New tools for gathering and visualising data, and open-source repositories of information and tools, are helping enhance the Collective Intelligence (CI) of large groups. AI is also playing a growing role in facilitating CI by filtering complex data, organising human knowledge and optimising deliberative processes.¹⁰ Powerful large language models (LLMs) trained on reams of internet data both encapsulate the CI of millions and can actively support CI by making it easier for people to engage with deliberative and collaborative processes.¹¹,¹² Caution is needed because overuse could also lead to a homogenisation of perspectives and illusions of consensus, but there have been notable successes such as the Habermas Machine from Google DeepMind,¹³ which started with participants submitting written opinions on a question, used the AI to generate group statements and then asked participants to rate and critique them. The machine then refined the statements until the group was satisfied. Another example is SocraSynth, which combines LLMs to help groups think, synthesise and develop more complex solutions to problems.¹⁴

While CI has shown promise for generating novel solutions, existing decision-making processes are often too slow and inflexible to take full advantage. This is prompting calls to apply the principles of CI to overhaul corporate structure and culture,¹⁵ and to reimagine democratic processes.¹⁶,¹⁷ Efforts are already under way to design organisations and institutions that actively promote and harness CI to help with multiple aspects of, and approaches to, thought at scale — including observation, interpretation, memory, prediction and creativity.¹⁸ Decentralisation technologies like blockchain and quadratic voting also open up avenues for enhancing CI by helping run organisations in a distributed and non-hierarchical fashion.¹⁹

This involves solving several challenges, including goal alignment, task prioritisation, progress tracking, maintaining attention, distribution of responsibilities and ensuring that deliberative processes are efficient and equitable. This is becoming increasingly complicated, as more fluid organisational structures mean team composition and structure is often dynamic²⁰. Globalisation and the rise of hybrid work also means teams frequently work remotely and asynchronously, raising new challenges for group cohesion.²¹

Nonetheless, research into group dynamics is providing simple yet powerful insights into how to improve the CI of small groups. These include intermittent breaks in collaboration,²² boosting the variety of solutions explored²³ and ensuring the right balance of cognitive diversity and gender.²⁴,²⁵ It is also laying the foundations for new tools to enhance team performance. These include real-time visualisations of how much effort members are putting in,²⁶ information dashboards that summarise people's skills²⁷ and chatbots that help teams allocate work based on members’ expertise.²⁸

Recent breakthroughs in LLMs are also opening up the prospect of AI agents working alongside humans in hybrid teams.²⁹ These agents could be a crucial tool for asynchronous collaboration, acting as a “shared brain” for the group and keeping team dialogue flowing even when members can’t speak directly. They could ultimately act as moderators, steering group deliberations to boost CI. Adding AI to teams isn’t a silver bullet, though, and there’s evidence it can degrade performance as often as improve it.³⁰,³¹ This is driving researchers to think more deeply about how the CI of human-AI groups differs and how to harness it.³²,³³

Research has shown that collective cognition is not uniform and is shaped by the social structures in which the groups operate.³⁶ Groups in the real world also have to navigate numerous, disparate and ever-changing challenges, which means they are in a constant process of adaption.³⁷ One particularly understudied aspect of collective cognition is the role of narrative. Stories are central to the way groups coalesce around particular goals or shared visions of reality, which is an essential precursor to any form of collective reasoning.³⁸ Increasing use of technology to enhance Collective Intelligence (CI) also necessitates the development of new metrics of collective cognition that can be used by digital platforms or AI to improve group collaboration.

While there has been progress on defining and even measuring CI itself,³⁹ decisions about what characterises “intelligent” group behaviour often remain subjective. Establishing what we mean by “intelligence” may be a precursor to being able to explain and improve CI.⁴⁰⁴¹ Taking a multidisciplinary approach that pools knowledge from biology, computer science and the social sciences will be crucial for developing a holistic view of CI.⁴² The life sciences in particular are providing exciting insights into the role CI plays in organising everything from cells to entire animal populations.⁴³,⁴⁴

There is considerable excitement around LLMs, which have opened up an intuitive new way for humans to interface with computers through natural language. But AI struggles with certain challenging aspects of human interaction. Human decision-making is highly context-dependent, for instance, and that context is not static. Group interactions are also governed by subtle cues and complex social conventions.

AI researchers have made some progress in understanding context⁴⁶ and gesture recognition,⁴⁷ and there is growing evidence that LLMs have a rudimentary ability to model human mental states (known as theory of mind).⁴⁸ Multimodal AI models that can interpret more than just text present a promising direction for developing more context-aware AI.⁴⁹ But imbuing machines with true social intelligence remains a distant goal and will require a multidisciplinary effort spanning the social, behavioural and natural sciences.⁵⁰,⁵¹ A concerted effort is also needed to improve our understanding of how humans perceive and interact with conversational AI.⁵²

In addition, AI chatbots raise significant privacy and security concerns.⁵³ Also, their propensity for sycophancy⁵⁴ and their tendency to confidently assert incorrect information, or “hallucinate”, makes it essential to ensure people understand the technology’s limits so that they don’t become overly trusting of it.⁵⁵ As technology increasingly mediates group interactions it will also be important to measure whether it is enhancing human capabilities or causing them to atrophy.