Massimo Chiriatti, Marianna Bergamaschi Ganapini, Enrico Panai
et al.
This paper introduces System 0, a conceptual framework for understanding how artificial intelligence functions as a cognitive extension preceding both intuitive (System 1) and deliberative (System 2) thinking processes. As AI systems increasingly shape the informational substrate upon which human cognition operates, they transform from passive tools into active cognitive partners. Building on the Extended Mind hypothesis and Heersmink's criteria for cognitive extension, we argue that AI systems satisfy key conditions for cognitive integration. These include reliability, trust, transparency, individualization, and the ability to enhance and transform human mental functions. However, AI integration creates a paradox: while expanding cognitive capabilities, it may simultaneously constrain thinking through sycophancy and bias amplification. To address these challenges, we propose seven evidence-based frameworks for effective human-AI cognitive integration: Enhanced Cognitive Scaffolding, which promotes progressive autonomy; Symbiotic Division of Cognitive Labor, strategically allocating tasks based on comparative strengths; Dialectical Cognitive Enhancement, countering AI sycophancy through productive epistemic tension; Agentic Transparency and Control, ensuring users understand and direct AI influence; Expertise Democratization, breaking down knowledge silos; Social-Emotional Augmentation, addressing affective dimensions of cognitive work; and Duration-Optimized Integration, managing the evolving human-AI relationship over time. Together, these frameworks provide a comprehensive approach for harnessing AI as a genuine cognitive extension while preserving human agency, critical thinking, and intellectual growth, transforming AI from a replacement for human cognition into a catalyst for enhanced thinking.
This paper presents COGENT3 (or Collective Growth and Entropy-modulated Triads System), a novel approach for emergent cognition integrating pattern formation networks with group influence dynamics. Contrasting with traditional strategies that rely on predetermined architectures, computational structures emerge dynamically in our framework through agent interactions. This enables a more flexible and adaptive system exhibiting characteristics reminiscent of human cognitive processes. The incorporation of temperature modulation and memory effects in COGENT3 closely integrates statistical mechanics, machine learning, and cognitive science.
Luca M. Schulze Buschoff, Konstantinos Voudouris, Elif Akata
et al.
Pre-trained vision language models still fall short of human visual cognition. In an effort to improve visual cognition and align models with human behavior, we introduce visual stimuli and human judgments on visual cognition tasks, allowing us to systematically evaluate performance across cognitive domains under a consistent environment. We fine-tune models on ground truth data for intuitive physics and causal reasoning and find that this improves model performance in the respective fine-tuning domain. Furthermore, it can improve model alignment with human behavior. However, we find that task-specific fine-tuning does not contribute to robust human-like generalization to data with other visual characteristics or to tasks in other cognitive domains.
Cornelia McCormick, Cornelia McCormick, Sven Lange
et al.
Mental visual imagery, especially the ability to construct naturalistic scenes seems central to vivid episodic autobiographical memory (AM). This mini review will first highlight the neural anatomy of different aspects of mental imagery, focusing on the roles of the hippocampus, ventromedial prefrontal cortex and posterior neocortex and the consequences of damage to these regions to AM. We will then contrast the consequences of missing images for AM in two special populations with no apparent brain damage: Congenital Aphantasia (i.e., lack of visual imagery) and congenital blindness (i.e., lack of visual perception). We propose that Aphantasia leads to impaired scene construction and reduced AM reliving. Despite limited evidence on AM in congenitally blind individuals, they seem to rely on auditory and tactile information to construct (scene) imagery, which in turn may support vivid AM reliving. The main findings here suggest that mental scene imagery, rather than visual encoding, is crucial for AM. This conclusion has far-reaching implications for understanding memory disorders, mental health, and a call to protect our imagination.
Language is known to play a crucial role in influencing how humans perceive and categorize sensory stimuli, including odors. This study investigated the impact of linguistic labeling on odor categorization among bilingual participants proficient in Chinese (L1) and English (L2). We hypothesized that L1-like linguistic labels would more robustly propel the learning of new olfactory categories compared to a condition without language, and more familiar labels would better support odor category learning. The analysis focused on comparing learning trajectories and odor categorization performance of four groups, three in which odors were paired with different sets of verbal labels and a control group that categorized odors without any verbal labeling. Following four days of intensive training, the results showed that the groups with verbal labels numerically outperformed the control group, and that the less familiar the labels sounded the more successful categorization became. However, between-group differences did not reach statistical significance. These findings, while not conclusively supporting our hypotheses, provide insights into the complex relationship between linguistic familiarity and odor category formation. The results are nested within Ad Hoc Cognition, highlighting that variations in linguistic familiarity may not induce robust enough contextual changes to differentially affect how odor categories are formed.
