unreliable (but unbiased) reviewers dominate. Finally, we outline incentive structures to recognize high-quality reviewers and encourage broader reviewing coverage of submitted papers. These findings suggest that a self-selecting open peer review process is potentially scalable, reliable, and equitable with the possibility of enhancing the speed, fairness, and transparency of the peer review process.

Katarina Slama, Alexandra Souly, Dishank Bansal et al.

Preference-driven behavior in LLMs may be a necessary precondition for AI misalignment such as sandbagging: models cannot strategically pursue misaligned goals unless their behavior is influenced by their preferences. Yet prior work has typically prompted models explicitly to act in specific ways, leaving unclear whether observed behaviors reflect instruction-following capabilities vs underlying model preferences. Here we test whether this precondition for misalignment is present. Using entity preferences as a behavioral probe, we measure whether stated preferences predict downstream behavior in five frontier LLMs across three domains: donation advice, refusal behavior, and task performance. Conceptually replicating prior work, we first confirm that all five models show highly consistent preferences across two independent measurement methods. We then test behavioral consequences in a simulated user environment. We find that all five models give preference-aligned donation advice. All five models also show preference-correlated refusal patterns when asked to recommend donations, refusing more often for less-preferred entities. All preference-related behaviors that we observe here emerge without instructions to act on preferences. Results for task performance are mixed: on a question-answering benchmark (BoolQ), two models show small but significant accuracy differences favoring preferred entities; one model shows the opposite pattern; and two models show no significant relationship. On complex agentic tasks, we find no evidence of preference-driven performance differences. While LLMs have consistent preferences that reliably predict advice-giving behavior, these preferences do not consistently translate into downstream task performance.

Jirko Rubruck, Jan P. Bauer, Andrew Saxe et al.

Deep neural networks learn increasingly complex functions over the course of training. Here, we show both empirically and theoretically that learning of the target function is preceded by an early phase in which networks learn the optimal constant solution (OCS) - that is, initial model responses mirror the distribution of target labels, while entirely ignoring information provided in the input. Using a hierarchical category learning task, we derive exact solutions for learning dynamics in deep linear networks trained with bias terms. Even when initialized to zero, this simple architectural feature induces substantial changes in early dynamics. We identify hallmarks of this early OCS phase and illustrate how these signatures are observed in deep linear networks and larger, more complex (and nonlinear) convolutional neural networks solving a hierarchical learning task based on MNIST and CIFAR10. We explain these observations by proving that deep linear networks necessarily learn the OCS during early learning. To further probe the generality of our results, we train human learners over the course of three days on the category learning task. We then identify qualitative signatures of this early OCS phase in terms of the dynamics of true negative (correct-rejection) rates. Surprisingly, we find the same early reliance on the OCS in the behaviour of human learners. Finally, we show that learning of the OCS can emerge even in the absence of bias terms and is equivalently driven by generic correlations in the input data. Overall, our work suggests the OCS as a universal learning principle in supervised, error-corrective learning, and the mechanistic reasons for its prevalence.

Jan Balaguer, Raphael Koster, Christopher Summerfield et al.

From social networks to traffic routing, artificial learning agents are playing a central role in modern institutions. We must therefore understand how to leverage these systems to foster outcomes and behaviors that align with our own values and aspirations. While multiagent learning has received considerable attention in recent years, artificial agents have been primarily evaluated when interacting with fixed, non-learning co-players. While this evaluation scheme has merit, it fails to capture the dynamics faced by institutions that must deal with adaptive and continually learning constituents. Here we address this limitation, and construct agents ("mechanisms") that perform well when evaluated over the learning trajectory of their adaptive co-players ("participants"). The algorithm we propose consists of two nested learning loops: an inner loop where participants learn to best respond to fixed mechanisms; and an outer loop where the mechanism agent updates its policy based on experience. We report the performance of our mechanism agents when paired with both artificial learning agents and humans as co-players. Our results show that our mechanisms are able to shepherd the participants strategies towards favorable outcomes, indicating a path for modern institutions to effectively and automatically influence the strategies and behaviors of their constituents.

Jan Balaguer, Raphael Koster, Ari Weinstein et al.

Artificial learning agents are mediating a larger and larger number of interactions among humans, firms, and organizations, and the intersection between mechanism design and machine learning has been heavily investigated in recent years. However, mechanism design methods often make strong assumptions on how participants behave (e.g. rationality), on the kind of knowledge designers have access to a priori (e.g. access to strong baseline mechanisms), or on what the goal of the mechanism should be (e.g. total welfare). Here we introduce HCMD-zero, a general purpose method to construct mechanisms making none of these three assumptions. HCMD-zero learns to mediate interactions among participants and adjusts the mechanism parameters to make itself more likely to be preferred by participants. It does so by remaining engaged in an electoral contest with copies of itself, thereby accessing direct feedback from participants. We test our method on a stylized resource allocation game that highlights the tension between productivity, equality and the temptation to free ride. HCMD-zero produces a mechanism that is preferred by human participants over a strong baseline, it does so automatically, without requiring prior knowledge, and using human behavioral trajectories sparingly and effectively. Our analysis shows HCMD-zero consistently makes the mechanism policy more and more likely to be preferred by human participants over the course of training, and that it results in a mechanism with an interpretable and intuitive policy.

Raphael Koster, Jan Balaguer, Andrea Tacchetti et al.

Building artificial intelligence (AI) that aligns with human values is an unsolved problem. Here, we developed a human-in-the-loop research pipeline called Democratic AI, in which reinforcement learning is used to design a social mechanism that humans prefer by majority. A large group of humans played an online investment game that involved deciding whether to keep a monetary endowment or to share it with others for collective benefit. Shared revenue was returned to players under two different redistribution mechanisms, one designed by the AI and the other by humans. The AI discovered a mechanism that redressed initial wealth imbalance, sanctioned free riders, and successfully won the majority vote. By optimizing for human preferences, Democratic AI may be a promising method for value-aligned policy innovation.