Glossary
Quick definitions for the terms we use.
A B C D E F G H I J K L M N O P Q R S T U V W X Y Z
A
Affect
Patterns in bodily and behavioral signals that reflect moment-to-moment state.
How we use it: input to EmPath that produces low-dimensional vectors for conditioning AI behavior.
Affect(ive) Input / Rich Affect Input
Multimodal signals (physio, motion, context) ingested together for greater state fidelity than any single channel.
Afferent States
State estimates inferred from incoming (afferent) human signals—e.g., arousal, approximate valence, tension, and engagement—before any downstream policy logic is applied.
Afferent State Vectors (ASVs)
Vectorized representation of afferent states. Today we claim: arousal, approximate valence, tension (primarily muscular), and engagement (attention/effort proxy). Roadmap: task-specific and emotion-specific embeddings with scale.
AI Agent Policy
The decision function an agent uses to choose actions. EmPath uses the database of biophysiological signals, afferent state vectors, and probabilistic emotional states to provide policy-conditioning weights and guardrails that modulate this policy in real time.
Atopia
ShadowMaker’s modular sensor platform (2-sEMG + 9-axis IMU per module; accessories for different placements). Used for games, HCI, and research prototypes. Ships data to a host for real-time + post analysis.
Arousal
Physiological activation level (low↔high). Our estimate blends signals (e.g., EDA, HR/HRV, motion artifacts) with context.
B
Bottom-Up Affect
Real-time ASVs flowing from human/context signals that modulate the agent’s policy weights dynamically.
C
Causal Maps
Structured graphs linking inputs (stimuli, context, biosignals), intermediate states (ASVs), and outputs (agent behavior). Used to reason about “why” an agent acted a certain way and to debug/verify safety.
E
E³T
Emotion, Empathy, and Ethics Test. Our internal/external evaluation framework that stress-tests agent behavior across curated scenario batteries; produces scores and gating thresholds for deployment.
EDA
Electrodermal Activity. Skin conductance changes tied to sympathetic activation. We use it as one contributor to arousal estimation.
ECG
Electrocardiogram; electrocardiography. Electrical activity of the heart. Used to derive heart rate and HRV as arousal/engagement features (non-diagnostic).
EEG
Electrical brain activity. Not in current baseline; considered for research integrations where appropriate.
Emotion-Weighted Dataset
Training/evaluation data where samples include ASVs, context labels, and outcome signals; weights emphasize ethically relevant states and edge cases to shape safer policies.
Emotional State
We expose operational states useful for interaction: arousal, valence, tension, and engagement. These are mapped to task-specific labels and personally calibration.
Emotional State Pathways
Graphical/temporal models of how ASVs evolve during scenarios (e.g., startle → tension spike → recovery). Useful for prediction and guardrail timing.
EmPath
Our middleware (“affect layer for AI”) that converts multimodal inputs into affective-state vectors and policy-conditioning weights that modulate an agent’s behavior. Model-agnostic.
Engagement
An attention/effort proxy inferred from multimodal patterns (e.g., HRV trends, motion stability, task performance correlates).
Ethical Conditioning
Applying policy weights, constraints, and guardrails (from E³T + EmPath) so agents behave within ethical bounds under real-time affect.
Eye-Tracking (roadmap)
Gaze direction, fixations, micro-saccades as attention/arousal features. Planned integrations via standard trackers.
F
Facial Expression Tracking (roadmap)
Computer-vision features (AU-style or learned embeddings) to augment ASVs; used with strict privacy controls.
H
Heart Rate (HR)
Beats per minute derived from ECG or PPG. Input to arousal/engagement estimates.
Heart Rate Variability (HRV)
Beat-to-beat interval variation. Lower short-term HRV can correlate with higher arousal; we use it as one feature.
Human Biosignals
Signals produced by the body.
Used now: sEMG, IMU (9-axis), ECG, EDA, PPG (where available), respiration (selected setups), skin temperature.
Planned: eye-tracking, facial expression tracking, keystroke/mouse dynamics, voice prosody, EEG (research).
I
IMU (9-axis)
Accelerometer + gyroscope + magnetometer. We use it for motion/pose, tremor, and artifact rejection.
K
Keyboard Use Tracking (roadmap)
Keystroke timing/dynamics as an engagement/fatigue proxy in desktop contexts; privacy-respecting and opt-in only.
M
Muscle Tension (our usage)
sEMG-derived activation magnitude and patterns (e.g., upper traps) as a proxy for tension or stress response in context.
N
Nexus, The
Our flagship interactive game / database building tool: a first-person experience controlled by wearable signals, showcasing EmPath’s real-time conditioning.
P
PESs
Probabilistic Emotional States. ASVs that are paired with emotion-specific contexts are associated at scale to create PESs, which are mathematical representations of contextualized emotional states. ASVs and PESs populate the database and provide reference nodes that AI agents pattern match against in order to predict probable outcomes for user queries / scenarios.
PPG
Photoplethysmography. Optical pulse signal (e.g., finger/camera sensors). Alternate path to HR/HRV where ECG isn’t available.
R
Respiration
Breathing rate/depth features (belt, camera, or derived) used in arousal/tension estimation where available.
S
Scenario Batteries (E³T)
Curated sets of test scenes (edge-cases, ambiguity, moral hazards) used to score and certify agent behavior under affect-conditioning.
sEMG (Surface Electromyography)
Electrical activity of muscles at the skin surface. Primary input to tension and interaction cues (e.g., intentional flex, posture).
Skin Temperature
Peripheral temperature trends; slow but useful complement to arousal/fatigue modeling.
Stimulus / Response
A stimulus is an input presented to a human and/or agent; response is the measured change in signals, ASVs, or actions. Used in calibration and E³T tests.
Stress Response (our usage)
A context-bound pattern across signals (e.g., spikes in sEMG/EDA, HR changes) indicating heightened activation.
T
Tension
Primarily muscular activation patterns (sEMG) and posture-related load; also incorporates co-occurring arousal features.
Top-Down Rules
Explicit constraints (policies, red lines) applied to an agent regardless of affect; combined with bottom-up affect for safer behavior.
Training Data
Opt-in, privacy-controlled recordings of signals, context, and outcomes used to improve mappings from signals → ASVs and ASVs → safe policies.
V
Valence (Approximate)
A coarse estimate of pleasant↔unpleasant orientation derived from patterns across modalities and context; not a claim of precise emotion or clinical mood.