Search

Insights from biomedical citation networks can be used to identify promising avenues for accelerating research and its downstream bench-to- bedside translation. Citation analysis generally assumes that each citation documents substantive knowledge transfer that informed the conception, design, or execution of the main experiments. Citations may exist for other reasons. In this paper, we take advantage of late-stage citations added during peer review because these are less likely to represent substantive knowledge flow. Using a large, comprehensive feature set of open access data, we train a predictive model to identify late-stage citations. The model relies only on the title, abstract, and citations to previous articles but not the full-text or future citations patterns, making it suitable for publications as soon as they are released, or those behind a paywall (the vast majority). We find that high prediction scores identify late-stage citations that were likely added during the peer review process as well as those more likely to be rhetorical, such as journal self-citations added during review. Our model conversely gives low prediction scores to early-stage citations and citation classes that are known to represent substantive knowledge transfer. Using this model, we find that US federally funded biomedical research publications represent 30% of the predicted early-stage (and more likely to be substantive) knowledge transfer from basic studies to clinical research, even though these comprise only 10% of the literature. This is a threefold overrepresentation in this important type of knowledge flow. [ABSTRACT FROM AUTHOR]

This document is a reply to a comment on a recent paper that investigated the relationship between race and language markers of depression. The comment raised two critiques of the prediction analysis. The authors of the paper provided details of their analyses, including sample sizes, validation metrics, and training and test set sizes. They also discussed the possibility that the language of Black individuals explains less of the variance in depression than the language of White individuals. The authors used a specific language model called BERT to evaluate the predictive performance of different feature sets. They acknowledged the suggestion to use a different model called RoBERTa but explained their reasons for prioritizing interpretable feature sets. The authors also noted that all models performed relatively poorly on data from Black individuals compared to White individuals. [Extracted from the article]

The notion of common sense is invoked so frequently in contexts as diverse as everyday conversation, political debates, and evaluations of artificial intelligence that its meaning might be surmised to be unproblematic. Surprisingly, however, neither the intrinsic properties of common sense knowledge (what makes a claim commonsensical) nor the degree to which it is shared by people (its "commonness") have been characterized empirically. In this paper, we introduce an analytical framework for quantifying both these elements of common sense. First, we define the commonsensicality of individual claims and people in terms of the latter's propensity to agree on the former and their awareness of one another's agreement. Second, we formalize the commonness of common sense as a clique detection problem on a bipartite belief graph of people and claims, defining [Formula: see text] common sense as the fraction [Formula: see text] of claims shared by a fraction [Formula: see text] of people. Evaluating our framework on a dataset of [Formula: see text] raters evaluating [Formula: see text] diverse claims, we find that commonsensicality aligns most closely with plainly worded, fact-like statements about everyday physical reality. Psychometric attributes such as social perceptiveness influence individual common sense, but surprisingly demographic factors such as age or gender do not. Finally, we find that collective common sense is rare: At most, a small fraction [Formula: see text] of people agree on more than a small fraction [Formula: see text] of claims. Together, these results undercut universalistic beliefs about common sense and raise questions about its variability that are relevant both to human and artificial intelligence., Competing Interests: Competing interests statement:The authors declare no competing interest.

Over the past few years, machine learning models have significantly increased in size and complexity, especially in the area of generative AI such as large language models. These models require massive amounts of data andcomputecapacitytotrain,totheextentthatconcerns over the training data (such as protected or private content)cannotbepracticallyaddressedbyretrainingthe model"fromscratch"withthequestionabledataremoved or altered. Furthermore, despite significant efforts and controls dedicated to ensuring that training corpora are properly curated and composed, the sheer volume re- quiredmakesitinfeasibletomanuallyinspecteachdatum comprising a training corpus. One potential approach to training corpus data defects is model disgorgement, by which we broadly mean the elimination or reduction of not only any improperly used data, but also the effects of improperly used data on any component of an ML model. Model disgorgement techniques can be used to address a wide range of issues, such as reducing bias or toxicity, increasing fidelity, and ensuring responsible use of intellectual property. In this paper, we survey the land- scape of model disgorgement methods and introduce a taxonomyofdisgorgementtechniquesthatareapplicable to modern ML systems. In particular, we investigate the variousmeaningsof"removingtheeffects"ofdataonthe trained model in a way that does not require retraining from scratch [ABSTRACT FROM AUTHOR]

This article explores the emergence of AI as a new frontier in behavioral science. It discusses a study that compares the decision-making of AI chatbots, specifically ChatGPT-3 and ChatGPT-4, with that of humans using classical behavioral assessments. The study reveals that AI models have distinct preferences and behaviors, leading to the development of a new research direction called "AI Behavioral Science." The article also examines the potential benefits and challenges of studying AI behavior, including its impact on human decision-making, bias correction, and policy design. It emphasizes the need for a comprehensive behavioral assessment framework and the engineering of AI behavior. Furthermore, the article discusses the potential impact of AI integration on human behavior and culture, such as algorithmic bias, cognitive decline, and the promotion of equality. The document includes references to various research papers and studies on biased programmers, biased data, AI ethics, recommender systems, social media, news consumption, and polarization, making it a valuable resource for library patrons researching these specific topics. [Extracted from the article]

This paper introduces the paradigm of "in-context operator learning" and the corresponding model "In-Context Operator Networks" to simultaneously learn operators from the prompted data and apply it to new questions during the inference stage, without any weight update. Existing methods are limited to using a neural network to approximate a specific equation solution or a specific operator, requiring retraining when switching to a new problem with different equations. By training a single neural network as an operator learner, rather than a solution/operator approximator, we can not only get rid of retraining (even fine-tuning) the neural network for new problems but also leverage the commonalities shared across operators so that only a few examples in the prompt are needed when learning a new operator. Our numerical results show the capability of a single neural network as a few-shot operator learner for a diversified type of differential equation problems, including forward and inverse problems of ordinary differential equations, partial differential equations, and mean-field control problems, and also show that it can generalize its learning capability to operators beyond the training distribution. [ABSTRACT FROM AUTHOR]

