While modern artificial intelligence systems have achieved remarkable proficiency in processing language, summarizing dense documents, and generating complex code, they remain profoundly socially deaf. Human communication has never been restricted to mere vocabulary; rather, it relies heavily on a complex symphony of hesitation, confidence, uncertainty, behavioral engagement, and subtle agreement. These non-verbal and contextual cues fundamentally shape how human beings interpret interactions, yet they have largely remained invisible to contemporary AI architectures.
To bridge this critical divide, Copenhagen-based startup Interhuman AI is pioneering the development of Artificial Social Intelligence (ASI). As one of a select few European technology companies independently building proprietary foundational models rather than relying entirely on American or Asian infrastructure, Interhuman is designing systems capable of interpreting human social signals in real time. The company’s flagship technology, the newly unveiled Inter-2 model, analyzes social cues across text, audio, and video streams simultaneously, marking a significant leap forward in how machines perceive human behavior.
The Genesis of Artificial Social Intelligence
The ideological foundation of Interhuman AI took shape following the mainstream commercial explosion of Large Language Models (LLMs) such as OpenAI’s ChatGPT. While LLMs fundamentally transformed the paradigm of human-computer interaction, co-founder and CEO Paula Petcu recognized a glaring limitation: these models operated in a perceptual vacuum, entirely blind to body language, vocal modulation, facial micro-expressions, and contextual behavioral cues.
At the time of the breakthrough realization, Petcu was working in the highly regulated pharmaceutical sector, investigating ways in which artificial intelligence could better comprehend and assess patient sentiment during complex clinical trials. Realizing that clinical outcomes and user experiences depend heavily on emotional and physiological context, Petcu envisioned a multi-modal approach. By integrating traditional large language models with specialized neural networks capable of analyzing audio and video inputs from standard microphones and cameras, the system could add an entirely new tier of sensory perception.
Petcu soon connected with co-founder and Chief Revenue Officer Line Clemmensen through an industry webinar, where the pair shared mutual research interests in behavioral computing. Together, they constructed the initial technological proof-of-concept: an interactive AI coaching prototype designed to respond not just to the semantic meaning of a user’s words, but to the behavioral delivery behind them. The founding team was subsequently rounded out by Chief Operating Officer Frederik Sally, who joined the venture through the Antler startup matchmaking accelerator program, bringing vital operational discipline and product management strategy to the growing enterprise.
Deconstructing the Inter-2 Model and Real-Time Signal Detection
The introduction of the Inter-2 model represents the first iteration in a planned family of advanced social intelligence architectures. Moving far beyond static text analysis, Inter-2 evaluates a comprehensive suite of behavioral markers, including facial expressions, vocal tonality, postural shifts, gaze direction, and head gestures. In doing so, the model is engineered to detect up to 12 distinct social signals in real time, ranging from engagement and hesitation to uncertainty, confusion, agreement, and disagreement.
According to technical benchmarks released by the company, Inter-2 delivers up to four times faster inference speeds compared to earlier experimental iterations, coupled with substantial improvements in standardized evaluation benchmarks. This performance optimization is critical, as low-latency inference is a mandatory prerequisite for deploying real-time behavioral analysis at commercial scale.
"Today’s models are remarkably capable and still socially deaf," explains COO Frederik Sally. "They respond to the words, not to the person who wrote them. A correct answer delivered at the wrong moment is still a bad answer."
By translating human behavioral science into structured machine-learning parameters, Interhuman AI aims to construct systems that possess the perceptual baseline humans utilize instinctively during everyday interpersonal communication.
The High-Stakes Imperative for Socially Aware AI
As artificial intelligence systems transition rapidly from passive digital assistants into active participants in critical infrastructure, healthcare, and robotics, the absence of social intelligence introduces severe operational risks. Major global AI laboratories have traditionally concentrated their massive capital expenditures on general intelligence, mathematical reasoning, and workplace productivity optimization, frequently overlooking the psychological and social dimensions of human-machine interaction.
Interhuman leadership warns that this narrow focus creates dangerous blind spots. For instance, in automated healthcare triage systems, patients frequently call emergency lines while experiencing acute distress, fear, or frustration. An artificial intelligence call-routing system that is incapable of parsing vocal stress, hesitation, or psychological distress risks misinterpreting the caller’s true condition, potentially resulting in disastrous clinical errors or delayed emergency interventions.
Similarly, in the realm of advanced robotics, physical safety often hinges on immediate social and spatial awareness. A humanoid robot operating in a shared human workspace must be capable of recognizing non-verbal cues indicating hesitation, apprehension, or an immediate command to halt. Without the ability to process these vital contextual signals, the risk of physical harm or operational failure increases exponentially.
