Build intelligence that
understands people.
Physical intelligence helps machines perceive and act in the world. We build social intelligence that helps them understand what people feel and intend, remember their preferences and history, reason about relationships, and predict how interactions unfold.
Social Intelligence Model
Our approach
Read the interaction.
Connect continuous multimodal signals to understand who is expressing what, how, and in response to whom.
- Input
- Video, audio, text, and the timing and context of an interaction.
- How it works
- Align words, tone, facial expression, gaze and body movement. Distinguish participants and trace turns, responses and the sequence of events.
- Output
- Behavioral and interaction signals with timestamps, sources and confidence, ready to update the models.
Understand the person.
Across time.
Connect current signals with memory and context to build a lasting understanding of each person.
- Input
- Current signals, preferences, interaction history, relationships and context.
- How it works
- Use persistent memory to connect past experience with current signals. Update the Human Model as new evidence arrives, retaining uncertainty.
- Output
- A contextual understanding of the person’s feelings, intentions, preferences and relationships.
Explore what happens
before taking action.
Change a condition and compare how the same people or groups respond and evolve in different situations.
- Input
- The Human Model, current context and candidate responses.
- How it works
- Compare possible next responses and explore how an interaction might unfold under different choices.
- Output
- Possible reactions and interaction outcomes, together with their uncertainty.
Bring understanding
back into the world.
Choose how and when to respond, then learn from what actually happens.
- Input
- Predicted reactions, interaction goals and real-world constraints.
- How it works
- Choose what to say, when to respond, when to wait and how to adapt. Compare the person’s real response with the prediction.
- Output
- A response suited to the person and context, plus feedback for perception and persistent memory.
Use cases
Robotics & Physical AI
Sales & Customer Conversations
Learning & Training
Research & Interviews
Meetings & Collaboration
