RESUMO
reasoning is a key ability for an intelligent system. Large language models (LMs) achieve above-chance performance on abstract reasoning tasks but exhibit many imperfections. However, human abstract reasoning is also imperfect. Human reasoning is affected by our real-world knowledge and beliefs, and shows notable "content effects"; humans reason more reliably when the semantic content of a problem supports the correct logical inferences. These content-entangled reasoning patterns are central to debates about the fundamental nature of human intelligence. Here, we investigate whether language models-whose prior expectations capture some aspects of human knowledge-similarly mix content into their answers to logic problems. We explored this question across three logical reasoning tasks: natural language inference, judging the logical validity of syllogisms, and the Wason selection task. We evaluate state of the art LMs, as well as humans, and find that the LMs reflect many of the same qualitative human patterns on these tasks-like humans, models answer more accurately when the semantic content of a task supports the logical inferences. These parallels are reflected in accuracy patterns, and in some lower-level features like the relationship between LM confidence over possible answers and human response times. However, in some cases the humans and models behave differently-particularly on the Wason task, where humans perform much worse than large models, and exhibit a distinct error pattern. Our findings have implications for understanding possible contributors to these human cognitive effects, as well as the factors that influence language model performance.
RESUMO
Organ-on-chip (OOC) technology has gained importance for biomedical studies and drug development. This technology involves microfluidic devices that mimic the structure and function of specific human organs or tissues. OOCs are a promising alternative to traditional cell-based models and animals, as they provide a more representative experimental model of human physiology. By creating a microenvironment that closely resembles in vivo conditions, OOC platforms enable the study of intricate interactions between different cells as well as a better understanding of the underlying mechanisms pertaining to diseases. OOCs can be integrated with other technologies, such as sensors and imaging systems to monitor real-time responses and gather extensive data on tissue behavior. Despite these advances, OOCs for many organs are in their initial stages of development, with several challenges yet to be overcome. These include improving the complexity and maturity of these cellular models, enhancing their reproducibility, standardization, and scaling them up for high-throughput uses. Nonetheless, OOCs hold great promise in advancing biomedical research, drug discovery, and personalized medicine, benefiting human health and well-being. Here, we review several recent OOCs that attempt to overcome some of these challenges. These OOCs with unique applications can be engineered to model organ systems such as the stomach, cornea, blood vessels, and mouth, allowing for analyses and investigations under more realistic conditions. With this, these models can lead to the discovery of potential therapeutic interventions. In this review, we express the significance of the relationship between mucosal tissues and vasculature in organ-on-chip (OOC) systems. This interconnection mirrors the intricate physiological interactions observed in the human body, making it crucial for achieving accurate and meaningful representations of biological processes within OOC models. Vasculature delivers essential nutrients and oxygen to mucosal tissues, ensuring their proper function and survival. This exchange is critical for maintaining the health and integrity of mucosal barriers. This review will discuss the OOCs used to represent the mucosal architecture and vasculature, and it can encourage us to think of ways in which the integration of both can better mimic the complexities of biological systems and gain deeper insights into various physiological and pathological processes. This will help to facilitate the development of more accurate predictive models, which are invaluable for advancing our understanding of disease mechanisms and developing novel therapeutic interventions.
RESUMO
Psychologists and neuroscientists extensively rely on computational models for studying and analyzing the human mind. Traditionally, such computational models have been hand-designed by expert researchers. Two prominent examples are cognitive architectures and Bayesian models of cognition. While the former requires the specification of a fixed set of computational structures and a definition of how these structures interact with each other, the latter necessitates the commitment to a particular prior and a likelihood function which - in combination with Bayes' rule - determine the model's behavior. In recent years, a new framework has established itself as a promising tool for building models of human cognition: the framework of meta-learning. In contrast to the previously mentioned model classes, meta-learned models acquire their inductive biases from experience, i.e., by repeatedly interacting with an environment. However, a coherent research program around meta-learned models of cognition is still missing to this day. The purpose of this article is to synthesize previous work in this field and establish such a research program. We accomplish this by pointing out that meta-learning can be used to construct Bayes-optimal learning algorithms, allowing us to draw strong connections to the rational analysis of cognition. We then discuss several advantages of the meta-learning framework over traditional methods and reexamine prior work in the context of these new insights.
