Cognitive Architecture Research Articles (Page 1)

People often hesitate to rely on algorithmic advice, even when it is objectively more accurate than human input-a phenomenon known as algorithm aversion. In two experiments, we investigated the cognitive mechanisms underlying this effect in a clinical decision-making context. Participants evaluated x-rays for bone fractures, with each image accompanied by advice purportedly from either an algorithm or a human source. Across experiments, we observed longer response times for algorithmic advice, indicating increased deliberation. Evidence accumulation modeling revealed that participants set higher decision thresholds when evaluating algorithmic advice, reflecting a more cautious decision strategy. This hesitancy, observed when the human advice was attributed to lay participants (Experiment 1), persisted when the human advice was attributed to expert radiologists (Experiment 2). Accumulation rates and prior preferences did not differ across advisor types, suggesting that algorithm aversion stems specifically from increased caution rather than reduced perceived reliability. These findings demonstrate that algorithm aversion manifests as a strategic shift in decision-making and highlight the value of formal cognitive models for understanding trust in artificial intelligence. Our findings advance the theoretical understanding of algorithm aversion by identifying response caution as a core mechanism. More broadly, the results demonstrate how formal models of decision-making can clarify the cognitive architecture of trust in automated systems, offering a foundation for future work on optimizing human-algorithm collaboration.

Embodied intelligent systems build upon the foundations of behavioral robotics and classical cognitive architectures. They integrate multimodal perception, world modeling, and adaptive control to support closed-loop interaction in dynamic and uncertain environments. Recent breakthroughs in Multimodal Large Models (MLMs) and World Models (WMs) are profoundly transforming this field, providing the tools to achieve its long-envisioned capabilities of semantic understanding and robust generalization. Targeting the central challenge of how modern MLMs and WMs jointly advance embodied intelligence, this review provides a comprehensive overview across key dimensions, including multimodal perception, cross-modal alignment, adaptive decision-making, and Sim-to-Real transfer. Furthermore, we systematize these components into a three-stage theoretical framework termed “Dynamic Perception–Task Adaptation (DP-TA)”. This framework integrates multimodal perception modeling, causally driven world state prediction, and semantically guided strategy optimization, establishing a comprehensive “perception–modeling–decision” loop. To support this, we introduce a “Feature-Conditioned Modal Alignment (F-CMA)” mechanism to enhance cross-modal fusion under task constraints.

Optimizing the Connectionist Dual-Process Model of Reading Aloud (CDP; Perry et al., Journal of Memory and Language, 134 , 104468) using large-scale empirical datasets has been shown to enable accurate predictions of independent datasets that were not used for optimization. Here, we investigated CDP’s generalization performance when optimized on small datasets consisting of words, nonwords, or a combination of both. The results showed CDP’s quantitative performance was similar on both small and large datasets except when optimized on small nonword-only datasets. Additionally, CDP’s predictions generally surpassed those derived from regression-based models, suggesting it had good generalization performance. Using sloppy parameter analyses, we also found that a small number of parameters determined most of CDP’s quantitative performance and that the parameters which did this were similar across both small and large datasets. These findings suggest that the CDP does not overfit the data, even when optimized on very small numbers of stimuli. They also give insight into the role the parameters play in generating psycholinguistic effects. More generally, the findings show that when an underlying cognitive architecture constrains behavior, complex systems like reading may be analyzed and understood using very limited data. This is important as it shows that computational modelling can be used in some situations where data is scarce but understanding the system remains crucial.

Research claims that metacognitive experiences can be classified as types of metacognitive regulation. Formulated in terms of the theory of Attention as Internal Action, this view raises questions about the timing of metacognitive experiences that occur in response to internal experiences. To investigate these questions, this work presents a method for cognitive computation that simulates consecutive internal decisions occurring during the process of taking a digital exam. A new version of the General Internal Model of Attention is proposed and supported by research. It is applied as cognitive architecture in a simulation system to reproduce cognitive phenomena such as the cognitive cycle, internal decision-making, imagery, body actions, learning, and metacognition. Two corresponding groups of Markov Decision Processes were designed as information stores for goal influence and learning, and a Hebbian machine learning algorithm was applied as an operator on the learning models. The timing and consecutiveness of metacognitive experiences were analyzed based on the cognitive cycle results, and several hypotheses were derived. One of them suggests that the first engagement in a metacognitive experience for each question in the exam is delayed over the course of the exam-taking process.

