Large Language Models and the Reverse Turing Test

Neural Computation, Journal Year: 2023, Volume and Issue: 35(3), P. 309 - 342

Published: Feb. 6, 2023

Large language models (LLMs) have been transformative. They are pretrained foundational that self-supervised and can be adapted with fine-tuning to a wide range of natural tasks, each which previously would required separate network model. This is one step closer the extraordinary versatility human language. GPT-3 and, more recently, LaMDA, both them LLMs, carry on dialogs humans many topics after minimal priming few examples. However, there has reactions debate whether these LLMs understand what they saying or exhibit signs intelligence. high variance exhibited in three interviews reaching wildly different conclusions. A new possibility was uncovered could explain this divergence. What appears intelligence may fact mirror reflects interviewer, remarkable twist considered reverse Turing test. If so, then by studying interviews, we learning about beliefs interviewer than LLMs. As become capable, transform way interact machines how other. Increasingly, being coupled sensorimotor devices. talk talk, but walk walk? road map for achieving artificial general autonomy outlined seven major improvements inspired brain systems turn used uncover insights into function.

Language: Английский

---

Are Deep Neural Networks Adequate Behavioral Models of Human Visual Perception?

Annual Review of Vision Science, Journal Year: 2023, Volume and Issue: 9(1), P. 501 - 524

Published: March 31, 2023

Deep neural networks (DNNs) are machine learning algorithms that have revolutionized computer vision due to their remarkable successes in tasks like object classification and segmentation. The success of DNNs as has led the suggestion may also be good models human visual perception. In this article, we review evidence regarding current adequate behavioral core recognition. To end, argue it is important distinguish between statistical tools computational understand model quality a multidimensional concept which clarity about modeling goals key. Reviewing large number psychophysical explorations recognition performance humans DNNs, highly valuable scientific but that, today, should only regarded promising-but not yet adequate-computational behavior. On way, dispel several myths surrounding science.

Language: Английский

---

Cognitive modelling of concepts in the mental lexicon with multilayer networks: Insights, advancements, and future challenges

Psychonomic Bulletin & Review, Journal Year: 2024, Volume and Issue: 31(5), P. 1981 - 2004

Published: March 4, 2024

Abstract The mental lexicon is a complex cognitive system representing information about the words/concepts that one knows. Over decades psychological experiments have shown conceptual associations across multiple, interactive levels can greatly influence word acquisition, storage, and processing. How semantic, phonological, syntactic, other types of be mapped within coherent mathematical framework to study how works? Here we review multilayer networks as promising quantitative interpretative for investigating lexicon. Cognitive map multiple at once, thus capturing different layers might co-exist This starts with gentle introduction structure formalism networks. We then discuss mechanisms phenomena could not observed in single-layer were only unveiled by combining lexicon: (i) multiplex viability highlights language kernels facilitative effects knowledge processing healthy clinical populations; (ii) community detection enables contextual meaning reconstruction depending on psycholinguistic features; (iii) layer analysis mediate latent interactions mediation, suppression, facilitation lexical access. By outlining novel perspectives where shed light representations, including next-generation brain/mind models, key limitations directions cutting-edge future research.

Language: Английский

---

The Violation-of-Expectation Paradigm: A Conceptual Overview

Published: Aug. 21, 2022

For over 35 years, the violation-of-expectation paradigm has been used to study development of expectations in first three years life. A wide range examined, including physical, psychological, sociomoral, biological, numerical, statistical, probabilistic, and linguistic expectations. Surprisingly, despite paradigm’s widespread use many seminal findings it contributed psychological science, so far no one tried provide a detailed in-depth conceptual overview paradigm. Here, we attempted do just that. We focus on rationale discuss how evolved time. then show improved descriptions infants’ looking behavior, together with addition rich panoply brain behavioral measures, have helped deepen our understanding responses violations. Next, review strengths limitations. Finally, end discussion challenges that leveled against years. Through all, goal was two-fold. First, sought psychologists other scientists interested an informed constructive analysis its theoretical origins development. Second, wanted take stock what revealed date about infants form events, surprise at unexpected or out laboratory, can lead learning, by prompting revise their working model world as more accurate future.