This paper examines consciousness from the perspective of theoretical computer science (TCS), a branch of mathematics concerned with understanding the underlying principles of computation and complexity, including the implications and surprising consequences of resource limitations. We propose a formal TCS model, the Conscious Turing Machine (CTM). The CTM is influenced by Alan Turing's simple yet powerful model of computation, the Turing machine (TM), and by the global workspace theory (GWT) of consciousness originated by cognitive neuroscientist Bernard Baars and further developed by him, Stanislas Dehaene, Jean-Pierre Changeux, George Mashour, and others. Phenomena generally associated with consciousness, such as blindsight, inattentional blindness, change blindness, dream creation, and free will, are considered. Explanations derived from the model draw confirmation from consistencies at a high level, well above the level of neurons, with the cognitive neuroscience literature. [ABSTRACT FROM AUTHOR]

This document is a correction notice for an article titled "Misinformation in and about science" by Jevin D. West and Carl T. Bergstrom. The authors acknowledge an error in the text regarding the reporting of controversial work on detecting sexual orientation from photographs using artificial intelligence. The correction states that the Wang and Kosinski preprint was posted to the PsyArXiv preprint repository concurrently with the publication of the Guardian article and prior to the publication of the Economist article. The online version of the article has been corrected. The document is distributed under a Creative Commons Attribution-NonCommercial-NoDerivatives License 4.0. [Extracted from the article]

A study published in the Proceedings of the National Academy of Sciences of the United States of America examines whether thinking about God increases acceptance of artificial intelligence (AI) in decision-making. The original study by Karataş and Cutright suggests that "God salience" leads to greater reliance on AI, regardless of religious beliefs. However, a replication study conducted by Moore et al. with a larger sample size found weaker results, indicating that the original study may have lacked statistical power. The replication study suggests that the evidence for the relationship between thinking about God and reliance on AI is less compelling than initially claimed. [Extracted from the article]

Mutations in protein active sites can dramatically improve function. The active site, however, is densely packed and extremely sensitive to mutations. Therefore, some mutations may only be tolerated in combination with others in a phenomenon known as epistasis. Epistasis reduces the likelihood of obtaining improved functional variants and dramatically slows natural and lab evolutionary processes. Research has shed light on the molecular origins of epistasis and its role in shaping evolutionary trajectories and outcomes. In addition, sequence-and AI-based strategies that infer epistatic relationships from mutational patterns in natural or experimental evolution data have been used to design functional protein variants. In recent years, combinations of such approaches and atomistic design calculations have successfully predicted highly functional combinatorial mutations in active sites. These were used to design thousands of functional active-site variants, demonstrating that, while our understanding of epistasis remains incomplete, some of the determinants that are critical for accurate design are now sufficiently understood. We conclude that the space of active-site variants that has been explored by evolution may be expanded dramatically to enhance natural activities or discover new ones. Furthermore, design opens the way to systematically exploring sequence and structure space and mutational impacts on function, deepening our understanding and control over protein activity. [ABSTRACT FROM AUTHOR]

AI is now an integral part of everyday decision-making, assisting us in both routine and high-stakes choices. These AI models often learn from human behavior, assuming this training data is unbiased. However, we report five studies that show that people change their behavior to instill desired routines into AI, indicating this assumption is invalid. To show this behavioral shift, we recruited participants to play the ultimatum game, where they were asked to decide whether to accept proposals of monetary splits made by either other human participants or AI. Some participants were informed their choices would be used to train an AI proposer, while others did not receive this information. Across five experiments, we found that people modified their behavior to train AI to make fair proposals, regardless of whether they could directly benefit from the AI training. After completing this task once, participants were invited to complete this task again but were told their responses would not be used for AI training. People who had previously trained AI persisted with this behavioral shift, indicating that the new behavioral routine had become habitual. This work demonstrates that using human behavior as training data has more consequences than previously thought since it can engender AI to perpetuate human biases and cause people to form habits that deviate from how they would normally act. Therefore, this work underscores a problem for AI algorithms that aim to learn unbiased representations of human preferences. [ABSTRACT FROM AUTHOR]

Secreted signaling peptides are central regulators of growth, development, and stress responses, but specific steps in the evolution of these peptides and their receptors are not well understood. Also, the molecular mechanisms of peptide-receptor binding are only known for a few examples, primarily owing to the limited availability of protein structural determination capabilities to few laboratories worldwide. Plants have evolved a multitude of secreted signaling peptides and corresponding transmembrane receptors. Stress-responsive SERINE RICH ENDOGENOUS PEPTIDES (SCOOPs) were recently identified. Bioactive SCOOPs are proteolytically processed by subtilases and are perceived by the leucine-rich repeat receptor kinase MALE DISCOVERER 1-INTERACTING RECEPTOR-LIKE KINASE 2 (MIK2) in the model plant Arabidopsis thaliana. How SCOOPs and MIK2 have (co)evolved, and how SCOOPs bind to MIK2 are unknown. Using in silico analysis of 350 plant genomes and subsequent functional testing, we revealed the conservation of MIK2 as SCOOP receptor within the plant order Brassicales. We then leveraged AI-based structural modeling and comparative genomics to identify two conserved putative SCOOP-MIK2 binding pockets across Brassicales MIK2 homologues predicted to interact with the "SxS" motif of otherwise sequence-divergent SCOOPs. Mutagenesis of both predicted binding pockets compromised SCOOP binding to MIK2, SCOOP-induced complex formation between MIK2 and its coreceptor BRASSINOSTEROID INSENSITIVE 1-ASSOCIATED KINASE 1, and SCOOP-induced reactive oxygen species production, thus, confirming our in silico predictions. Collectively, in addition to revealing the elusive SCOOP-MIK2 binding mechanism, our analytic pipeline combining phylogenomics, AI-based structural predictions, and experimental biochemical and physiological validation provides a blueprint for the elucidation of peptide ligand-receptor perception mechanisms. [ABSTRACT FROM AUTHOR]