Introduction As cyber-physical systems become increasingly virtualized, digital twins have emerged as essential components for real-time monitoring, simulation, and control. However, their growing complexity and exposure to dynamic network environments make them vulnerable to sophisticated cyber threats. Traditional rule-based and machine-learning-based security models often fail to adapt in real time to evolving attack patterns, particularly in decentralized and resource-constrained settings. Methods This study introduces the Neuromorphic Cyber-Twin (NCT), a brain-inspired architectural framework that integrates spiking neural networks (SNNs) and event-driven cognition to enhance adaptive cyber defense. The NCT leverages neuromorphic principles such as sparse coding, temporal encoding, and spike-timing-dependent plasticity (STDP) to transform telemetry data from the digital-twin layer into spike-based sensory inputs. A layered cognitive architecture continuously monitors behavioral deviations, infers anomalies, and autonomously adapts its defensive responses in alignment with system dynamics. Results Lightweight prototype simulations demonstrate the feasibility of NCT-based event-driven anomaly detection and adaptive defense. The results highlight advantages in low-latency detection, contextual awareness, and energy efficiency compared with conventional machine-learning models. Discussion The NCT framework represents a biologically inspired paradigm for scalable, self-evolving cybersecurity in virtualized ecosystems. Potential applications include infrastructure monitoring, autonomous transportation, and industrial control systems. Comprehensive benchmarking and large-scale validation are identified as future research directions.

Does epistemic diversity necessarily promote scientific progress, or does this idea persist more as a normative principle than as an empirical finding? In this hypothesis-essay, I argue that scientific pluralism does not automatically entail epistemic fluidity. Distinct scientific communities may share publication venues and rhetorical commitments while remaining epistemically insulated. Building on recent debates about epistemic bubbles and echo chambers, I suggest that even within science, traditionally conceived as a self-correcting enterprise, mechanisms of selective exposure and institutional filtering can restrict genuine epistemic permeability. Frequently presented as an inherently diverse discipline, ethnobiology offers a revealing context for exploring whether declared plurality translates into dialogical openness or stabilizes into parallelism. The argument does not treat ethnobiology as a confirmed case, but as a field in which this hypothesis of pluralism without fluidity can be examined. I invite a reconsideration of how epistemic contact, reflexivity, and institutional design shape the moral and cognitive architecture of scientific progress.

Large Language Models (LLMs) offer new opportunities to devise automated implementation generation methods that can tackle problem solving beyond traditional methods, which usually require algorithmic specifications and use only static domain knowledge. LLMs can support devising new methods to support activities in tackling open-ended problems, like problem framing, exploring possible solving approaches, feature elaboration and combination, advanced implementation assessment, and handling unexpected situations. This paper presents a detailed overview of the current work on LLMs, including model prompting, retrieval-augmented generation (RAG), and reinforcement learning. It then proposes a novel, LLM-based Cognitive Architecture (CA) to generate programming code starting from verbal discussions in natural language, a particular kind of problem-solving activity. The CA uses four strategies, three top-down and one bottom-up, to elaborate, adaptively process, memorize, and learn. Experiments are devised to study the CA performance, e.g., convergence rate, semantic fidelity, and code correctness.

Enabling human–CPS cognitive interoperability: Cognitive architectures as technologies for human-like cognitive digital twins

Dynamic spectrum access enables efficient anti-jamming in cognitive radio systems. However, in a multi-user distributed decision scenario, differences in spectrum states make collaboration among users a major challenge, especially when the sensing devices are heterogeneous. In order to solve this issue, we propose a collaborative anti-jamming cognitive radio system architecture based on historical jamming knowledge. Devices exhibiting high sensing performance support those exhibiting low sensing performance. An online reinforcement learning algorithm is used to learn the jamming patterns in real time. Finally, a multi-user collaborative anti-jamming system is developed using a software-defined radio platform. The anti-jamming performance of the system is verified experimentally under both internal communication jamming and external malicious jamming scenarios, achieving a jamming probability below 0.1.

Detecting available road space is a fundamental task for autonomous driving vehicles, requiring robust image feature extraction methods that operate reliably across diverse sensor-captured scenarios. However, existing approaches process each input independently without leveraging Accumulated Experiential Knowledge (AEK), limiting their adaptability and reliability. In order to explore the impact of AEK, we introduce MemRoadNet, a Memory-Augmented (MA) semantic segmentation framework that integrates human-inspired cognitive architectures with deep-learning models for free road space detection. Our approach combines an InternImage-XL backbone with a UPerNet decoder and a Human-like Memory Bank system implementing episodic, semantic, and working memory subsystems. The memory system stores road experiences with emotional valences based on segmentation performance, enabling intelligent retrieval and integration of relevant historical patterns during training and inference. Experimental validation on the KITTI road, Cityscapes, and R2D benchmarks demonstrates that our single-modality RGB approach achieves competitive performance with complex multimodal systems while maintaining computational efficiency and achieving top performance among single-modality methods. The MA framework represents a significant advancement in sensor-based computer vision systems, bridging computational efficiency and segmentation quality for autonomous driving applications.

