Collective Intelligence: When We Know More Together Than We Think Possible

Abstract

Sociotechnical collective intelligence (STCI) represents a transformative paradigm in knowledge production, in which integrated systems of human and artificial intelligence (AI) achieve synergistic capabilities that exceed the scope of either system alone. This article establishes its philosophical foundations by bridging the gap between Michael Polanyi's classic paradox of tacit knowledge — that ‘we know more than we can express’ — and modern cognitive theories of entropic memory (which harnesses productive disorder for creativity) and associative memory (which connects disparate concepts). From this synthesis, three fundamental architectural pillars for effective STCI are derived: Augmented transactional memory, which merges human experiential, contextual, and tacit knowledge with scalable data processing and AI pattern recognition; Coordinated transactional attention, which dynamically allocates cognitive resources across the system, balancing focus and exploratory entropy to drive adaptive problem solving; Hybrid transactional reasoning, where collective judgements arise from the interaction between computational rationality and human intuitive, ethical, and evaluative judgement. Using principles from network science, we demonstrate how a system's connection topology dictates its collective effectiveness depending on the type of problem. While efficient, densely connected networks excel at quickly solving well-defined problems, inefficient, more sparse networks with longer paths are crucial for maintaining the cognitive diversity and potential for variation needed to tackle novel, ambiguous, or complex challenges. This framework reveals tensions inherent in current AI, particularly in large language models (LLMs), which often act as homogenising “central tendencies” that can suppress the diversity required by STCI. Applications are examined in: Collaborative medical diagnosis, where AI crosses references from a vast medical literature with the doctor's patient-specific tacit knowledge; AI-augmented emergency response, where human teams coordinate with AI simulations for real-time logistics and scenario modelling; Hybrid scientific research, where AI-driven data analysis and hypothesis generation are guided by scientists' theoretical vision and experimental intuition. In each domain, knowledge is shown to be a genuinely distributed property of the socio-technical network, emerging from collective processes in which controlled indeterminacy—managed randomness and diversity—becomes a source of innovation without compromising systemic reliability. The analysis also addresses fundamental ethical and epistemological challenges: the allocation of distributed responsibility in hybrid decisions, the mitigation of the risks of algorithmic homogenisation, and the management of the tension between the democratisation of knowledge and the rise of a new techno-elitism. It concludes that STCI is not simply a tool that extends human cognition, but a fundamental evolution in the architecture of knowledge itself. It signifies a shift towards heterogeneous cognitive ecosystems capable of building genuinely emergent knowledge. This requires cultivating a new form of hybrid competence: the literacy and wisdom to manage these systems in ways that preserve essential human virtues: autonomy, moral responsibility, and contextual wisdom.