Collaborative intelligence refers to systems where humans and AI agents work jointly, each contributing distinct cognitive strengths. The concept builds on sociotechnical systems theory and hybrid intelligence research, emphasizing human oversight, ethical reasoning, and contextual grounding, while AI provides computation, pattern detection, and scale. Research focuses on task allocation, interface design, and trust calibration between agents.

Large language models like GPT-4 operate by predicting likely word sequences based on probabilistic associations across large corpora. They excel at linguistic generation, zero- or few-shot generalization, and latent pattern recognition. However, they lack grounded reasoning, access to live knowledge unless integrated with external tools (e.g., RAG systems), and do not possess theory of mind or stable memory. Known limitations include hallucination, superficial coherence, and vulnerability to prompt manipulation.