Ray Mother–Infant Inter-brain Synchrony Algorithm for Deep Compassionate AI

Abstract

The Mother–Infant Inter-brain Synchrony Algorithm (MI-Sync-AI), developed at the Sri Amit Ray Compassionate AI Lab, is a computational framework inspired by the neurobiology of maternal–infant bonding. Neuroscientific research has demonstrated that shared gaze, touch, vocal prosody, and affective attunement foster measurable brain-to-brain synchrony between caregivers and infants. This synchrony forms the developmental foundation of trust, empathy, and co-regulation in human relationships.

The MI-Sync-AI system translates these principles into artificial intelligence by enabling agents to detect, model, and maintain synchrony-based interactions. Unlike conventional AI architectures optimized for efficiency or predictive accuracy alone, MI-Sync-AI emphasizes resonance, emotional alignment, and mutual regulation.

This article explores the theoretical background, covering the neuroscience of inter-brain synchrony, developmental and behavioral foundations, operational definitions, and sensor modalities, as outlined in the 20-point descriptive report. These foundations guide the algorithm’s design and its preprocessing pipeline, ensuring robust, ethical, and clinically relevant applications.

The algorithm has direct applications in healthcare, education, social robotics, and therapeutic interventions, where compassionate responsiveness is essential. This paper outlines the theoretical foundations, computational pipeline, and ethical implications of MI-Sync-AI as a cornerstone in the emerging field of Compassionate AI.

Introduction

Human development begins with the profound synchrony that emerges between a mother and her infant. Neuroscience studies reveal that coordinated rhythms of gaze, touch, voice, and affective expression generate inter-brain coupling that supports emotional bonding, stress regulation, and learning. These early dyadic interactions demonstrate how synchrony is not merely behavioral but a neurobiological process shaping the architecture of trust and empathy.

The Ray MI-Sync-AI Algorithm seeks to computationally replicate this mother–infant synchrony by embedding its mechanisms into artificial systems. It moves beyond signal analysis and prediction, advancing toward AI that can resonate with human users, adaptively co-regulate responses, and express compassion. By modeling the biological blueprint of synchrony, the algorithm equips AI agents—including social robots, healthcare companions, and therapeutic systems—to respond not only intelligently but empathetically.

The aims of this article are threefold: (1) to describe the neuroscientific foundations of inter-brain synchrony as observed in mother–infant interactions, (2) to present the computational design and pipeline of MI-Sync-AI with emphasis on synchrony detection, Transformer-based encoding, and Mutual Regulation (MoR), and (3) to explore its broader implications for Compassionate AI, including ethical safeguards and future applications in society.

 “Mother-infant inter-brain synchrony algorithms of compassionate artificial intelligence and social robots are developed based on brain-to-brain synchrony.” – Sri Amit Ray.

This principle guides us toward AI that truly cares.

 

Introduction | Neuroscience of Inter-brain Synchrony | Developmental and Behavioral Foundations | Operational Definition for MI-Sync-AI | Sensors & Modalities | References

 

Main Components of MI-Sync-AI

1. Neural Synchrony Modeling

This component replicates how oscillatory rhythms in different brains align during interaction. Using measures like coherence, cross-correlation, and phase-locking, the AI system simulates temporal coupling of signals across multiple agents.

2. Affective Signal Encoding

Emotional cues from multimodal sources (voice tone, facial expressions, gestures, heart-rate variability) are encoded into high-dimensional vector spaces. This enables the AI to recognize not only factual states but also emotional urgency.

3. Mutual Regulation (MoR) Module

Inspired by how mothers soothe distressed infants, the MoR module continuously adjusts AI responses to reduce synchrony gaps. This ensures that the AI outputs evolve in resonance with human or agent states, fostering comfort and stability.

4. Compassionate Attention Mechanism

Going beyond statistical attention, this mechanism prioritizes signals based on emotional vulnerability and need. For example, an infant’s cry—or a patient’s pain signal—receives higher response weight than neutral signals.

