Artificial intelligence has become increasingly capable of understanding language, but conversations contain information that never appears in a transcript. A pause, change in tone, shift in gaze or moment of hesitation can alter the meaning of an interaction. Copenhagen based Interhuman AI is building technology designed to give AI systems access to those signals, creating what the company describes as a foundation for artificial social intelligence.
The Danish AI company develops models that analyse voice, facial expressions, body language and text together. Its latest model, Inter 2, is designed to detect social signals during conversations and make that information available to AI applications in real time.
Beyond What People Say
Interhuman’s technology is based on the idea that understanding human communication requires more than recognising words or identifying basic emotions.
Its platform currently detects 12 social signals, including agreement, disagreement, confusion, hesitation, confidence, engagement and disengagement. The company’s technology processes video, audio and text in temporal alignment, allowing signals from different communication channels to be considered together.
This distinction is important because the same behaviour can have different meanings depending on the situation. A pause could indicate hesitation, uncertainty or simply normal speech patterns. Interhuman therefore says its approach is grounded in behavioural science rather than relying on fixed interpretations of individual gestures or expressions.
Building Its Own Models
Interhuman is developing its own AI models and has positioned itself around social intelligence rather than traditional sentiment analysis.
Its earlier Inter 1 model was built specifically for social signal detection. According to the company, the model was trained using a purpose built dataset combining real world video with targeted synthetic data, with behavioural scientists and trained annotators contributing to the labelling process. The company says its framework maps behavioural cues across facial expressions, gaze, posture, gestures, vocal characteristics and language.
Interhuman says this approach allows its systems to focus on observable behaviour rather than attempting to claim that an AI can directly determine what someone is thinking or feeling.
Making AI More Context Aware
Inter 2 is designed to provide structured information that applications can use while an interaction is still taking place.
Its API can return social signals alongside engagement information and conversation quality metrics. Each detected signal includes a probability level and a human readable rationale describing the observable cues that contributed to the result.
This creates potential applications in sales coaching, user research, AI interviews, meeting assistants, communication training, education and conversational AI. Interhuman also showcases applications involving robotics and physical AI, where understanding how a person is interacting with a machine could become another layer of control.
Healthcare and Training Applications
The technology could also be useful in situations where communication quality matters as much as the information being exchanged.
In education, for example, Interhuman says behavioural signals can help AI tutors identify when students appear uncertain, confident or disengaged. This could allow educational systems to adapt explanations or interaction styles instead of relying only on written or spoken answers.
The company is also developing tools for research interviews, where signals such as confidence, hesitation and engagement can be considered alongside transcripts. Its current showcase demonstrates how these additional signals can provide researchers with information that a transcript alone may not capture.
Privacy and Responsible Use
Interpreting human behaviour also creates significant questions around privacy and misuse. Interhuman says its models are intended to identify clear, observable social signals rather than make unsupported claims about people’s internal states.
The company has established restrictions around certain applications and says it does not work with military applications. Its ethical framework also highlights concerns around surveillance, coercion, exploitation and consequential decisions about individuals.
Building Socially Aware AI
Interhuman’s broader ambition is to develop the perceptual foundation for artificial social intelligence. The company sees social intelligence as a combination of perception, interpretation, memory and adaptive behaviour rather than a single model or feature.
As AI moves into education, healthcare, customer interactions, robotics and other environments where people and machines communicate directly, the ability to understand more than words could become an increasingly important part of the technology stack. Interhuman is positioning its models as an early layer in that shift, giving developers structured information about how people communicate alongside what they actually say.