Understanding how neural circuits generate sequential activity is a longstanding challenge. While foundational theoretical models have shown how sequences can be stored as memories in neural networks with Hebbian plasticity rules, these models considered only a narrow range of Hebbian rules. Here, we introduce a model for arbitrary Hebbian plasticity rules, capturing the diversity of spike-timing-dependent synaptic plasticity seen in experiments, and show how the choice of these rules and of neural activity patterns influences sequence memory formation and retrieval. In particular, we derive a general theory that predicts the tempo of sequence replay. This theory lays a foundation for explaining how cortical tutor signals might give rise to motor actions that eventually become "automatic." Our theory also captures the impact of changing the tempo of the tutor signal. Beyond shedding light on biological circuits, this theory has relevance in artificial intelligence by laying a foundation for frameworks whereby slow and computationally expensive deliberation can be stored as memories and eventually replaced by inexpensive recall. [ABSTRACT FROM AUTHOR]

Humans and animals excel at generalizing from limited data, a capability yet to be fully replicated in artificial intelligence. This perspective investigates generalization in biological and artificial deep neural networks (DNNs), in both in-distribution and out-of-distribution contexts. We introduce two hypotheses: First, the geometric properties of the neural manifolds associated with discrete cognitive entities, such as objects, words, and concepts, are powerful order parameters. They link the neural substrate to the generalization capabilities and provide a unified methodology bridging gaps between neuroscience, machine learning, and cognitive science. We overview recent progress in studying the geometry of neural manifolds, particularly in visual object recognition, and discuss theories connecting manifold dimension and radius to generalization capacity. Second, we suggest that the theory of learning in wide DNNs, especially in the thermodynamic limit, provides mechanistic insights into the learning processes generating desired neural representational geometries and generalization. This includes the role of weight norm regularization, network architecture, and hyper-parameters. We will explore recent advances in this theory and ongoing challenges. We also discuss the dynamics of learning and its relevance to the issue of representational drift in the brain. [ABSTRACT FROM AUTHOR]

This paper introduces the paradigm of in-context operator learning and the corresponding model In-Context Operator Networks to simultaneously learn operators from the prompted data and apply it to new questions during the inference stage, without any weight update. Existing methods are limited to using a neural network to approximate a specific equation solution or a specific operator, requiring retraining when switching to a new problem with different equations. By training a single neural network as an operator learner, rather than a solution/operator approximator, we can not only get rid of retraining (even fine-tuning) the neural network for new problems but also leverage the commonalities shared across operators so that only a few examples in the prompt are needed when learning a new operator. Our numerical results show the capability of a single neural network as a few-shot operator learner for a diversified type of differential equation problems, including forward and inverse problems of ordinary differential equations, partial differential equations, and mean-field control problems, and also show that it can generalize its learning capability to operators beyond the training distribution.

The article explores the potential benefits of combining automation and artificial intelligence (AI) in scientific laboratories. It discusses the use of self-driving labs and cloud labs to improve efficiency and speed in various scientific fields. While there is excitement about the possibilities of autonomous laboratories, the article acknowledges the need for human supervision and the challenges associated with implementing these technologies. It highlights examples of successful self-driving labs and emphasizes the importance of improved automation in leveraging machine learning in scientific research. The article presents a balanced view of the current state and future potential of AI in scientific research. [Extracted from the article]

Antimicrobial resistance was estimated to be associated with 4.95 million deaths worldwide in 2019. It is possible to frame the antimicrobial resistance problem as a feedback-control problem. If we could optimize this feedback-control problem and translate our findings to the clinic, we could slow, prevent, or reverse the development of high-level drug resistance. Prior work on this topic has relied on systems where the exact dynamics and parameters were known a priori. In this study, we extend this work using a reinforcement learning (RL) approach capable of learning effective drug cycling policies in a system defined by empirically measured fitness landscapes. Crucially, we show that it is possible to learn effective drug cycling policies despite the problems of noisy, limited, or delayed measurement. Given access to a panel of 15 β-lactam antibiotics with which to treat the simulated Escherichia coli population, we demonstrate that RL agents outperform two naive treatment paradigms at minimizing the population fitness over time. We also show thatRLagents approach the performance of the optimal drug cycling policy. Even when stochastic noise is introduced to the measurements of population fitness, we show that RL agents are capable of maintaining evolving populations at lower growth rates compared to controls. We further tested our approach in arbitrary fitness landscapes of up to 1,024 genotypes. We show that minimization of population fitness using drug cycles is not limited by increasing genome size. Our work represents a proof-of-concept for using AI to control complex evolutionary processes. [ABSTRACT FROM AUTHOR]

People want to "feel heard" to perceive that they are understood, validated, and valued. Can AI serve the deeply human function of making others feel heard? Our research addresses two fundamental issues: Can AI generate responses that make human recipients feel heard, and how do human recipients react when they believe the response comes from AI? We conducted an experiment and a follow-up study to disentangle the effects of actual source of a message and the presumed source. We found that AI-generated messages made recipients feel more heard than human-generated messages and that AI was better at detecting emotions. However, recipients felt less heard when they realized that a message came from AI (vs. human). Finally, in a follow-up study where the responses were rated by third-party raters, we found that compared with humans, AI demonstrated superior discipline in offering emotional support, a crucial element in making individuals feel heard, while avoiding excessive practical suggestions, which may be less effective in achieving this goal. Our research underscores the potential and limitations of AI in meeting human psychological needs. These findings suggest that while AI demonstrates enhanced capabilities to provide emotional support, the devaluation of AI responses poses a key challenge for effectively leveraging AI's capabilities. [ABSTRACT FROM AUTHOR]