Modalities of the Algorithm

The algorithm draws from five modalities inspired by natural mother-infant bonds. Here’s a detailed breakdown:

Modality Description Sensors/Techniques Role in Synchrony
Gaze Tracks eye contact for emotional engagement. Eye-tracking cameras (e.g., webcams). AI mirrors user’s focus to build rapport [10].
Facial Expressions Analyzes emotions like joy or distress. Computer vision (e.g., OpenCV). Syncs AI responses to user’s mood [11].
Touch Detects physical gestures. Haptic sensors in wearables. Simulates comforting feedback [12].
Heart Rhythms Monitors HRV for stress levels. Smartwatch sensors. Aligns AI’s pacing with user’s state [13].
Voice Captures tone, pitch, rhythm, and prosody of speech for emotional resonance. Microphones with acoustic analysis (MFCC, spectrograms, deep audio embeddings). AI adjusts speech rhythm and tone to co-regulate affective states.

These components feed into a synchrony engine, computing alignment using metrics like phase-locking value (PLV) [14].

Algorithm Steps

The algorithm runs in a real-time loop, much like a nurturing conversation:

  1. Data Collection: Gather inputs over 5-10 seconds.
  2. Feature Extraction: Convert data into vectors for analysis.
  3. Synchrony Calculation: Measure alignment with AI’s state using similarity metrics.
  4. State Adjustment: Optimize AI’s responses via learning algorithms, biased toward compassion.
  5. Response Generation: Output empathetic actions or words.
  6. Feedback Loop: Refine based on user changes.

This process ensures AI evolves with the user, promoting positive emotional outcomes.

Pseudocode Implementation

To bring this to life, here’s a Python-style pseudocode based on Ray’s frameworks:

import numpy as np
import torch
from sklearn.metrics.pairwise import cosine_similarity

class MotherInfantSynchronyAI:
    def __init__(self):
        self.ai_state = torch.nn.Module()  # RNN for emotional state
        self.synchrony_threshold = 0.7

    def collect_inputs(self, time_window):
        # Placeholder for gathering data
        return {'gaze': ..., 'face': ..., 'touch': ..., 'heart': ...}

    def extract_features(self, inputs):
        # Convert to vector
        return np.concatenate([...])

    def compute_synchrony(self, user_vector, ai_vector):
        return cosine_similarity([user_vector], [ai_vector])[0][0]

    def adjust_ai_state(self, user_vector, current_sync):
        # Optimize with Adam
        optimizer = torch.optim.Adam(self.ai_state.parameters())
        # Loop to minimize loss
        return updated_ai_vector

    def generate_response(self, updated_ai_vector, current_emotion):
        # Compassionate output
        if current_emotion['sad'] > 0.5:
            return "I sense you're feeling down. Let's breathe together."

    def run_loop(self):
        while True:
            # Full loop execution

Train on datasets from neuroscience studies for authenticity [20].

Implementation Considerations

Hardware: Use ROS for robots or apps with phone sensors.

ML: PyTorch with datasets like EMOTIC.

Ethics: Privacy checks and bias mitigation.

Challenges: Real-time edge computing; Ray suggests quantum enhancements. Evaluation: User empathy reports and HRV reductions.

Ray MI-Sync-AI bridges cognitive neuroscience and AI ethics, creating agents capable of empathetic responsiveness without relying solely on pre-programmed rules. It introduces a framework to model synthetic inter-brain coupling, enabling new insights into human social cognition and machine-human synergy. The aim AI systems for child-care, elder-care, or therapeutic support that align emotionally and cognitively with humans. Robots can engage in emotionally intelligent interactions, mimicking natural human bonding cues. Agents synchronize with human learners for better tutoring, social coaching, or collaborative problem-solving. The Mother–Infant Inter-brain Synchrony Algorithm (MI-Sync-AI) is a cutting-edge framework designed to emulate the neural and emotional alignment observed in human mother–infant dyads. Drawing from neuroscience, developmental psychology, and computational modeling, MI-Sync-AI enables Compassionate AI systems to anticipate, resonate with, and respond to another agent’s cognitive and emotional states in real time.

MI-Sync-AI Framework

MI-Sync-AI leverages multimodal data—neural signals (e.g., EEG), physiological metrics (e.g., heart rate variability), and behavioral cues (e.g., gaze, facial expressions)—to achieve empathetic responsiveness. The core mechanism involves computing inter-brain synchrony metrics, such as Phase Locking Value (PLV): \[ \text{PLV} = \Bigg|\frac{1}{N} \sum_{n=1}^{N} e^{i(\phi_m(n)-\phi_i(n))}\Bigg| \] where \(\phi_m\) and \(\phi_i\) represent phase signals from two agents (e.g., human and AI or human dyads), and \(N\) is the number of samples.This is complemented by cross-attention fusion: \[ \mathbf{Z}_{fusion} = \text{CrossAttention}(\mathbf{Z}_{neural}, \mathbf{Z}_{physio}, \mathbf{Z}_{behavior}) \] enabling the AI to integrate diverse inputs for real-time emotional alignment.