Claudia A.F. Wascher, Gemini Waterhouse, Bret A. Beheim
Abstract Vocal mimicry, the copying of sounds produced by another species or the environment, is commonly described in vocal learners, such as songbirds. Understanding the functions of vocal mimicry can help to uncover the evolutionary drivers of vocal learning. Different adaptive functions like interspecific and intraspecific communication have been suggested, as well as the possibility of vocal mimicry to be a mistake during vocal learning. In the present study, we review the occurrence of mimicry in the family of corvids and investigate the socio-ecological factors driving the evolution of vocal mimicry in this group of birds. We recorded evidence of vocal mimicry from primary (xeno-canto recordings) as well as secondary sources (published literature) and found evidence for vocal mimicry in 39 out of 128 corvid species (30%). Socio-ecological factors like breeding system, habitat and trophic niche did not have a significant effect. We used Bayesian modelling based on existing data from primary and secondary sources to estimate the occurrence of mimicry, suggesting that vocal mimicry may be more widespread among corvids than currently documented, with many species potentially being ‘hidden mimics’. Our study for the first time systematically reviews the occurrence of vocal mimicry across the family of corvids and investigates a range of socio-ecological factors driving the behaviour, hopefully inspiring future field work on those species.
While several therapeutic options for insomnia are currently available, they often require long-term use and come with certain disadvantages. Given insomnia's significant impact on health overall, more effective treatments are warranted. Here, we report two patients with moderate to severe insomnia whose symptoms significantly improved following the intravenous administration of ex vivo-expanded bone marrow-derived mesenchymal stem cells (MSC).
Abstract The aim of the present study was to test the replicability of the text generation effect for learning with expository texts while systematically varying contextual factors that—based on extant literature—can be assumed to affect the occurrence and magnitude of the text generation effect. Seven experiments were conducted in which participants either read (control condition) or unscrambled sentences (generation condition) in expository texts. The experiments varied systematically on intentionality of learning, learning time constraint, retention interval, and study design. Contrary to expectations, no text generation effect could be found. Instead, some of the experiments even revealed a learning disadvantage for text generation compared to the reading control condition. In only one experiment (Experiment 6) and for just one of the learning measures, learners performed better when they had generated the texts. In sum, the results indicate that a generation effect is most likely to occur when learning is intentional, when learning time is unrestricted, and for immediate testing. The findings suggest that the applications of text generation in educational contexts are rather limited.
Semantic equivalence in the affective domain is always a matter of degree, even for the words that may seem uncontroversial. For example, a word may be quoted in dictionaries as the semantic equivalent of another word and be used in practice as its most frequent translation equivalent, and yet those two words may significantly differ in meaning. This study focuses on one such case – that of the English term frustration and its cognates in Spanish (frustración), French (frustration) and German (Frustration). Using data from corpora and self-report, we find that, while frustration terms in Spanish, French and German reflect a cross-culturally stable type of low-power anger, or can denote affective experiences other than anger, English frustration refers to a prototypical anger experience characterized by high power. Converging evidence is presented from two psycholinguistic and two linguistic studies employing elicited and observational data. We offer a possible explanation for the observed semantic differences based on psychological appraisal theory and cross-cultural psychology. The novelties and limitations of our findings are discussed, along with their implications for researchers in the affective sciences.
Large Language Models (LLMs) have demonstrated remarkable capabilities in performing tasks across various domains without needing explicit retraining. This capability, known as In-Context Learning (ICL), while impressive, exposes LLMs to a variety of adversarial prompts and jailbreaks that manipulate safety-trained LLMs into generating undesired or harmful output. In this paper, we propose a novel interpretation of ICL in LLMs through the lens of cognitive neuroscience, by drawing parallels between learning in human cognition with ICL. We applied the principles of Cognitive Load Theory in LLMs and empirically validate that similar to human cognition, LLMs also suffer from cognitive overload a state where the demand on cognitive processing exceeds the available capacity of the model, leading to potential errors. Furthermore, we demonstrated how an attacker can exploit ICL to jailbreak LLMs through deliberately designed prompts that induce cognitive overload on LLMs, thereby compromising the safety mechanisms of LLMs. We empirically validate this threat model by crafting various cognitive overload prompts and show that advanced models such as GPT-4, Claude-3.5 Sonnet, Claude-3 OPUS, Llama-3-70B-Instruct, Gemini-1.0-Pro, and Gemini-1.5-Pro can be successfully jailbroken, with attack success rates of up to 99.99%. Our findings highlight critical vulnerabilities in LLMs and underscore the urgency of developing robust safeguards. We propose integrating insights from cognitive load theory into the design and evaluation of LLMs to better anticipate and mitigate the risks of adversarial attacks. By expanding our experiments to encompass a broader range of models and by highlighting vulnerabilities in LLMs' ICL, we aim to ensure the development of safer and more reliable AI systems.