Navigating the Messiness of Human Behavior and Transparency
Interpreting human behavior is an inherently complex endeavor due to individual idiosyncrasies, cultural variations, and diverse communication styles. A momentary pause in conversation, for instance, can signify deep uncertainty and confusion in one context, while simply indicating thoughtful reflection in another.
To address these profound ambiguities without resorting to opaque or biased algorithmic judgments, Interhuman AI has prioritized radical model transparency. Unlike many deep-learning black boxes where outputs are generated without explanation, Inter-2 is intentionally designed to provide traceable rationales for its assessments.
"We thought it was important to include the rationale behind the model’s outputs," Sally notes. "You should be able to trace what caused a particular signal to be identified. We’re trying to make the model as transparent as possible so you can understand what was behind an assessment. As a human being, if an AI is determining something about my future, I want to be able to question it."
Addressing cultural, linguistic, generational, and neurodiverse variations in communication requires rigorous, meticulously curated datasets. Interhuman initially leveraged publicly available behavioral datasets to establish baseline problem-solving frameworks and refine its annotation pipelines. Today, the startup supplements this foundation through controlled internal data collection—with employees recording diverse conversational scenarios—and partnerships with external specialized data providers. Crucially, the company collaborates directly with certified behavioral scientists, psychologists, and expert human annotators to ensure that the training data reflects genuine psychological phenomena rather than algorithmic artifacts.
Ethical Guardrails and Regulatory Compliance in Europe
The integration of real-time behavioral analysis and social intelligence into commercial software naturally raises profound ethical questions regarding data privacy, psychological manipulation, and corporate overreach. Recognizing these vulnerabilities, Interhuman leadership emphasizes that social intelligence must be deployed strictly as a tool to enhance communication context rather than an invasive mechanism for psychological profiling or coercion.
"Social intelligence should not be about judging people or claiming to know what someone is thinking," Paula Petcu states. "It should give AI more context to understand human communication, while keeping people in control of how that information is used."
The company has already established strict ethical boundaries regarding potential commercial partnerships. Interhuman leadership notably turned down investment offers from military and defense entities interested in weaponizing or repurposing the underlying surveillance technology. Furthermore, Sally points to the proliferation of AI-enabled smart glasses as a high-risk application area, noting the ethical hazards of deploying real-time behavioral analysis on unsuspecting strangers in public spaces.
Operating out of Copenhagen places Interhuman squarely within the jurisdiction of the European Union’s rigorous regulatory ecosystem, including the General Data Protection Regulation (GDPR) and the newly enacted EU AI Act. Rather than viewing these mandates as commercial burdens, the startup’s founders view European compliance as a distinct competitive advantage. Interhuman is currently undergoing rigorous third-party security certification audits. Because the company remains relatively lean, executive leadership maintains direct visibility into client use cases, allowing them to actively audit customer applications and reject deployments that violate data privacy laws or ethical thresholds.
Sector-Agnostic Applications and Future Horizons
Interhuman delivers its technology via a flexible API, rendering the Inter-2 model sector-agnostic and accessible to developers across numerous industries. Current enterprise customers are already leveraging the API to build sophisticated AI-driven role-play applications for corporate training. These platforms allow employees to practice high-stakes scenarios—such as difficult management conversations, salary negotiations, or formal job interviews—receiving actionable, comprehensive feedback that evaluates both the semantic content and the behavioral delivery of their responses.
In the digital health sector, Interhuman is collaborating with clinical psychologists, including researchers at the Centre for Mental Health for Children and Adolescents in Denmark, to power applications designed to help parents navigate high-stress behavioral challenges with children. Additional use cases span digital health coaching, automated sales intelligence platforms, and qualitative market research, where AI interviewers can dynamically probe consumer sentiment based on real-time micro-expressions and vocal cues.
Looking ahead, the company anticipates that social intelligence will serve as an indispensable foundational layer for emerging technologies, including physical robotics, elderly care assistants, empathetic companion avatars, and voice-first computing interfaces. As developers increasingly shift away from text-based chat toward natural voice interactions with coding agents and digital systems, Interhuman is actively developing specialized voice-only models to process audio streams lacking visual data.
With plans to announce two additional proprietary models in the coming months, Interhuman AI is cementing Europe’s position at the vanguard of responsible, socially aware artificial intelligence development. By anchoring machine learning in rigorous behavioral science and unwavering ethical transparency, the Copenhagen startup is successfully teaching machines to navigate the profound complexities of the human condition.
