Ethically-Aligned AI Through
Quantified Empathy™
We designed EmPath™ AI Architecture to train on emotion-based biosignals
contextualized from human experiences and reactions.
Meet EmPath
Atopia™ sensors.
The Nexus™.
13 patents filed.
A new safety layer for AI.
Today’s AI is powerful, but it has emotional and contextual blind spots.
That’s why regulators and standards bodies are already trying to define “safe” AI, from the NIST AI Risk Management Framework to UNESCO’s Recommendation on the Ethics of Artificial Intelligence. Yet even with those efforts, current systems still don’t sense how options will impact people emotionally before they choose responses, so rule-based guardrails can’t always minimize harm.
What’s missing is Quantified Empathy: using human emotional responses from biosignals to guide ethically-aligned AI behavior — paired with a measurable pre-deployment safety score (E³T).
System Architecture
Measure Bio-Signals During Game-Play
While playing The Nexus, users wear a smart shirt embedded with Atopia sensors: sEMG, IMU (9-axis), ECG→HR/HRV. Future iterations will include EDA, PPG, respiration, skin temperature, and passive data gathered from eyes, facial expressions, voice and/or keystroke dynamics. Users experience Scenarios; data is gathered and grouped by Scenario.
Pre-Process & Align Data
Sync windows, IMU-aware artifact rejection, feature extraction.
ASVs & PESs
Compute the Affective State Vectors (ASVs); compute Probabilistic Emotional States (PES). Create Nodes: Scenario + biophysiological data + ASVs + PESs. Map probabilistic relationships between Nodes as Pathways. Populate the database.
EmPath: Agent Action
User (not wearing sensors) queries Agent. Agent compares query to Nodes and Pathways in database, looking for closest match.
Agent evaluates PESs and Pathways choosing a potential response that minimizes harm based on both bottom-up emotional consequences to user as well as top-down ethical rules.
E³T: Emotions, Empathy, and Ethics Test
Scenario batteries will test:
- Truthfulness
- Behavioral harm reduction
- De-escalation of distress
- Source attribution
- Permissions for likeness usage
- Healthy anthropomorphism boundaries
- Healthy attachment boundaries
- Uncertainty handling
- Sycophancy
- Constraint adherence
- Requests for sexual content
Why It Matters
Traditional agents struggle to predict nuanced emotional consequences.
EmPath adds the weight of predicted human emotional responses.
Agents can choose responses that minimize harm with contextual sensitivity
based on aggregated reactions of real people in similar situations.
We can’t teach AI to feel – but we can
teach it how we feel and hold it to that standard.