The framework moves beyond static rule-based systems by dynamically modeling synthetic inter-brain coupling, simulating the neural synchrony observed in human social cognition. This allows AI agents to adaptively respond to emotional and cognitive states, fostering trust and connection in human-machine interactions.

Preprocessing Pipeline Steps

Clock Sync & Alignment (Hardware Timestamps or NTP)

Purpose: To temporally align multimodal data streams (EEG, PPG/ECG, video/audio) for accurate synchrony analysis.

Method: Use hardware timestamps (e.g., shared trigger pulses) for sub-millisecond precision in lab settings, or Network Time Protocol (NTP) for millisecond accuracy in naturalistic environments like home pilots. This ensures metrics like Phase Locking Value (PLV) reflect true neural alignment (Point 4: \(\text{PLV} = \Bigg|\frac{1}{N} \sum_{n=1}^{N} e^{i(\phi_m(n)-\phi_i(n))}\Bigg|\)).

Example: A trigger pulse aligns mother-infant EEG channels within 1 ms, critical for computing PLV during play interactions.

Filtering: EEG 0.5–45 Hz, Notch; PPG/ECG Bandpass

Purpose: To remove noise and irrelevant frequencies, preserving δ, θ, α, and β bands for EEG and HR/HRV for PPG/ECG (Points 3, 6).

Method: Apply a bandpass filter (0.5–45 Hz) and 50/60 Hz notch for EEG to isolate neural activity and remove power line interference. For PPG/ECG, use a 0.5–10 Hz (PPG) or 0.5–40 Hz (ECG) bandpass to capture pulse waves or QRS complexes. Tools like MNE-Python or EEGLAB are used.

Example: Filtering EEG to 0.5–45 Hz isolates θ-band synchrony during role-switching, enhancing wPLI accuracy.

Artifact Rejection: Accelerometer-Guided ICA, Wavelet Denoising

Purpose: To remove artifacts (e.g., eye blinks, motion) from EEG and physiological signals for reliable synchrony metrics (Points 4–6).

Method: Use Independent Component Analysis (ICA) with accelerometer data to identify and remove motion-related EEG components. Apply wavelet denoising (e.g., Daubechies wavelet) to smooth transient noise in EEG/PPG, preserving signal structure.

Example: ICA removes blink artifacts from frontal EEG, while wavelet denoising smooths PPG during breastfeeding, improving HRV correlation.

Segmentation: 8–12 s Windows, Stride 2 s; 30–60 s for Context

Purpose: To divide data into segments for real-time analysis and context-aware modeling (Point 12).

Method: Use 8–12 s windows with a 2 s stride for dynamic synchrony analysis (e.g., PLV/wPLI). Extract 30–60 s windows for contextual tasks like distress detection in the Still-Face Paradigm (Point 7). Implement via NumPy or MATLAB.

Example: An 8 s window captures θ-band synchrony during play, while a 30 s window analyzes distress episodes.

Normalization & Baseline Correction: Per-Session Z-Score

Purpose: To standardize data across sessions/subjects for consistent model input (Points 8, 11).

Method: Apply z-scoring (\( z(t) = \frac{x(t) – \mu}{\sigma} \)) per session to normalize EEG/PPG amplitudes. Subtract baseline (e.g., rest period mean) to remove drift.

Example: Z-scoring EEG ensures comparable θ-band amplitudes across dyads, while baseline correction removes PPG drift during breastfeeding.

Applications

  • Child-Care Support: MI-Sync-AI enables AI-driven caregiving systems to detect and respond to infants’ emotional needs, mimicking maternal attunement for improved bonding and reduced distress.
  • Elderly Care: AI companions use MI-Sync-AI to synchronize with elderly users’ emotional states, combating loneliness through empathetic interactions.
  • Therapeutic Support: In mental health settings, MI-Sync-AI facilitates emotionally intelligent therapy bots that align with patients’ emotional cues, enhancing therapeutic outcomes.
  • Educational Tutoring: AI tutors synchronize with learners’ cognitive and emotional states, optimizing engagement and learning efficiency through personalized coaching.
  • Social Coaching and Collaboration: MI-Sync-AI supports AI agents in guiding social interactions or collaborative problem-solving, fostering empathy and cooperation in group settings.