RESUMO
Many cognitive models provide valuable insights into human behavior. Yet the algorithmic complexity of candidate models can fail to capture how human reaction times scale with increasing input complexity. In the current work, we investigate the algorithms underlying human cognitive processes. Computer science characterizes algorithms by their time and space complexity scaling with problem size. We propose to use participants' reaction times to study how human computations scale with increasing input complexity. We tested this approach in a task where participants had to sort sequences of rectangles by their size. Our results showed that reaction times scaled close to linearly with sequence length and that participants learned and actively used latent structure whenever it was provided. This behavior was in line with a computational model that used the observed sequences to form hypotheses about the latent structures, searching through candidate hypotheses in a directed fashion. These results enrich our understanding of plausible cognitive models for efficient mental sorting and pave the way for future studies using reaction times to investigate the scaling of mental computations across psychological domains.
RESUMO
The ability to acquire abstract knowledge is a hallmark of human intelligence and is believed by many to be one of the core differences between humans and neural network models. Agents can be endowed with an inductive bias towards abstraction through meta-learning, where they are trained on a distribution of tasks that share some abstract structure that can be learned and applied. However, because neural networks are hard to interpret, it can be difficult to tell whether agents have learned the underlying abstraction, or alternatively statistical patterns that are characteristic of that abstraction. In this work, we compare the performance of humans and agents in a meta-reinforcement learning paradigm in which tasks are generated from abstract rules. We define a novel methodology for building "task metamers" that closely match the statistics of the abstract tasks but use a different underlying generative process, and evaluate performance on both abstract and metamer tasks. We find that humans perform better at abstract tasks than metamer tasks whereas common neural network architectures typically perform worse on the abstract tasks than the matched metamers. This work provides a foundation for characterizing differences between humans and machine learning that can be used in future work towards developing machines with more human-like behavior.
Assuntos
Formação de Conceito , Aprendizado de Máquina , Humanos , Inteligência , Conhecimento , Redes Neurais de ComputaçãoRESUMO
When exposed to perceptual and motor sequences, people are able to gradually identify patterns within and form a compact internal description of the sequence. One proposal of how sequences can be compressed is people's ability to form chunks. We study people's chunking behavior in a serial reaction time task. We relate chunk representation with sequence statistics and task demands, and propose a rational model of chunking that rearranges and concatenates its representation to jointly optimize for accuracy and speed. Our model predicts that participants should chunk more if chunks are indeed part of the generative model underlying a task and should, on average, learn longer chunks when optimizing for speed than optimizing for accuracy. We test these predictions in two experiments. In the first experiment, participants learn sequences with underlying chunks. In the second experiment, participants were instructed to act either as fast or as accurately as possible. The results of both experiments confirmed our model's predictions. Taken together, these results shed new light on the benefits of chunking and pave the way for future studies on step-wise representation learning in structured domains.
Assuntos
Aprendizagem , Memória , Humanos , Tempo de ReaçãoRESUMO
Computer scientists have long recognized that naive implementations of algorithms often result in a paralyzing degree of redundant computation. More sophisticated implementations harness the power of memory by storing computational results and reusing them later. We review the application of these ideas to cognitive science, in four case studies (mental arithmetic, mental imagery, planning, and probabilistic inference). Despite their superficial differences, these cognitive processes share a common reliance on memory that enables efficient computation.
Assuntos
Algoritmos , Memória , HumanosRESUMO
As modern deep networks become more complex, and get closer to human-like capabilities in certain domains, the question arises as to how the representations and decision rules they learn compare to the ones in humans. In this work, we study representations of sentences in one such artificial system for natural language processing. We first present a diagnostic test dataset to examine the degree of abstract composable structure represented. Analyzing performance on these diagnostic tests indicates a lack of systematicity in representations and decision rules, and reveals a set of heuristic strategies. We then investigate the effect of training distribution on learning these heuristic strategies, and we study changes in these representations with various augmentations to the training set. Our results reveal parallels to the analogous representations in people. We find that these systems can learn abstract rules and generalize them to new contexts under certain circumstances-similar to human zero-shot reasoning. However, we also note some shortcomings in this generalization behavior-similar to human judgment errors like belief bias. Studying these parallels suggests new ways to understand psychological phenomena in humans as well as informs best strategies for building artificial intelligence with human-like language understanding.