One of the fundamental aspects of cognitive architectures is their ability to encode and manipulate knowledge. Without a consistent, well-designed, and scalable knowledge management scheme, an architecture will be unable to move past toy problems and tackle the broader problems of cognition. In this paper, we document some of the challenges we have faced in developing the knowledge stack for the Companion cognitive architecture and discuss the tools, representations, and practices we have developed to overcome them. We also lay out a series of potential next steps that will allow Companion agents to play a greater role in managing their own knowledge. It is our hope that these observations will prove useful to other cognitive architecture developers facing similar challenges.
Kazuki Sawada, Hikaru Koike, Arata Murayama
et al.
Art appreciation evokes feelings of being moved, which, in turn, may inspire creative activities. However, the experience of being aesthetically moved can involve various processes, and it remains unclear which appreciation processes specifically lead to being moved and inspired. This experiment investigated whether three appreciation processes—perceived beauty, feelings of awe, and meaning making—facilitate feelings of inspiration through feelings of being moved. Participants (33 undergraduate and graduate students; 14 females and 19 meles; mean age = 21.48, SD = 2.28) were instructed to appreciate paintings by either inferring their meaning or evaluating their beauty, and were then required to view a painting and draft a short creative story about it. They were also asked to report the extent to which they felt inspired during the writing phase and felt moved, perceived beauty, awe, and meaning-making during the appreciation phase. Mediation analyses indicated that meaning-making promoted feelings of inspiration through increased feelings of being moved in both the meaning inference and impression evaluation conditions, and that feelings of awe promoted feelings of inspiration in the meaning inference condition, but not in the impression evaluation condition. Furthermore, correlation and partial correlation analyses indicated that perceived beauty was not significantly correlated with feelings of inspiration. Our results provide a deeper understanding of the psychological mechanisms of inspiration through art appreciation from the perspectives of feelings of being moved, awe, and meaning-making.
This article explores the nature and dynamics of mnemonic communities within the context of social media platforms and proposes to identify mnemonic communities using hashtag co-occurrence analysis. The article distinguishes between ‘explicit’ and ‘latent’ mnemonic communities, arguing that while some digital mnemonic communities may exhibit characteristics of offline communities, others exist latently as discursive spaces or semiospheres without direct awareness. On platforms like Instagram, hashtags function as semiotic markers, but also as user-chosen indexes to the content. As hashtags link the social and semantic aspects of community formation, hashtag co-occurrence analysis offers a robust framework for understanding and mapping these communities. This method allows to detect and analyse patterns of hashtag use that suggest the presence of networked community structures that may not be apparent or conscious to the social media users themselves. Additionally, a metric is introduced for determining the degree of ‘latentness’ of communities that quantifies the cohesion within communities compared to their external connections. The article demonstrates this approach by applying hashtag co-occurrence analysis to a dataset of Instagram posts tagged with #Juneteenth, a popular hashtag used to commemorate the ending of slavery in the United States. It identifies 87 mnemonic communities that reflect the diversity and complexity of how platforms facilitate memory-sharing practices and the role of semiotic markers in forming (latent) mnemonic networks.
Communication. Mass media, Consciousness. Cognition
With the deterioration of the marine environment today, in order to promote the management of marine plastic waste, relevant influencing factors are explored to better enhance public participation in the management of marine plastic waste. Using the fuzzy set qualitative comparative analysis method (fsQCA), this study explores the influencing factors of public participation in marine plastic waste management in Shanghai. The results indicate that: (1) In the configuration of factors influencing participation in governance, environmental cognition, policy identification, participation awareness, and participation pathways all appear as core conditions. (2) Environmental cognition and participation consciousness appear as core conditions in both configurations, playing a major role; Economic level, policy identification, environmental behavior, and participation pathways play an auxiliary role in influencing governance factors. (3) Based on the controllability of conditional variables, strengthening public awareness of the marine environment and increasing public participation awareness are effective ways to enhance the influencing factors of marine plastic waste management. The combined effect of multiple factors reveals the complexity of factors influencing public participation in governance. Strengthening public environmental awareness and participation awareness is an effective choice to enhance public participation in governance.