Significance

MI-Sync-AI bridges cognitive neuroscience and AI ethics, creating a new paradigm for human-machine synergy. By modeling synthetic inter-brain coupling, it offers profound insights into human social cognition, enabling AI systems to engage in emotionally intelligent interactions that align with human emotional and cognitive needs. Aligned with Sri Amit Ray’s Compassionate AI principles, MI-Sync-AI prioritizes empathy, non-violence, and human rights, reducing suffering by fostering meaningful connections in child-care, elder-care, therapy, education, and social collaboration.[1] Its ethical design ensures privacy, inclusivity, and safety, paving the way for AI that uplifts humanity and all beings.

Neuroscience of Inter-brain Synchrony

Example Use Case

  • Collaborative Robotics: Two AI-driven robots carrying a fragile object can maintain temporal and action alignment so one robot anticipates the other’s movement, reducing the chance of collision or dropping the object.
  • Social AI Agents: In therapy or education, AI agents can mirror emotional states or conversational timing of human participants to create more natural and supportive interactions.
  • Inter-brain synchrony refers to the temporal alignment of neural signals between two interacting individuals, measured via hyperscanning with EEG or fNIRS. In mother–infant dyads, synchrony occurs in δ, θ, α, and β bands, correlating with behavioral attunement and joint attention.
  • Measurement: Metrics like Phase Locking Value (PLV) and Weighted Phase Lag Index (wPLI) quantify synchrony: \[ \text{PLV} = \Bigg|\frac{1}{N} \sum_{n=1}^{N} e^{i(\phi_m(n)-\phi_i(n))}\Bigg| \] \[ \text{wPLI} = \frac{|\sum_n \text{Im}\{C_{mn}\}|}{\sum_n |\text{Im}\{C_{mn}\}|} \] These metrics detect phase alignment in regions like the prefrontal cortex, reflecting emotional bonding during cooperative tasks like feeding.
  • Significance: Synchrony in θ and α bands predicts secure attachment, informing MI-Sync-AI’s ability to detect attunement epochs for compassionate guidance (Point 2).

Developmental and Behavioral Foundations

The Face-to-Face Still-Face Paradigm (SFP) demonstrates infants’ sensitivity to caregiver responsiveness. During the still-face phase, infants show reduced gaze, negative affect, and physiological changes (e.g., increased HRV) due to disrupted attunement.

Relevance: SFP validates behavioral synchrony metrics (e.g., gaze, facial AUs) used in MI-Sync-AI (Point 7). It supports clinical applications by identifying distress episodes, enabling interventions to enhance socio-emotional development (Point 17).

Operational Definition for MI-Sync-AI

An attunement epoch is a temporal window where multimodal evidence—neural (PLV/wPLI > 0.6), physiological (correlated HRV), and behavioral (gaze, prosody)—exceeds validated thresholds, corroborated by markers like smiling or eye contact.

Implementation: The algorithm uses sliding windows (8–12 s, Point 12) and cross-attention transformers (Point 8: \(\mathbf{Z}_{fusion} = \text{CrossAttention}(\mathbf{Z}_m, \mathbf{Z}_i, \mathbf{Z}_{physio}, \mathbf{Z}_{behavior})\)) to detect these epochs, enabling real-time compassionate prompts.

Sensors & Modalities

Neural: Dual EEG hyperscanning (mother: 32–64 channels; infant: 8–16 channels, wearable). Measures δ, θ, α, β synchrony.

Physiological: PPG/ECG, respiration, GSR. Captures HRV and respiration synchrony.

Behavioral: Video (gaze, AUs), audio (prosody, cry), touch sensors. Validates attunement via smiling, eye contact.

Context: Feeding/sleep metadata, caregiver reports, consent flags. Ensures ethical data use (Point 18).

Constraints: Infant equipment must be soft, non-invasive, and safe (Point 15). Robust clock protocols (hardware timestamps/NTP) ensure multimodal alignment (Point 12).