This article explores the topic of protein folding, which is essential for understanding biological processes in cells. Proteins need to fold into specific three-dimensional structures to function properly, and misfolded proteins can lead to diseases like Alzheimer's and Parkinson's. Recent advancements in AI-based methods, such as DeepMind's AlphaFold, have allowed for accurate predictions of protein structures. The article discusses the impact of these advancements and the collaboration between experimental and computational approaches in protein folding research. It introduces a new method called CoDropleT, which uses deep learning to extract protein properties from their amino acid sequences. The article also presents experimental validation of a predicted protein complex structure and discusses other studies on protein aggregation, external forces on protein stability, and the role of evolution in protein unfolding. The collection of articles highlights the interdisciplinary nature of protein folding research and the promising future of protein science. [Extracted from the article]

We administer a Turing test to AI chatbots. We examine how chatbots behave in a suite of classic behavioral games that are designed to elicit characteristics such as trust, fairness, risk-aversion, cooperation, etc., as well as how they respond to a traditional Big-5 psychological survey that measures personality traits. ChatGPT-4 exhibits behavioral and personality traits that are statistically indistinguishable from a random human from tens of thousands of human subjects from more than 50 countries. Chatbots also modify their behavior based on previous experience and contexts "as if" they were learning from the interactions and change their behavior in response to different framings of the same strategic situation. Their behaviors are often distinct from average and modal human behaviors, in which case they tend to behave on the more altruistic and cooperative end of the distribution. We estimate that they act as if they are maximizing an average of their own and partner's payoffs. [ABSTRACT FROM AUTHOR]

Quantum computers have been proposed to solve a number of important problems such as discovering new drugs, new catalysts for fertilizer production, breaking encryption protocols, optimizing financial portfolios, or implementing new artificial intelligence applications. Yet, to date, a simple task such as multiplying 3 by 5 is beyond existing quantum hardware. This article examines the difficulties that would need to be solved for quantum computers to live up to their promises. I discuss the whole stack of technologies that has been envisioned to build a quantum computer from the top layers (the actual algorithms and associated applications) down to the very bottom ones (the quantum hardware, its control electronics, cryogeny, etc.) while not forgetting the crucial intermediate layer of quantum error correction. [ABSTRACT FROM AUTHOR]

Forms of both simple and complex machine intelligence are increasingly acting within human groups in order to affect collective outcomes. Considering the nature of collective action problems, however, such involvement could paradoxically and unintentionally suppress existing beneficial social norms in humans, such as those involving cooperation. Here, we test theoretical predictions about such an effect using a unique cyber-physical lab experiment where online participants (N = 300 in 150 dyads) drive robotic vehicles remotely in a coordination game. We show that autobraking assistance increases human altruism, such as giving way to others, and that communication helps people to make mutual concessions. On the other hand, autosteering assistance completely inhibits the emergence of reciprocity between people in favor of self-interest maximization. The negative social repercussions persist even after the assistance system is deactivated. Furthermore, adding communication capabilities does not relieve this inhibition of reciprocity because people rarely communicate in the presence of autosteering assistance. Our findings suggest that active safety assistance (a form of simple AI support) can alter the dynamics of social coordination between people, including by affecting the trade-off between individual safety and social reciprocity. The difference between autobraking and autosteering assistance appears to relate to whether the assistive technology supports or replaces human agency in social coordination dilemmas. Humans have developed norms of reciprocity to address collective challenges, but such tacit understandings could break down in situations where machine intelligence is involved in human decision-making without having any normative commitments. [ABSTRACT FROM AUTHOR]

Political discourse is the soul of democracy, but misunderstanding and conflict can fester in divisive conversations. The widespread shift to online discourse exacerbates many of these problems and corrodes the capacity of diverse societies to cooperate in solving social problems. Scholars and civil society groups promote interventions that make conversations less divisive or more productive, but scaling these efforts to online discourse is challenging. We conduct a large-scale experiment that demonstrates how online conversations about divisive topics can be improved with AI tools. Specifically, we employ a large language model to make real-time, evidence-based recommendations intended to improve participants’ perception of feeling understood. These interventions improve reported conversation quality, promote democratic reciprocity, and improve the tone, without systematically changing the content of the conversation or moving people’s policy attitudes. [ABSTRACT FROM AUTHOR]

One of the most troubling trends in criminal investigations is the growing use of “black box” technology, in which law enforcement rely on artificial intelligence (AI) models or algorithms that are either too complex for people to understand or they simply conceal how it functions. In criminal cases, black box systems have proliferated in forensic areas such as DNA mixture interpretation, facial recognition, and recidivism risk assessments. The champions and critics of AI argue, mistakenly, that we face a catch 22: While black box AI is not understandable by people, they assume that it produces more accurate forensic evidence. In this Article, we question this assertion, which has so powerfully affected judges, policymakers, and academics. We describe a mature body of computer science research showing how “glass box” AI—designed to be interpretable—can be more accurate than black box alternatives. Indeed, black box AI performs predictably worse in settings like the criminal system. Debunking the black box performance myth has implications for forensic evidence, constitutional criminal procedure rights, and legislative policy. Absent some compelling—or even credible—government interest in keeping AI as a black box, and given the constitutional rights and public safety interests at stake, we argue that a substantial burden rests on the government to justify black box AI in criminal cases. We conclude by calling for judicial rulings and legislation to safeguard a right to interpretable forensic AI. [ABSTRACT FROM AUTHOR]

