Our Thesis: Social Intelligence Framework for AI

0%*

of interaction is non-verbal.

Standard AI is socially blind. It misses your personality, the tone of your voice, and the look on your face. Citrus captures and acts on the 93% of the interaction that happens between the lines.

*Based on the Mehrabian non-verbal communication study.

As per social psychologist Kurt Lewin, behavior is shaped by both one's personality and the environment around them. We can use this framework to analyze how humans and AI behave as a human-AI interaction unfolds.

Lewin Social Psychologist Kurt Lewin, 1936
Behavior = f(Personality, Environment)

Applying Lewin's equation to the human side of the interaction, users' behavior in the form of multimodal signals is a function of both their personality and the conversation setting.

Humans
Multimodal signals, emotions, and behavior = f(Personality, Conversation Setting)

On the AI side of the interaction, AI's behavior, including activations, is also a reflection of its own training and the conversation context.

LLMs
Activations, generated behavior = f(Model Identity, Prompt & Conversation History)

In human-AI conversations, AI's Conversation History partially comprises user's Behavior, so these equations loop and become a dynamic system. This means that our framework can be used to understand how AI influences humans, as well as how humans influence AI.

Combining these perspectives gives Citrus a social-intelligence framework for understanding both people and AI, modeling interaction context, and adapting AI behavior in real time.