Language: Английский

---

Commonsense psychology in human infants and machines

Cognition, Journal Year: 2023, Volume and Issue: 235, P. 105406 - 105406

Published: Feb. 16, 2023

Human infants are fascinated by other people. They bring to this fascination a constellation of rich and flexible expectations about the intentions motivating people's actions. Here we test 11-month-old state-of-the-art learning-driven neural-network models on "Baby Intuitions Benchmark (BIB)," suite tasks challenging both machines make high-level predictions underlying causes agents' Infants expected actions be directed towards objects, not locations, demonstrated default rationally efficient goals. The failed capture infants' knowledge. Our work provides comprehensive framework in which characterize commonsense psychology takes first step testing whether human knowledge human-like artificial intelligence can built from foundations cognitive developmental theories postulate.

Language: Английский

---

A Comprehensive Review and Analysis of Deep Learning-Based Medical Image Adversarial Attack and Defense

Mathematics, Journal Year: 2023, Volume and Issue: 11(20), P. 4272 - 4272

Published: Oct. 13, 2023

Deep learning approaches have demonstrated great achievements in the field of computer-aided medical image analysis, improving precision diagnosis across a range disorders. These developments not, however, been immune to appearance adversarial attacks, creating possibility incorrect with substantial clinical implications. Concurrently, has seen notable advancements defending against such targeted adversary intrusions deep diagnostic systems. In context this article provides comprehensive survey current attacks and their accompanying defensive strategies. addition, conceptual analysis is presented, including several strategies designed for interpretation images. This survey, which draws on qualitative quantitative findings, concludes thorough discussion problems attack mechanisms that are unique systems, opening up new directions future research. We identified main defense imaging include dataset labeling, computational resources, robustness target evaluation transferability adaptability, interpretability explainability, real-time detection response, multi-modal fusion. The area might move toward more secure, dependable, therapeutically useful systems by filling these research gaps following objectives.

Language: Английский

---

The Enemy of My Enemy is My Friend: Exploring Inverse Adversaries for Improving Adversarial Training

2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Journal Year: 2023, Volume and Issue: 30, P. 24678 - 24687

Published: June 1, 2023

Although current deep learning techniques have yielded superior performance on various computer vision tasks, yet they are still vulnerable to adversarial examples. Adversarial training and its variants been shown be the most effective approaches defend against A particular class of these methods regularize difference between output probabilities for an corresponding natural example. However, it may a negative impact if example is misclassified. To circumvent this issue, we propose novel scheme that encourages model produce similar "inverse adversarial" counterpart. Particularly, counterpart generated by maximizing likelihood in neighborhood Extensive experiments datasets architectures demonstrate our method achieves state-of-the-art robustness as well accuracy among robust models. Furthermore, using universal version inverse examples, improve single-step at low computational cost.

Language: Английский

---

The relational bottleneck as an inductive bias for efficient abstraction

Trends in Cognitive Sciences, Journal Year: 2024, Volume and Issue: 28(9), P. 829 - 843

Published: May 9, 2024

Language: Английский

---

Evidence synthesis, digital scribes, and translational challenges for artificial intelligence in healthcare

Cell Reports Medicine, Journal Year: 2022, Volume and Issue: 3(12), P. 100860 - 100860

Published: Dec. 1, 2022

Language: Английский

---

Quantifying the Impact of Large Language Models on Collective Opinion Dynamics

Published: Jan. 1, 2024

The process of opinion expression and exchange is a critical component democratic societies. As people interact with large language models (LLMs) in the shaping different from traditional media, impacts LLMs are increasingly recognized being concerned. However, knowledge about how affect social networks very limited. Here, we create an network dynamics model to encode opinions LLMs, cognitive acceptability usage strategies individuals, simulate impact on variety scenarios. outcomes simulations inform effective demand-oriented interventions. results this study suggested that output has unique positive effect collective difference. marginal formation nonlinear shows decreasing trend. When partially rely becomes more intense diversity favorable. In fact, there 38.6% when all compared prohibiting use entirely. optimal was found fractions who do not use, on, fully reached roughly 4:12:1. Our experiments also find introducing extra agents opposite/neutral/random opinions, can effectively mitigate biased/toxic LLMs. findings provide valuable insights into age highlighting need for customized interventions tailored specific scenarios address drawbacks improper

Language: Английский

---