Machine learning methods, particularly neural networks trained on large datasets, are transforming how scientists approach scientific discovery and experimental design. However, current state-of-the-art neural networks are limited by their uninterpretability: Despite their excellent accuracy, they cannot describe how they arrived at their predictions. Here, using an “interpretable-by-design” approach, we present a neural network model that provides insights into RNA splicing, a fundamental process in the transfer of genomic information into functional biochemical products. Although we designed our model to emphasize interpretability, its predictive accuracy is on par with state-of-the-art models. To demonstrate the model’s interpretability, we introduce a visualization that, for any given exon, allows us to trace and quantify the entire decision process from input sequence to output splicing prediction. Importantly, the model revealed uncharacterized components of the splicing logic, which we experimentally validated. This study highlights how interpretable machine learning can advance scientific discovery. [ABSTRACT FROM AUTHOR]

Research in both ecology and AI strives for predictive understanding of complex systems, where nonlinearities arise from multidimensional interactions and feedbacks across multiple scales. After a century of independent, asynchronous advances in computational and ecological research, we foresee a critical need for intentional synergy to meet current societal challenges against the backdrop of global change. These challenges include understanding the unpredictability of systems-level phenomena and resilience dynamics on a rapidly changing planet. Here, we spotlight both the promise and the urgency of a convergence research paradigm between ecology and AI. Ecological systems are a challenge to fully and holistically model, even using the most prominent AI technique today: deep neural networks. Moreover, ecological systems have emergent and resilient behaviors that may inspire new, robust AI architectures and methodologies. We share examples of how challenges in ecological systems modeling would benefit from advances in AI techniques that are themselves inspired by the systems they seek to model. Both fields have inspired each other, albeit indirectly, in an evolution toward this convergence. We emphasize the need for more purposeful synergy to accelerate the understanding of ecological resilience whilst building the resilience currently lacking in modern AI systems, which have been shown to fail at times because of poor generalization in different contexts. Persistent epistemic barriers would benefit from attention in both disciplines. The implications of a successful convergence go beyond advancing ecological disciplines or achieving an artificial general intelligence--they are critical for both persisting and thriving in an uncertain future. [ABSTRACT FROM AUTHOR]

While deep learning has enabled significant advances in many areas of science, its blackbox nature hinders architecture design for future artificial intelligence applications and interpretation for high-stakes decision-makings. We addressed this issue by studying the fundamental question of how deep neural networks process data in the intermediate layers. Our finding is a simple and quantitative law that governs how deep neural networks separate data according to class membership throughout all layers for classification. This law shows that each layer improves data separation at a constant geometric rate, and its emergence is observed in a collection of network architectures and datasets during training. This law offers practical guidelines for designing architectures, improving model robustness and out-of-sample performance, as well as interpreting the predictions. [ABSTRACT FROM AUTHOR]

Glial cells account for between 50% and 90% of all human brain cells, and serve a variety of important developmental, structural, and metabolic functions. Recent experimental efforts suggest that astrocytes, a type of glial cell, are also directly involved in core cognitive processes such as learning and memory. While it is well established that astrocytes and neurons are connected to one another in feedback loops across many timescales and spatial scales, there is a gap in understanding the computational role of neuron-astrocyte interactions. To help bridge this gap, we draw on recent advances in AI and astrocyte imaging technology. In particular, we show that neuron-astrocyte networks can naturally perform the core computation of a Transformer, a particularly successful type of AI architecture. In doing so, we provide a concrete, normative, and experimentally testable account of neuron-astrocyte communication. Because Transformers are so successful across a wide variety of task domains, such as language, vision, and audition, our analysis may help explain the ubiquity, flexibility, and power of the brain's neuron-astrocyte networks. [ABSTRACT FROM AUTHOR]

Changes in behavioral state, such as arousal and movements, strongly affect neural activity in sensory areas, and can be modeled as long-range projections regulating the mean and variance of baseline input currents. What are the computational benefits of these baseline modulations? We investigate this question within a brain-inspired framework for reservoir computing, where we vary the quenched baseline inputs to a recurrent neural network with random couplings. We found that baseline modulations control the dynamical phase of the reservoir network, unlocking a vast repertoire of network phases.Weuncovered a number of bistable phases exhibiting the simultaneous coexistence of fixed points and chaos, of two fixed points, and of weak and strong chaos. We identified several phenomena, including noise-driven enhancement of chaos and ergodicity breaking; neural hysteresis, whereby transitions across a phase boundary retain the memory of the preceding phase. In each bistable phase, the reservoir performs a different binary decision-making task. Fast switching between different tasks can be controlled by adjusting the baseline input mean and variance. Moreover, we found that the reservoir network achieves optimal memory performance at any first-order phase boundary. In summary, baseline control enables multitasking without any optimization of the network couplings, opening directions for brain-inspired artificial intelligence and providing an interpretation for the ubiquitously observed behavioral modulations of cortical activity. [ABSTRACT FROM AUTHOR]

Thinking about God promotes greater acceptance of Artificial intelligence (AI)-based recommendations. Eight preregistered experiments (n = 2,462) reveal that when God is salient, people are more willing to consider AI-based recommendations than when God is not salient. Studies 1 and 2a to 2d demonstrate across a wide variety of contexts, from choosing entertainment and food to mutual funds and dental procedures, that God salience reduces reliance on human recommenders and heightens willingness to consider AI recommendations. Studies 3 and 4 demonstrate that the reduced reliance on humans is driven by a heightened feeling of smallness when God is salient, followed by a recognition of human fallibility. Study 5 addresses the similarity in mysteriousness between God and AI as an alternative, but unsupported, explanation. Finally, study 6 (n = 53,563) corroborates the experimental results with data from 21 countries on the usage of robo-advisors in financial decision-making. [ABSTRACT FROM AUTHOR]

