eBICA: Emotional Cognitive Architecture

• eBICA is a cognitive architecture that unifies biologically-based emotion with cognitive processes through dynamic, overlapping networks.
• It employs neuromodulator-inspired gating and salience tagging to modulate learning, memory, and attention in real time.
• Implemented in diverse domains, it shows promise in resource management, smart automation, and adaptive tutoring systems.

An Emotional Biologically Inspired Cognitive Architecture (eBICA) is a class of computational systems designed to integrate biologically grounded models of emotion with cognitive processing, facilitating attention, learning, memory, and action selection in ways that parallel vertebrate brains. Unlike architectures that treat emotion as a detached module, eBICA frameworks instantiate emotion as an intrinsic, pervasive influence on perception, decision-making, resource allocation, memory consolidation, and behavior, drawing directly from principles in neurobiology, affective neuroscience, and systems theory. Instances of eBICA unify mechanistic models of neuromodulator-based control, salience-gated attention, emotion-augmented memory, and motivational drives, and operationalize these in distributed, modular, or hybrid software and hardware frameworks.

1. Architectural Principles and Core Modules

Fundamentally, eBICA design mandates the integration of emotion and cognition at all system levels, eschewing strict module boundaries in favor of dynamic, overlapping networks or microservice-based distributed modules. Notable implementations include:

• Distributed/Modular Designs: For example, an architecture composed of independently deployable servers for arousal (neuromodulator modeling), thalamic gating (sensory filtering), and cortical processing (perceptual, classification, contextual prediction, emotional tagging) (Remmelzwaal et al., 2020). RESTful APIs facilitate orchestration and modular replacement.
• Intrinsic Neuromodulatory Control: Synthetic analogs to neuromodulators (e.g., noradrenaline, dopamine, serotonin) pervasively bias activation thresholds, gating, scheduling, and memory access across all cognitive processes—mirroring diffuse monoaminergic projections in mammalian brains (Vallverdú et al., 2016, Béroule et al., 2019).
• Overlapping Network Ethos: Rather than serial flow between discrete components, the non-modular eBICA instantiates overlapping, dynamically reconfigurable “valuation,” “priority map,” “executive control,” and “action selection” networks, wherein emotion and motivation percolate simultaneously through all activity (Pessoa, 2019).
• Memory-Episode Tagging: All events and episodes are tagged with affective vectors, so that current context and behavioral decisions are referenced against a history of emotionally scored experiences (0901.4963, Borotschnig, 1 May 2025).
• Appraisal–Somatic–Emotion—Action Loop: eBICA architectures frequently couple cognitive appraisal, real or simulated somatic feedback, emotional state update, and emotion-biased behavior selection into a recurrent control loop (Yamauchi et al., 25 Dec 2025).

The table below summarizes key structural elements as realized in several benchmark eBICA systems:

System/Framework	Biologically-Inspired Features	Emotional Integration Mode
(Remmelzwaal et al., 2020)	Modular servers: arousal, thalamus, cortex	Noradrenaline gating, salience-driven tagging, RESTful API
(Vallverdú et al., 2016)	Monoaminergic neuromodulation, Von Neumann	D/S/N bias CPU, memory, bus at microarchitecture level
(Pessoa, 2019)	Overlapping networks (“priority map,” etc.)	Affect modulates perception, memory, action competition
(0901.4963)	pseudo-hippocampus, pseudo-amygdala	Affective episode tagging, memory retrieval bias
(Yamauchi et al., 25 Dec 2025)	Appraisal, somatic, emotion feedback loop	Emotion as control state for smart-home action selection

2. Computation and Emotional Modulation Mechanisms

eBICA implementations formalize emotional impact using tractable mathematical and algorithmic models:

• Emotional Salience Tagging: Salience is typically computed as a weighted sum or dot product between network activations and learned salience weights, e.g.,

$E(t) = \mathbf{w}^\top \mathbf{a}(t)$

where $E(t)$ is the instantaneous emotional salience, triggering further neuromodulator release if above threshold (Remmelzwaal et al., 2020).