Purpose: Decision-making under uncertainty remains a foundational challenge in cognitive science and artificial intelligence. Classical Bayesian Probability Models (CBM) often fail to explain paradoxical cognitive behaviors such as order effects, ambiguity aversion, and context-dependent reasoning. This study seeks to compare Quantum Probability Theory (QPT) and Classical Bayesian Models in their ability to capture the dynamics of human decision-making. It aims to determine which framework more accurately reflects the cognitive mechanisms underlying reasoning under uncertainty. Methodology: A qualitative, exploratory research design was adopted, involving in-depth semi-structured interviews with 16 experts across psychology, philosophy, artificial intelligence, and cognitive neuroscience. Participants were purposively selected for their theoretical and empirical expertise in probabilistic reasoning. Data were analyzed using reflexive thematic analysis, guided by the Dual-Process Theory and Busemeyer’s Quantum Cognition framework. The analysis emphasized participants’ perspectives on theoretical assumptions, cognitive plausibility, and predictive utility between QPT and CBM paradigms. Findings: Thematic findings reveal that Quantum Probability Theory offers superior explanatory power in contexts involving cognitive ambiguity, contextual dependence, and non-commutativity of mental operations. Participants consistently reported that QPT better models real-world reasoning tasks where classical logic collapses, capturing the fluid and context-sensitive nature of human judgment. Conversely, while CBM remains effective in structured, low-uncertainty scenarios, it fails to accommodate superposition and interference effects inherent in human cognition. Unique Contribution to Theory, Practice, and Policy (Recommendations): The study contributes theoretically by demonstrating how quantum probabilistic models expand existing theories of bounded rationality and probabilistic reasoning in cognitive science. Practically, it encourages interdisciplinary collaboration between cognitive scientists, AI researchers, and philosophers to refine decision models that mirror human intuition more closely. Policy-wise, the findings support the integration of quantum-inspired approaches in the design of intelligent decision-support systems and cognitive architectures. The study recommends continued empirical validation of QPT within applied domains—such as behavioral economics, machine learning, and cognitive modeling—to strengthen its predictive and explanatory robustness.

In post-disaster search and rescue (SAR) operations, unmanned aerial vehicles (UAVs) are essential tools, yet the large volume of raw visual data often overwhelms human operators by providing isolated, context-free information. This paper presents an innovative system with a novel cognitive–agentic architecture that transforms the UAV from an intelligent tool into a proactive reasoning partner. The core innovation lies in the LLM’s ability to perform high-level semantic reasoning, logical validation, and robust self-correction through internal feedback loops. A visual perception module based on a custom-trained YOLO11 model feeds the cognitive core, which performs contextual analysis and hazard assessment, enabling a complete perception–reasoning–action cycle. The system also incorporates a physical payload delivery module for first-aid supplies, which acts on prioritized, actionable recommendations to reduce operator cognitive load and accelerate victim assistance. This work, therefore, presents the first developed LLM-driven architecture of its kind, transforming a drone from a mere data-gathering tool into a proactive reasoning partner and demonstrating a viable path toward reducing operator cognitive load in critical missions.

Reinforcement learning at the interface of artificial intelligence and cognitive science.

This study examines how domain-general (processing speed and receptive vocabulary) and domain-specific (symbolic and non-symbolic comparison) cognitive skills contribute to early informal mathematical thinking in preschoolers. The aim was to assess the invariance of these predictive relationships across two sociocultural contexts: Chilean and Spanish samples. A total of 130 children participated, and structural equation modeling was used to estimate latent structures and test multigroup invariance. The results revealed a consistent latent structure across samples and a significant contribution of symbolic and non-symbolic comparison to early math performance, while processing speed and vocabulary showed context-specific variations. These findings indicate that although foundational mathematical competencies rely on common cognitive mechanisms, cultural and educational contexts modulate the strength of these associations. This study contributes to understanding the cognitive architecture underlying early numeracy and highlights the importance of culturally sensitive assessment and intervention strategies.

NEW & NOTEWORTHY Essential to critical thinking skill development in human physiology students is a focus on the causal relationships between sequential elements of physiological processes. Causality is essential to our predictive abilities within complex systems. In essence, physiological sequences should be explained as a series of logical "triads" (stimulus causally linked to effect). This helps limit the short-term memory load, enhancing the storage of long-term memory and the formation of essential cognitive architecture required for critical thinking.