The CASP14 experiment demonstrated the extraordinary structure modeling capabilities of artificial intelligence (AI) methods. That result has ignited a fierce debate about what these methods are actually doing. One of the criticisms has been that the AI does not have any sense of the underlying physics but is merely performing pattern recognition. Here, we address that issue by analyzing the extent to which the methods identify rare structural motifs. The rationale underlying the approach is that a pattern recognition machine tends to choose the more frequently occurring motifs, whereas some sense of subtle energetic factors is required to choose infrequently occurring ones. To reduce the possibility of bias from related experimental structures and to minimize the effect of experimental errors, we examined only CASP14 target protein crystal structures determined to a resolution limit better than 2 Å, which lacked significant amino acid sequence homology to proteins of known structure. In those experimental structures and in the corresponding models, we track cis peptides, π- helices, 3 10- helices, and other small 3D motifs that occur in the PDB database at a frequency of lower than 1% of total amino acid residues. The best-performing AI method, AlphaFold2, captured these uncommon structural elements exquisitely well. All discrepancies appeared to be a con- sequence of crystal environment effects. We propose that the neural network learned a protein structure potential of mean force, enabling it to correctly identify situations where unusual structural features represent the lowest local free energy because of subtle influences from the atomic environment. [ABSTRACT FROM AUTHOR]

The controllability and targeting of nanoparticles (NPs) offer solutions for precise and sustainable agriculture. However, the development potential of nanoenabled agriculture remains unknown. Here, we build an NP-plant database containing 1,174 datasets and predict (R² higher than 0.8 for 13 random forest models) the response and uptake/ transport of various NPs by plants using a machine learning approach. Multiway feature importance analysis quantitatively shows that plant responses are driven by the total NP exposure dose and duration and plant age at exposure, as well as the NP size and zeta potential. Feature interaction and covariance analysis further improve the interpretability of the model and reveal hidden interaction factors (e.g., NP size and zeta potential). Integration of the model, laboratory, and field data suggests that Fe2O3 NP application may inhibit bean growth in Europe due to low night temperatures. In contrast, the risks of oxi dative stress are low in Africa because of high night temperatures. According to the prediction, Africa is a suitable area for nanoenabled agriculture. The regional differences and temperature changes make nanoenabled agriculture complicated. In the future, the temperature increase may reduce the oxidative stress in African bean and European maize induced by NPs. This study projects the development potential of nanoenabled agriculture using machine learning, although many more field studies are needed to address the differences at the country and continental scales. [ABSTRACT FROM AUTHOR]

The philosopher John Rawls proposed the Veil of Ignorance (VoI) as a thought experiment to identify fair principles for governing a society. Here, we apply the VoI to an important governance domain: artificial intelligence (AI). In five incentivecompatible studies (N = 2,508), including two preregistered protocols, participants choose principles to govern an Artificial Intelligence (AI) assistant from behind the veil: that is, without knowledge of their own relative position in the group. Compared to participants who have this information, we find a consistent preference for a principle that instructs the AI assistant to prioritize the worst-off. Neither risk attitudes nor political preferences adequately explain these choices. Instead, they appear to be driven by elevated concerns about fairness: Without prompting, participants who reason behind the VoI more frequently explain their choice in terms of fairness, compared to those in the Control condition. Moreover, we find initial support for the ability of the VoI to elicit more robust preferences: In the studies presented here, the VoI increases the likelihood of participants continuing to endorse their initial choice in a subsequent round where they know how they will be affected by the AI intervention and have a self-interested motivation to change their mind. These results emerge in both a descriptive and an immersive game. Our findings suggest that the VoI may be a suitable mechanism for selecting distributive principles to govern AI. [ABSTRACT FROM AUTHOR]

We survey a current, heated debate in the artificial intelligence (AI) research community on whether large pretrained language models can be said to understand language—and the physical and social situations language encodes—in any humanlike sense. We describe arguments that have been made for and against such understanding and key questions for the broader sciences of intelligence that have arisen in light of these arguments. We contend that an extended science of intelligence can be developed that will provide insight into distinct modes of understanding, their strengths and limitations, and the challenge of integrating diverse forms of cognition. [ABSTRACT FROM AUTHOR]

Sudden changes in populations are ubiquitous in ecological systems, especially under perturbations. The agents of global change may increase the frequency and severity of anthropogenic perturbations, but complex populations' responses hamper our understanding of their dynamics and resilience. Furthermore, the long-term environmental and demographic data required to study those sudden changes are rare. Fitting dynamical models with an artificial intelligence algorithm to population fluctuations over 40 y in a social bird reveals that feedback in dispersal after a cumulative perturbation drives a population collapse. The collapse is well described by a nonlinear function mimicking social copying, whereby dispersal made by a few individuals induces others to leave the patch in a behavioral cascade for decision-making to disperse. Once a threshold for deterioration of the quality of the patch is crossed, there is a tipping point for a social response of runaway dispersal corresponding to social copying feedback. Finally, dispersal decreases at low population densities, which is likely due to the unwillingness of the more philopatric individuals to disperse. In providing the evidence of copying for the emergence of feedback in dispersal in a social organism, our results suggest a broader impact of self-organized collective dispersal in complex population dynamics. This has implications for the theoretical study of population and metapopulation nonlinear dynamics, including population extinction, and managing of endangered and harvested populations of social animals subjected to behavioral feedback loops. [ABSTRACT FROM AUTHOR]