• Adaptive Gating via Neuromodulators: Neuromodulators like noradrenaline are modeled as dynamical variables (e.g., with exponential decay), modulating thresholds in sensory gating and attention circuits:

$$N(t+\Delta t) = N(t) e^{-\lambda \Delta t},\quad \tau(t) = \tau_0 - k N(t)$$

where elevated $N(t)$ lowers the gating threshold $\tau(t)$, increasing sensitivity to emotionally salient inputs (Remmelzwaal et al., 2020, Vallverdú et al., 2016).

• Priority and Utility Maps: Action selection is driven by priority scores aggregating bottom-up salience, top-down goal relevance, affective significance, and motivational significance:

$P(x) = w_S S(x) + w_G G(x) + w_E E(x) + w_M M(x)$

Utility $U_a$ of actions combines cognitive prediction and emotional/motivational biases (Pessoa, 2019).

• Guided Propagation Networks: Some eBICA variants employ chains of elementary processing units whose activation and “anticipatory flows” are controlled by neuromodulator analogs (dopamine, serotonin), adjusting excitability and context-weight in temporal logic chains (Béroule et al., 2019).
• Emotional Homeostasis via Task Allocation: Agents optimize the empirical emotion distribution toward a prespecified “character” (target distribution $\pi$), minimizing KL-divergence through constrained mixture models and short-term corrective motivational drives (Gros, 2021).
• Episode-based Memory Bias: Episodic traces are retained with affective vectors; retrieval mechanisms compute similarity between past and present, biasing action toward emotionally satisfying behaviors via pattern mining and memory consolidation algorithms (0901.4963, Borotschnig, 1 May 2025).

3. Learning, Memory, and Attention in eBICA

eBICA architectures incorporate emotion-modulated learning and episodic memory:

• Emotion-Biased Hebbian and Stentian Learning: Synaptic modification occurs preferentially when internal emotion neurons (e.g., eFEAR, ePLEASURE) are highly active, enabling context-sensitive plasticity and associative conditioning reminiscent of Pavlovian learning (McDonald et al., 2020).
• Multiple-Trace Episodic Memory Consolidation: Each event appends a new, affectively annotated trace; pattern mining identifies frequent event/valence temporal patterns (0901.4963). The consolidation process averages affective valences across pattern occurrences, ensuring both frequency and emotional intensity modulate retrieval weighting.
• Emotional Tagging for Priority Recall: When new events share characteristics with previously tagged “high-salience” episodes, eBICA architectures trigger heightened attention and resource allocation through dynamic threshold adaptation (Remmelzwaal et al., 2020).
• Predictive Coding with Affective Error Signals: Generative models in eBICA predict not only sensory signals but expected affective and motivational values; prediction error signals across these axes drive memory update and executive resource reallocation (Pessoa, 2019).

4. Biological and Theoretical Foundations

eBICA directly leverages neurobiological findings:

• Monoaminergic Control: Simulation of global, modulatory axonal projections from loci such as the substantia nigra or dorsal raphe, instantiated as synthetic dopamine, serotonin, and noradrenaline (Vallverdú et al., 2016, Béroule et al., 2019).
• Rich Club Topology and Overlap: Cortical/subcortical connectivity in the brain forms high-density overlapping “rich clubs,” supporting rapid, multi-pathway signal integration—mirrored in dynamic membership and non-modularity of eBICA networks (Pessoa, 2019).
• Embodiment and Somatic Loops: Physiological signals (e.g., heart rate, respiratory gases, facial expression) can be modeled directly, or via simulated “somatic markers,” feeding into emotional state estimation and behavior selection (Yamauchi et al., 25 Dec 2025, McDonald et al., 2020).
• Phylogenetic and Functional Abstraction: eBICA models exploit phylogenetic progression—from primitive binary affective states to domain-general, abstract valuation sets—enabling scalable architectures spanning simple organisms to advanced cognition (Gros, 2021).

