Last week I participated in a short course on collective intelligence at Santa Fe Institute (SFI). I’m engaged in collective intelligence, leadership and development from an adult development perspective, so I thought it could be interesting to get insights from the field of complexity studies that SFI engages in. One could argue that collective intelligence is an emergent phenomenon from individuals coordinating and acting together. And this doesn’t necessarily imply humans but could also involve insects, birds, cells and any living organism – as well as artificial intelligence.
The course took place over three days at Santa Fe Institute and online (where I participated) mainly through presentations and some panel debates. Collective intelligence, most of the presenters argued, is more relevant than ever due to the complex challenges of our time and examples of the opposite which could be referred to as groupthink or collective stupidity. But what do we mean by collective intelligence (what is intelligence?), which assumptions or first-order principles do we rest upon when exploring this topic?
Collective intelligence can be understood as individuals coordinating their behaviour with other individuals to form coherent collective decision-making and behaviour. You are an individual, but you are also made up of various parts in organs, tissue and cells that work together, so the question can be approached from several levels.
Coordinating action and collective decision-making require communication to take place between the members or parts. This can be in the form of visible cues such as bodily movements like a stock of fish avoiding a predator or aiming for a target or fireflies coordinating temporally when they light up. It can also be in the form of chemical signals, pheromones, that ants use to coordinate their movements when foraging for food. A shorter path to the food source means more ant traffic intensity, which implies a denser pheromone trail inviting more ants to choose the shorter way.
Another way of passing on information could be in the form of stigmergy, introduced by Guy Theraulaz and coined by Pierre-Paul Grassé, which is when an insect’s action is a stimulus to another insect’s following action. For instance, if a social wasp encounters a half-built comb with an apparent “hole” in it, i.e. a position with many adjacent walls, this will be a stimulus for this to build a new cell.
These are also relevant for us humans as we may assume that communication is verbal in nature. Is passing an unwashed coffee cup in the sink or seeing some litter on the street enough stimulus for us to act? It could also be a half-developed theory that invites for completion. For us, these actions from non-verbal stimuli in others’ actions can be more or less conscious and we may get drawn into a collective behaviour we never intended. Design can be understood in terms of stigmergy as a way to encourage certain behaviours.
From an information theoretical perspective, a common question is how to communicate and store information with minimal loss or distortion of information in the process. This has been exemplified in the work with LLMs, or Large Language Models, such as Chat GPT. This is trained on a vast amount of data, but the internal representation of this, the neural network, much be a very condensed representation of this. How well and accurately it can perform is then very much a function of how it can represent this training text corpus of linguistic data.
One question here that relates to the second half of the term collective intelligence is how we assess how intelligent an AI model is. Which measures are relevant to use? Typically, we only assess the output of the model and this was also the prevailing perspective in the other studies and presentations. There were several discussions and examples of mathematical modelling and simulations of collective behaviour based on the appearance and behaviour of the collectives. But there were very few arguments and insights around what’s going on inside the individuals.
One such exception was the study of the perspectives of fish presented by Iain Couzin which aimed to take the (visual) perspective of the fish. He also argues that computation can be understood as an emergent property of the entire network of fish and that this network can sense environmental changes before any individual can. Computation is made by the network in a similar way as individual fish are practising “voting with your feet”. If a sufficient amount of fish (or influential fish individuals) aims for a target, the entire school will at some point choose that destination. This has similarities with decision-making in mammals’ brains.
Seeing the entire school of fish as a collective with emergent computational, and thus cognitive, properties is a very useful perspective to be applied to us humans as well. How can we consider an organisation, a movement, a country or our species as a collective with emergent cognitive properties and shared identity? There is, however, a difference between studying human and non-human collectives in that we have access to the interior of the former. We can experience human collectives from the inside whereas the non-human collectives were typically addressed from an outside third-person perspective as black boxes to be modelled as simply as possible. But can we as individuals really grasp and reflect on our collective cognition and behaviour?
The only presentation that explicitly addressed social and collective intelligence in human social settings and addressed the interior was by Mirta Galesic who introduced a framework, or possibly a metatheory, for addressing collective adaptation. This had the three main components of social integration strategies: social integration strategies (how we in a psychological sense conceptualise our environment), social environments (addressing the relational and collective aspects of adapting) and problem structures (comprising the challenges we face), three aspects that I find quite generic (psychological, social and physical/structural). It was argued here that a social system needs to be understood in relation to its history to grasp its current behaviour and response to different situations, which reflects a processual awareness of social systems.
Concluding, the quality of the presentations was high with leading researchers in the respective fields of complexity studies. The theoretical level was also challengingly high, it felt more like a conference in the respective research fields than a course. I think it was very affordable with only $100 for online participation and $300 for on-site participation.
Participants in the collective intelligence course at Santa Fe Institute.
I see, however, some potential for further development of the course design. The level of interactivity was restricted to short Q&As following the presentations and to the public Zoom chat for the online participants (except for a poster session and off-schedule interactions for the on-site participants). From my experience with well-facilitated online events where the participants are invited to engage and interact more (in the Inner Development Goals initiative and the CADRA project), I felt the opportunities to exercise collective intelligence among the participants could be developed for the coming events. Some glimpses from the audience interaction revealed that there was plenty of competence from the participants that weren’t properly harnessed.
I also recognised that my field of adult development psychology could significantly contribute to collective intelligence, not only for human systems but also for assessing the cognitive performance of other species and insights around first-order principles that were discussed here but far from established. I hope to get back with further insights on how to integrate these fields in a useful way.
Many thanks to Santa Fe Institute, the arrangers and all presenters!