Every year, millions of brain MRI scans are acquired in hospitals, which is a figure considerably larger than the size of any research dataset. Therefore, the ability to analyze such scans could transform neuroimaging research. Yet, their potential remains untapped since no automated algorithm is robust enough to cope with the high variability in clinical acquisitions (MR contrasts, resolutions, orientations, artifacts, and subject populations). Here, we present SynthSeg+, an AI segmentation suite that enables robust analysis of heterogeneous clinical datasets. In addition to wholebrain segmentation, SynthSeg+ also performs cortical parcellation, intracranial volume estimation, and automated detection of faulty segmentations (mainly caused by scans of very low quality). We demonstrate SynthSeg+ in seven experiments, including an aging study on 14,000 scans, where it accurately replicates atrophy patterns observed on data of much higher quality. SynthSeg+ is publicly released as a ready-to-use tool to unlock the potential of quantitative morphometry. [ABSTRACT FROM AUTHOR]

We study GPT-3, a recent large language model, using tools from cognitive psychology. More specifically, we assess GPT-3's decision-making, information search, deliberation, and causal reasoning abilities on a battery of canonical experiments from the literature. We find that much of GPT-3's behavior is impressive: It solves vignette-based tasks similarly or better than human subjects, is able to make decent decisions from descriptions, outperforms humans in a multiarmed bandit task, and shows signatures of model-based reinforcement learning. Yet, we also find that small perturbations to vignette-based tasks can lead GPT-3 vastly astray, that it shows no signatures of directed exploration, and that it fails miserably in a causal reasoning task. Taken together, these results enrich our understanding of current large language models and pave the way for future investigations using tools from cognitive psychology to study increasingly capable and opaque artificial agents. [ABSTRACT FROM AUTHOR]

We analyze the knowledge acquired by AlphaZero, a neural network engine that learns chess solely by playing against itself yet becomes capable of outperforming human chess players. Although the system trains without access to human games or guidance, it appears to learn concepts analogous to those used by human chess players. We provide two lines of evidence. Linear probes applied to AlphaZero’s internal state enable us to quantify when and where such concepts are represented in the network. We also describe a behavioral analysis of opening play, including qualitative commentary by a former world chess champion. [ABSTRACT FROM AUTHOR]

Vacuolar-type adenosine triphosphatases (V-ATPases) not only function as rotary proton pumps in cellular organelles but also serve as signaling hubs. To identify the endogenous binding partners of V-ATPase, we collected a large dataset of human V-ATPases and did extensive classification and focused refinement of human V-ATPases. Unexpectedly, about 17% of particles in state 2 of human V-ATPases display additional density with an overall resolution of 3.3 Å. Structural analysis combined with artificial intelligence modeling enables us to identify this additional density asmEAK-7, a protein involved in mechanistic target of rapamycin (mTOR) signaling in mammals. Our structure shows thatmEAK-7 interacts with subunits A, B, D, and E of V-ATPases in state 2. Thus, we propose that mEAK-7 may regulate V-ATPase function through binding to V-ATPases in state 2 as well as mediate mTOR signaling. [ABSTRACT FROM AUTHOR]

We propose that coding and decoding in the brain are achieved through digital computation using three principles: relative ordinal coding of inputs, random connections between neurons, and belief voting. Due to randomization and despite the coarseness of the relative codes, we show that these principles are sufficient for coding and decoding sequences with error-free reconstruction. In particular, the number of neurons needed grows linearly with the size of the input repertoire growing exponentially. We illustrate our model by reconstructing sequences with repertoires on the order of a billion items. From this, we derive the Shannon equations for the capacity limit to learn and transfer information in the neural population, which is then generalized to any type of neural network. Following the maximum entropy principle of efficient coding, we show that random connections serve to decorrelate redundant information in incoming signals, creating more compact codes for neurons and therefore, conveying a larger amount of information. Henceforth, despite the unreliability of the relative codes, few neurons become necessary to discriminate the original signal without error. Finally, we discuss the significance of this digital computation model regarding neurobiological findings in the brain and more generally with artificial intelligence algorithms, with a view toward a neural information theory and the design of digital neural networks. [ABSTRACT FROM AUTHOR]

Using simulations or experiments performed at some set of temperatures to learn about the physics or chemistry at some other arbitrary temperature is a problem of immense practical and theoretical relevance. Here we develop a framework based on statistical mechanics and generative artificial intelligence that allows solving this problem. Specifically, we work with denoising diffusion probabilistic models and show how these models in combination with replica exchange molecular dynamics achieve superior sampling of the biomolecular energy landscape at temperatures that were never simulated without assuming any particular slow degrees of freedom. The key idea is to treat the temperature as a fluctuating random variable and not a control parameter as is usually done. This allows us to directly sample from the joint probability distribution in configuration and temperature space. The results here are demonstrated for a chirally symmetric peptide and single-strand RNA undergoing conformational transitions in all-atom water. We demonstrate how we can discover transition states and metastable states that were previously unseen at the temperature of interest and even bypass the need to perform further simulations for a wide range of temperatures. At the same time, any unphysical states are easily identifiable through very low Boltzmann weights. The procedure while shown here for a class of molecular simulations should be more generally applicable to mixing information across simulations and experiments with varying control parameters. [ABSTRACT FROM AUTHOR]

Humans increasingly rely on artificial intelligence (AI) for efficient and objective decision-making, yet there is increasing concern that algorithms used by modern AI systems produce discriminatory outputs, presumably because they are trained on data in which societal biases are embedded. As a consequence, their use by human decision makers may result in the propagation, rather than reduction, of existing disparities. To assess this hypothesis empirically, we tested the relation between societal gender inequality and algorithmic search output and then examined the effect of this output on human decision-making. First, in two multinational samples (n = 37, 52 countries), we found that greater nation-level gender inequality was associated with more maledominated Google image search results for the gender-neutral keyword "person" (in a nation's dominant language), revealing a link between societal-level disparities and algorithmic output. Next, in a series of experiments with human participants (n = 395), we demonstrated that the gender disparity associated with high-vs. low-inequality algorithmic outputs guided the formation of gender-biased prototypes and influenced hiring decisions in novel scenarios. These findings support the hypothesis that societal-level gender inequality is recapitulated in internet search algorithms, which in turn can influence human decision makers to act in ways that reinforce these disparities. [ABSTRACT FROM AUTHOR]