5. Practical Applications and Empirical Validation

eBICA has been instantiated in a variety of operational and experimental contexts:

• Distributed Microservice Architectures: Modular flask-server instantiation allows for high extensibility, cross-machine deployment, and component-level research (Remmelzwaal et al., 2020).
• Resource Management in Computing Systems: NEUCOGAR demonstrates how neuromodulatory control can dynamically bias resource scheduling and memory allocation, validated with spiking neural network simulations (Vallverdú et al., 2016).
• Smart Environment Automation: Real-world smart-home proof-of-concept shows that eBICA emotion-guided automation (through appraisal, somatic markers, and target-goal bias) can significantly modulate user anxiety, with effect sizes $r=0.69$ (Main group) and $r=0.75$ (Long group), $p<.01$ for both, as measured by STAI-S (Yamauchi et al., 25 Dec 2025).
• Virtual Organisms and Associative Learning: Minimal bio-inspired virtual architectures, such as Ortus, display emotion-driven associative conditioning (e.g., water+fear learning), with direct modulation of synaptic weights measured as post-conditioning affective response boost $\Delta A\approx 0.6$ (McDonald et al., 2020).
• Episodic Memory-Guided Tutoring Systems: CTS showcases real-time improvement in behavioral selection in high-stakes procedural instruction, leveraging emotion-modulated frequent pattern mining to bias toward compassion-centric hinting paths (0901.4963).

6. Open Challenges, Extensions, and Theoretical Limits

Despite demonstrated success, several challenges and open questions are prominent within the eBICA paradigm:

• Modularity versus Overlap: The field debates the sufficiency of module-based grafts of emotion onto cognition—Pessoa's “Dolores test” posits that only architectures with inseparable, overlapping network-based emotion-cognition integration can achieve lifelike autonomy (Pessoa, 2019).
• Richness and Stability of Emotional Repertoires: Scaling to multi-valent or abstract synthetic emotions (beyond primitive “fear” and “pleasure”) is nontrivial given combinatoric growth and the need for compressed representations and dynamic balance (McDonald et al., 2020, Gros, 2021).
• Homeostatic and Meta-learning Modules: Incorporating adaptive character vectors, multi-timescale emotion hierarchies, and meta-learning to evolve target distributions (π) remains a future direction (Gros, 2021).
• Phenomenological Boundaries: Some eBICA models are argued to be “affective zombies”—displaying functionally rich emotional behavior without the architectural prerequisites for conscious affect or moral standing (Borotschnig, 1 May 2025). Architectural complexity, recursive self-models, and broadcasting global workspaces are identified as necessary (but not sufficient) preconditions for machine sentience.
• Robust Real-Time Integration: Integrating physiological (somatic), cognitive (appraisal), and motivational (goal/need) signals in real time is still a practical bottleneck in applied environments; current systems often rely on simplified proxies or staged input (Yamauchi et al., 25 Dec 2025).
• Experimental Standardization and Theory-Experiment Synchronization: Unlike classical architectures, many eBICAs lack widely accepted quantitative benchmarks or cross-task comparison frameworks, given their dependency on task context, affective scoring, and architectural reconfigurability.

7. Summary and Significance

eBICA establishes a biologically and theoretically validated basis for emotion-cognition unification in artificial intelligent systems, spanning micro- to macro-scale phenomena. Through the formalization of neuromodulatory control, salience-oriented attention, emotional tagging of episodic memory, and emotion-driven behavioral modulation, eBICAs offer a robust framework for constructing lifelike, contextually adaptive, and emotionally intelligent agents. Instantiations demonstrate advantages in selective computation, learning efficiency, and bio-realistic attention allocation, while the field continues to advance on issues of architecture complexity, emotional expressivity, and the delineation between functional mimicry and true affective awareness (Remmelzwaal et al., 2020, Vallverdú et al., 2016, Pessoa, 2019, 0901.4963, Gros, 2021, Borotschnig, 1 May 2025, McDonald et al., 2020, Yamauchi et al., 25 Dec 2025, Béroule et al., 2019).