Abstract The majority of theories and models of social influence tend to focus on the social-behavioral level and implicitly discount the potential role of cognitive explanations to ground out social phenomena in cognitive operations. The present study describes a preliminary simulation of social influence and conformity using three possible cognitive mechanisms: the first is a homophily-based model that weighs belief updating based on generating a latent trust magnitude relying entirely on strategy selection, and the second two are novel similarity-learning and structure-mapping mechanisms embodied in the ACT-R cognitive architecture (Anderson 1990) that spread association via learned similarities based on communicating similar beliefs from other agents. Across three simulations (homophily, similarity, similarity+structure mapping), the homophily model was the only model capable of capturing both influence and conformity effects. The similarity model quickly flipped based on the statistics of the messages (i.e., more opposing messages than supporting messages) while the similarity+structure mapping eventually led to a convergence and averaging across beliefs. Each model exhibited initial learning and larger belief updating while settling to a relatively-stable state over time. While more research is required, the current simulation implies that architectural solutions may not be sufficiently robust to adequately account for influence and conformity effects, while the homophily-based trust model was able to model these effects driven by the dynamics of human memory.

Resolving the dichotomy between the human-like yet constrained reasoning processes of cognitive architectures (CAs) and the broad but often noisy inference behavior of large language models (LLMs) remains a challenging yet exciting pursuit, aimed at enabling reliable machine reasoning capabilities in LLMs. Previous approaches that employ off-the-shelf LLMs in manufacturing decision-making face challenges in complex reasoning tasks, often exhibiting human-level yet unhuman-like behaviors due to insufficient grounding. This present article start to address this gap by asking whether LLMs can replicate cognition from CAs to make human-like decisions. We introduce cognitive LLMs , which are hybrid decision-making architectures comprised of a CA and an LLM through a knowledge transfer mechanism LLM-ACTR . Cognitive LLMs extract and embed knowledge of CA’s internal decision-making process as latent neural representations, inject this information into trainable LLM adapter layers, and fine-tune the LLMs for downstream prediction tasks. We find that, after knowledge transfer through LLM-ACTR , the cognitive LLMs offers better representations of human decision-making behaviors on a novel design for manufacturing problem, compared to an LLM-only model that employs chain-of-thought. Taken together, the results open up new research directions for equipping LLMs with the necessary knowledge to computationally model and replicate the internal mechanisms of human cognitive decision-making. We release the code and data samples at https://github.com/SiyuWu528/LLM-ACTR .

Abstract Inspired by the ”thinking fast and slow” cognitive theory of human decision making, we propose a multi-agent cognitive architecture (SOFAI) that is based on ”fast”/”slow” solvers and a metacognitive module. We then present experimental results on the behavior of an instance of this architecture for AI systems that make decisions about navigating in a constrained environment. We show that combining the two decision modalities through a separate metacognitive function allows for higher decision quality with less resource consumption compared to employing only one of the two modalities. Analyzing how the system achieves this, we also provide evidence for the emergence of several human-like behaviors, including skill learning, adaptability, and cognitive control.

Predicting Drivers’ situation awareness and response times in the emergency situation using an integrated cognitive architecture

This article challenges the dominant paradigm of “situational modification” in productivity discourse, which advocates environmental control through eliminating technological distractions. Drawing on theories of Extended Mind (Clark & Chalmers) and 4E cognition (embodied, embedded, extended, enactive), we propose an alternative conceptualization of tools as “probes”—dynamic agents that catalyze subject transformation through productive tension between new possibilities and mastery challenges. Within the framework of Eco-Centered Psychological Facilitation (ECPF), we develop a Six-Phase Model of Probe Mastery (6PMPM): attraction, frustration, tension, transition, integration, and emergence. Each phase represents distinct patterns of cognitive-somatic experience essential for genuine transformation rather than mere skill acquisition. Special attention is given to artificial intelligence as a paradigmatic contemporary probe, requiring fundamental restructuring of cognitive architecture rather than simple technical adaptation. Our research reveals the “mirror crisis” phenomenon—a specific pattern where generative AI externalizes users' thinking patterns, creating unprecedented conditions for metacognitive awareness and transformation. The probe concept has significant implications for education, psychological practice, and organizational development, suggesting a shift from defensive strategies that limit technological exposure to active integration approaches where tools become catalysts for expanding human potential. While acknowledging limitations including individual variability and cultural specificity, this framework offers a productive perspective for understanding human-technology co-evolution in an era of rapid change. The choice between avoidance and integration strategies represents a fundamental existential decision about human development direction. We argue for conscious engagement with tool-probes as a path toward co-creative becoming, where each new technology becomes not a threat to identity but an invitation to expand human potential through deliberate cognitive transformation.