Assuntos
Idioma , Aprendizado de Máquina , Processamento de Linguagem Natural , Compreensão , Heurística , HumanosRESUMO
Bayesian theories of cognition assume that people can integrate probabilities rationally. However, several empirical findings contradict this proposition: human probabilistic inferences are prone to systematic deviations from optimality. Puzzlingly, these deviations sometimes go in opposite directions. Whereas some studies suggest that people underreact to prior probabilities (base rate neglect), other studies find that people underreact to the likelihood of the data (conservatism). We argue that these deviations arise because the human brain does not rely solely on a general-purpose mechanism for approximating Bayesian inference that is invariant across queries. Instead, the brain is equipped with a recognition model that maps queries to probability distributions. The parameters of this recognition model are optimized to get the output as close as possible, on average, to the true posterior. Because of our limited computational resources, the recognition model will allocate its resources so as to be more accurate for high probability queries than for low probability queries. By adapting to the query distribution, the recognition model learns to infer. We show that this theory can explain why and when people underreact to the data or the prior, and a new experiment demonstrates that these two forms of underreaction can be systematically controlled by manipulating the query distribution. The theory also explains a range of related phenomena: memory effects, belief bias, and the structure of response variability in probabilistic reasoning. We also discuss how the theory can be integrated with prior sampling-based accounts of approximate inference. (PsycInfo Database Record (c) 2020 APA, all rights reserved).
Assuntos
Teorema de Bayes , Viés , Aprendizagem , Memória , Modelos Teóricos , Teoria da Probabilidade , Cognição , Heurística , Humanos , Resolução de ProblemasRESUMO
Bayesian models of cognition assume that people compute probability distributions over hypotheses. However, the required computations are frequently intractable or prohibitively expensive. Since people often encounter many closely related distributions, selective reuse of computations (amortized inference) is a computationally efficient use of the brain's limited resources. We present three experiments that provide evidence for amortization in human probabilistic reasoning. When sequentially answering two related queries about natural scenes, participants' responses to the second query systematically depend on the structure of the first query. This influence is sensitive to the content of the queries, only appearing when the queries are related. Using a cognitive load manipulation, we find evidence that people amortize summary statistics of previous inferences, rather than storing the entire distribution. These findings support the view that the brain trades off accuracy and computational cost, to make efficient use of its limited cognitive resources to approximate probabilistic inference.
Assuntos
Cognição , Modelos Psicológicos , Pensamento , Adulto , Teorema de Bayes , Feminino , Humanos , Masculino , Memória , Método de Monte Carlo , ProbabilidadeRESUMO
Why are human inferences sometimes remarkably close to the Bayesian ideal and other times systematically biased? In particular, why do humans make near-rational inferences in some natural domains where the candidate hypotheses are explicitly available, whereas tasks in similar domains requiring the self-generation of hypotheses produce systematic deviations from rational inference. We propose that these deviations arise from algorithmic processes approximating Bayes' rule. Specifically in our account, hypotheses are generated stochastically from a sampling process, such that the sampled hypotheses form a Monte Carlo approximation of the posterior. While this approximation will converge to the true posterior in the limit of infinite samples, we take a small number of samples as we expect that the number of samples humans take is limited. We show that this model recreates several well-documented experimental findings such as anchoring and adjustment, subadditivity, superadditivity, the crowd within as well as the self-generation effect, the weak evidence, and the dud alternative effects. We confirm the model's prediction that superadditivity and subadditivity can be induced within the same paradigm by manipulating the unpacking and typicality of hypotheses. We also partially confirm our model's prediction about the effect of time pressure and cognitive load on these effects.
Assuntos
Algoritmos , Modelos Estatísticos , Teorema de Bayes , Humanos , Método de Monte CarloRESUMO
Minimal change nephropathy (MCN) is the third most common cause of primary nephrotic syndrome in adults. Most patients with MCN respond to corticosteroid therapy, but relapse is common. In children, steroid-dependent patients are often given alternative agents to spare the use of steroids and to avoid the cumulative steroid toxicity. In this respect, levamisole has shown promise due to its ability to effectively maintain remission in children with steroid-sensitive or steroid-dependent nephrotic syndrome. Despite clinical effectiveness, there is a complete lack of molecular evidence to explain its mode of action and there are no published reports on the use of this compound in adult patients. We studied the effectiveness of levamisole in a small cohort of adult patients and also tested the hypothesis that levamisole's mode of action is attributable to its direct effects on podocytes. In the clinic, we demonstrate that in our adult patients, cohort levamisole is generally well tolerated and clinically useful. Using conditionally immortalized human podocytes, we show that levamisole is able to induce expression of glucocorticoid receptor (GR) and to activate GR signalling. Furthermore, levamisole is able to protect against podocyte injury in a puromycin aminonucleoside (PAN)-treated cell model. In this model the effects of levamisole are blocked by the GR antagonist mifepristone (RU486), suggesting that GR signalling is a critical target of levamisole's action. These results indicate that levamisole is effective in nephrotic syndrome in adults, as well as in children, and point to molecular mechanisms for this drug's actions in podocyte diseases.