This work investigates online learning techniques for a cognitive radar network utilizing feedback from a central coordinator. The available spectrum is divided into channels, and each radar node must transmit in one channel per time step. The network attempts to optimize radar tracking accuracy by learning the optimal channel selection for spectrum sharing and radar performance. We define optimal selection for such a network in relation to the radar observation quality obtainable in a given channel. This is a difficult problem since the network must seek the optimal assignment from nodes to channels, rather than just seek the best overall channel. Since the presence of primary users appears as interference, the approach also improves spectrum sharing performance. In other words, maximizing radar performance also minimizes interference to primary users. Each node is able to learn the quality of several available channels through repeated sensing. We define hybrid cognition as the condition where both the independent radar nodes as well as the central coordinator are modeled as cognitive agents, with restrictions on the amount of information that can be exchanged between the radars and the coordinator. Importantly, each part of the network acts as an online learner, observing the environment to inform future actions. We show that in interference-limited spectrum, where the signal-to-interference-plus-noise ratio varies by channel and over time for a target with fixed radar cross section, a cognitive radar network is able to use information from the central coordinator in order to reduce the amount of time necessary to learn the optimal channel selection. We also show that even limited use of a central coordinator can eliminate collisions, which occur when two nodes select the same channel.
Visual Question Answering (VQA) is a challenging task that requires cross-modal understanding and reasoning of visual image and natural language question. To inspect the association of VQA models to human cognition, we designed a survey to record human thinking process and analyzed VQA models by comparing the outputs and attention maps with those of humans. We found that although the VQA models resemble human cognition in architecture and performs similarly with human on the recognition-level, they still struggle with cognitive inferences. The analysis of human thinking procedure serves to direct future research and introduce more cognitive capacity into modeling features and architectures.
Yonatan Sanz Perl, Carla Pallavicini, Juan Piccinini
et al.
Summary: Brain states are frequently represented using a unidimensional scale measuring the richness of subjective experience (level of consciousness). This description assumes a mapping between the high-dimensional space of whole-brain configurations and the trajectories of brain states associated with changes in consciousness, yet this mapping and its properties remain unclear. We combine whole-brain modeling, data augmentation, and deep learning for dimensionality reduction to determine a mapping representing states of consciousness in a low-dimensional space, where distances parallel similarities between states. An orderly trajectory from wakefulness to patients with brain injury is revealed in a latent space whose coordinates represent metrics related to functional modularity and structure-function coupling, increasing alongside loss of consciousness. Finally, we investigate the effects of model perturbations, providing geometrical interpretation for the stability and reversibility of states. We conclude that conscious awareness depends on functional patterns encoded as a low-dimensional trajectory within the vast space of brain configurations.
Adi Wijaya, Noor Akhmad Setiawan, Asma Hayati Ahmad
et al.
Background: Mild cognitive impairment (MCI) is often considered a precursor to Alzheimer's disease (AD) due to the high rate of progression from MCI to AD. Sensitive neural biomarkers may provide a tool for an accurate MCI diagnosis, enabling earlier and perhaps more effective treatment. Despite the availability of numerous neuroscience techniques, electroencephalography (EEG) is the most popular and frequently used tool among researchers due to its low cost and superior temporal resolution. Objective: We conducted a scoping review of EEG and MCI between 2012 and 2022 to track the progression of research in this field. Methods: In contrast to previous scoping reviews, the data charting was aided by co-occurrence analysis using VOSviewer, while data reporting adopted a Patterns, Advances, Gaps, Evidence of Practice, and Research Recommendations (PAGER) framework to increase the quality of the results. Results: Event-related potentials (ERPs) and EEG, epilepsy, quantitative EEG (QEEG), and EEG-based machine learning were the research themes addressed by 2310 peer-reviewed articles on EEG and MCI. Conclusion: Our review identified the main research themes in EEG and MCI with high-accuracy detection of seizure and MCI performed using ERP/EEG, QEEG and EEG-based machine learning frameworks.