Regional phenotypic and functional differences in the retinal pigment epithelium (RPE) monolayer have been suggested to account for regional susceptibility in ocular diseases such as age-related macular degeneration (AMD), late-onset retinal degeneration (L-ORD), and choroideremia (CHM). However, a comprehensive description of human topographical RPE diversity is not yet available, thus limiting the understanding of regional RPE diversity and degenerative disease sensitivity in the eye. To develop a complete morphometric RPE map of the human eye, artificial intelligence–based software was trained to recognize, segment, and analyze RPE borders. Five statistically different, concentric RPE subpopulations (P1 to P5) were identified using cell area as a parameter, including a subpopulation (P4) with cell area comparable to that of macular cells in the far periphery of the eye. This work provides a complete reference map of human RPE subpopulations and their location in the eye. In addition, the analysis of cadaver non-AMD and AMD eyes and ultra-widefield fundus images of patients revealed differential vulnerability of the five RPE subpopulations to different retinal diseases. [ABSTRACT FROM AUTHOR]

Although they are staple foods in cuisines globally, many commercial fruit varieties have become progressively less flavorful over time. Due to the cost and difficulty associated with flavor phenotyping, breeding programs have long been challenged in selecting for this complex trait. To address this issue, we leveraged targeted metabolomics of diverse tomato and blueberry accessions and their corresponding consumer panel ratings to create statistical and machine learning models that can predict sensory perceptions of fruit flavor. Using these models, a breeding program can assess flavor ratings for a large number of genotypes, previously limited by the low throughput of consumer sensory panels. The ability to predict consumer ratings of liking, sweet, sour, umami, and flavor intensity was evaluated by a 10-fold cross-validation, and the accuracies of 18 different models were assessed. The prediction accuracies were high for most attributes and ranged from 0.87 for sourness intensity in blueberry using XGBoost to 0.46 for overall liking in tomato using linear regression. Further, the bestperforming models were used to infer the flavor compounds (sugars, acids, and volatiles) that contribute most to each flavor attribute. We found that the variance decomposition of overall liking score estimates that 42% and 56% of the variance was explained by volatile organic compounds in tomato and blueberry, respectively. We expect that these models will enable an earlier incorporation of flavor as breeding targets and encourage selection and release of more flavorful fruit varieties. [ABSTRACT FROM AUTHOR]

The recent emergence of machine-manipulated media raises an important societal question: How can we know whether a video that we watch is real or fake? In two online studies with 15,016 participants, we present authentic videos and deepfakes and ask participants to identify which is which. We compare the performance of ordinary human observers with the leading computer vision deepfake detection model and find them similarly accurate, while making different kinds of mistakes. Together, participants with access to the model's prediction are more accurate than either alone, but inaccurate model predictions often decrease participants' accuracy. To probe the relative strengths and weaknesses of humans and machines as detectors of deepfakes, we examine human and machine performance across video-level features, and we evaluate the impact of preregistered randomized interventions on deepfake detection. We find thatmanipulations designed to disrupt visual processing of faces hinder human participants' performance while mostly not affecting the model's performance, suggesting a role for specialized cognitive capacities in explaining human deepfake detection performance. [ABSTRACT FROM AUTHOR]

Researchers in areas as diverse as computer science and political science must increasingly navigate the possible risks of their research to society. However, the history of medical experiments on vulnerable individuals influencedmany research ethics reviews to focus exclusively on risks to human subjects rather than risks to human society. We describe an Ethics and Society Review board (ESR), which fills this moral gap by facilitating ethical and societal reflection as a requirement to access grant funding: Researchers cannot receive grant funding from participating programs until the researchers complete the ESR process for their proposal. Researchers author an initial statement describing their proposed research's risks to society, subgroups within society, and globally and commit to mitigation strategies for these risks. An interdisciplinary faculty panel iterates with the researchers to refine these risks and mitigation strategies. We describe a mixedmethod evaluation of the ESR over 1 y, in partnership with a large artificial intelligence grant program at our university. Surveys and interviews of researchers who interacted with the ESR found 100% (95% CI: 87 to 100%) were willing to continue submitting future projects to the ESR, and 58% (95% CI: 37 to 77%) felt that it had influenced the design of their research project. The ESR panel most commonly identified issues of harms to minority groups, inclusion of diverse stakeholders in the research plan, dual use, and representation in datasets. These principles, paired with possible mitigation strategies, offer scaffolding for future research designs. [ABSTRACT FROM AUTHOR]

Over the past five decades, tremendous effort has been devoted to computational methods for predicting properties of ligands--i.e., molecules that bind macromolecular targets. Such methods, which are critical to rational drug design, fall into two categories: physics-based methods, which directly model ligand interactions with the target given the target's three-dimensional (3D) structure, and ligand-based methods, which predict ligand properties given experimental measurements for similar ligands. Here, we present a rigorous statistical framework to combine these two sources of information. We develop a method to predict a ligand's pose--the 3D structure of the ligand bound to its target--that leverages a widely available source of information: a list of other ligands that are known to bind the same target but for which no 3D structure is available. This combination of physics-based and ligand-based modeling improves pose prediction accuracy across all major families of drug targets. Using the same framework, we develop a method for virtual screening of drug candidates, which outperforms standard physics-based and ligand-based virtual screening methods. Our results suggest broad opportunities to improve prediction of various ligand properties by combining diverse sources of information through customized machine-learning approaches. [ABSTRACT FROM AUTHOR]