Assuntos
Glucocorticoides/uso terapêutico , Levamisol/uso terapêutico , Síndrome Nefrótica/tratamento farmacológico , Adolescente , Adulto , Células Cultivadas/efeitos dos fármacos , Quimioterapia Combinada , Feminino , Humanos , Levamisol/efeitos adversos , Levamisol/antagonistas & inibidores , Levamisol/farmacologia , Masculino , Pessoa de Meia-Idade , Mifepristona/farmacologia , Síndrome Nefrótica/metabolismo , Síndrome Nefrótica/patologia , Uso Off-Label , Podócitos/efeitos dos fármacos , Podócitos/metabolismo , Prednisolona/uso terapêutico , Puromicina Aminonucleosídeo/antagonistas & inibidores , Puromicina Aminonucleosídeo/farmacologia , Receptores de Glucocorticoides/metabolismo , Transdução de Sinais/efeitos dos fármacos , Adulto JovemRESUMO
Reactive oxygen species (ROS) play a key role in the pathogenesis of proteinuria in glomerular diseases like diabetic nephropathy. Glomerular endothelial cell (GEnC) glycocalyx covers the luminal aspect of the glomerular capillary wall and makes an important contribution to the glomerular barrier. ROS are known to depolymerise glycosaminoglycan (GAG) chains of proteoglycans, which are crucial for the barrier function of GEnC glycocalyx. The aim of this study is to investigate the direct effects of ROS on the structure and function of GEnC glycocalyx using conditionally immortalised human GEnC. ROS were generated by exogenous hydrogen peroxide. Biosynthesis and cleavage of GAG chains was analyzed by radiolabelling (S(35) and (3)H-glucosamine). GAG chains were quantified on GEnC surface and in the cell supernatant using liquid chromatography and immunofluorescence techniques. Barrier properties were estimated by measuring trans-endothelial passage of albumin. ROS caused a significant loss of WGA lectin and heparan sulphate staining from the surface of GEnC. This lead to an increase in trans-endothelial albumin passage. The latter could be inhibited by catalase and superoxide dismutase. The effect of ROS on GEnC was not mediated via the GAG biosynthetic pathway. Quantification of radiolabelled GAG fractions in the supernatant confirmed that ROS directly caused shedding of HS GAG. This finding is clinically relevant and suggests a mechanism by which ROS may cause proteinuria in clinical conditions associated with high oxidative stress.
Assuntos
Células Endoteliais/metabolismo , Glicocálix/metabolismo , Glicosaminoglicanos/metabolismo , Glomérulos Renais/metabolismo , Espécies Reativas de Oxigênio/metabolismo , Albuminas/metabolismo , Linhagem Celular , Sobrevivência Celular , Humanos , Aglutininas do Germe de Trigo/metabolismoRESUMO
A buried ionizable residue can have a drastic effect on the stability of a native protein, but there has been only limited investigation of how burial of an ionizable residue affects the kinetics of protein folding. In this study, the effect of burial of ionizable residues on the thermodynamics and kinetics of folding and unfolding of monellin has been investigated. The stability of wild-type (wt) monellin is known to decrease with an increase in pH from 4 to 10. The Glu24 â Ala mutation makes the stability of the resultant E24A mutant protein independent of pH in the range from 4 to 8. An additional mutation, Cys42 â Ala, results in the stability becoming independent of pH in the range from 4 to 10. Like the wt protein, E24A folds via very fast, fast, and slow folding pathways. Compared to that of the wt protein, the rate of slow folding pathway of E24A is ~7-fold faster, the rate of fast folding pathway is ~1.5-fold faster, while the rate of very fast folding pathway is similar. E24A unfolds ~7-fold slower than the wt. The extent of stabilization of the transition state (TS) observed for the slow pathway of refolding and for unfolding is the same, indicating that unfolding occurs via the TS populated on the slow pathway of refolding. The stabilization of the TS of folding (1.1 kcal mol(-1)) is less than that of the native state (2.3 kcal mol(-1)) of E24A, indicating that structure has only partially formed in the vicinity of Glu24 in the TS of folding.