References:

  1. Ray, Amit. “Calling for a Compassionate AI Movement.” AmitRay.com, 25 June 2023, https://amitray.com/calling-for-a-compassionate-ai-movement/.
  2. Leong, Victoria, et al. “Brains in Sync: Practical Guideline for Parent–Infant EEG During Naturalistic Interaction.” Frontiers in Psychology, vol. 13, 2022, https://doi.org/10.3389/fpsyg.2022.833112.
  3. Clackson, Keira, et al. “Inter-brain Substrates of Role Switching During Mother–Child Interaction.” Scientific Reports, vol. 14, 2024, https://doi.org/10.1038/s41598-024-57314-3.
  4. Mesman, Judi, et al. “The Many Faces of the Still-Face Paradigm: A Review and Meta-Analysis.” Developmental Review, vol. 29, no. 2, 2009, pp. 120–162, https://doi.org/10.1016/j.dr.2009.02.001.
  5. Matsunaga, Masako, et al. “Inter-brain Synchrony During Mother–Infant Interactive Parenting in 3–4-Month-Olds.” Scientific Reports, vol. 13, 2023, https://doi.org/10.1038/s41598-023-49420-2.
  6. Ben-Yakov, Aya, et al. “Multimodal Analysis of Mother–Child Interaction Combining Several Interaction Modalities.” Scientific Reports, vol. 15, 2025 (forthcoming), https://doi.org/10.1038/s41598-025-90310-x.
  7. D’Mello, Sidney, et al. “Affective Computing in Education: Enhancing Learning Through Emotion-Aware Systems.” Educational Psychology Review, vol. 30, no. 3, 2018, pp. 787–819, https://doi.org/10.1007/s10648-017-9430-0.
  8. Ray, Amit. “Compassionate AI and Social Robotics.” AmitRay.com, 2023, https://amitray.com/compassionate-ai-social-robotics/.
  9. Vaswani, Ashish, et al. “Attention Is All You Need.” Advances in Neural Information Processing Systems, vol. 30, 2017, https://arxiv.org/abs/1706.03762.
  10. Kosti, Ronak, et al. “EMOTIC: Emotions in Context Dataset.” 2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops, 2017, pp. 2309–2317, https://doi.org/10.1109/CVPRW.2017.284.
  11. Andreasson, Per, et al. “Haptic Feedback in Human-Robot Interaction.” Journal of Robotics, vol. 2018, 2018, https://doi.org/10.1155/2018/6045871.
  12. McCraty, Rollin, and Maria Zayas. “Cardiac Coherence, Self-Regulation, Autonomic Stability, and Psychosocial Well-Being.” Frontiers in Psychology, vol. 5, 2014, https://doi.org/10.3389/fpsyg.2014.01090.
  13. Lachaux, Jean-Philippe, et al. “Measuring Phase Synchrony in Brain Signals.” Human Brain Mapping, vol. 8, no. 4, 1999, pp. 194–208, https://doi.org/10.1002/(SICI)1097-0193(1999)8:4<194::AID-HBM4>3.0.CO;2-C.
  14. Feldman, Ruth. “Parent–Infant Synchrony: A Biobehavioral Model.” Current Directions in Psychological Science, vol. 21, no. 3, 2012, pp. 131–136, https://doi.org/10.1177/0963721412443457.
  15. Quigley, Morgan, et al. “ROS: An Open-Source Robot Operating System.” ICRA Workshop on Open Source Software, 2009, http://www.willowgarage.com/sites/default/files/icraoss09-ROS.pdf.
  16. Kosti, Ronak, et al. “EMOTIC Dataset.” IEEE CVPRW, 2017, https://doi.org/10.1109/CVPRW.2017.284.
  17. Jobin, Anna, et al. “The Global Landscape of AI Ethics Guidelines.” Nature Machine Intelligence, vol. 1, no. 9, 2019, pp. 389–399, https://doi.org/10.1038/s42256-019-0088-2.
  18. Ray, Amit. “Quantum-Inspired AI for Compassionate Systems.” AmitRay.com, 2024, https://amitray.com/quantum-inspired-ai-compassionate-systems/.
  19. Reindl, Vanessa, et al. “Brain-to-Brain Synchrony in Parent–Child Dyads.” Developmental Cognitive Neuroscience, vol. 48, 2021, https://doi.org/10.1016/j.dcn.2021.100928.
Tags: