KOI-Pond: The creation of a synthetic deme
TLDR: Metagov is experimenting with whether a community’s myths, knowledge, practices, rules and norms can be activated through LLMs and other technologies by creating a synthetic deme that mirrors its natural counterpart.
The KOI-Pond is Metagov’s experiment in creating a synthetic approximation of our existing group, including our collective knowledge, practices, myths, interests and rules. To achieve this, we are deploying an infrastructure first developed by BlockScience called Knowledge Organisation Infrastructure. Among other uses, KOI-Pond is a vehicle through which we can make use of the otherwise unshaped intelligence that resides in an LLM. The harnessing of an LLM by a group through its self-governed software infrastructure is a process of ‘re-deme-ification’. The purpose of this article is to explain what I mean by ‘re-deme-ificaton’, which requires a short summary of a field called Cultural Science. If my framing of KOI-Pond is correct then our experiment provides a basis for groups of any kind — organisations, communities, political parties, nations — to access new capabilities via generative AI that align with their existing culture and rules. To understand “re-deme-ification” I first need to define “demes” and “de-deme-ication”.
What is a deme?
Cultural Science, first formulated by cultural studies luminary John Hartley and developed with innovation economist Jason Potts[1], commences from the insight that culture makes groups and groups make knowledge. The cultural science word for a group is ‘deme’ (from the Greek dêmos, meaning a political group or population or closely-related organisms). Demes are ‘“we”-groups’ that develop through social learning, shared language, culture and ideas. When they collide, demes can produce innovation or descend into conflict. Cultural science described natural demes and their evolutionary function, bringing together cultural studies and evolutionary economics to argue that
· “the evolutionary function of culture is to create and sustain groups;
· the cultural function of groups is to make knowledge and act accordingly, while inter-group competitive conflict is a productive force for newness and innovation; and
· knowledge is the ‘currency’ of both economics (growth) and politics (contestation)” (Hartley 2020)
In a recent article, Jason Potts draws on Stephen Wolfram’s insight that LLMs are “embeddings of embeddings” to argue that they commit ‘de-demification’. An LLM embedding is a mathematical process that captures semantic similarities in a low-dimensional space. Potts argues that “[t]hese new A.I.s made of embeddings of human language (the training set), they are us. And that training set, as language, is an embedding on human social, biological and physical reality”. The Cultural Science insight is that culture itself is an embedding, in that our semantic universe is coded into written language and in turn into our cultural mini-worlds (demes). However, while LLMs are trained on that the technical process of a mathematical embedding, they do so across a vast array of culture. The result is that knowledge is de-deme-ified:
A trained LLM has ingested an enormously vast amount of culture. But it hasn’t experienced culture, in the way humans do. Rather, it has created an embedding of a vast sea of cultural elements. In the theory of cultural science (Hartley and Potts 2014), culture creates groups (demes) and groups create knowledge. That knowledge, de-deme-ified, has been reprocessed into an embedding (Potts 2024)
An LLM only becomes knowledge when it is prompted. Until that point it lies latent, not knowing what it knows. Potts’s conclusion is that we can now create what Hartley first dubbed “The Culturetron” (like a particle-collider for culture), “a powerful instrument to smash cultural elements together and study its properties in very precise ways”.
What is KOI Pond?
Metagov is a place where people come to share knowledge and collaborate on “a governance layer for the internet that is empowering, creative, interconnected, and accountable”. Knowledge-sharing currently occurs through seminars, written-outputs and discussion (meetings or chat), and is integral to our development of new projects. Metagov’s KOI-Pond aims for Metagov’s collective knowledge to be accessible and deployable through a wider network of people and their inventions.
KOI stands for Knowledge Organisation Infrastructure. Our KOI-Pond combines the technologies of a knowledge management system (KMS) and reference IDs (RID), resulting in a graph-based database to support relationships between knowledge objects, users and groups within Metagov. The KOI-Pond can be accessed by a large language model (LLM) or other technologies. We intend for Metagov’s KOI to be guided by Metagov and imbued with its priorities, which will become clearer through the practice of incorporating knowledge and rules into KOI. Therefore, this project is much more than setting up an LLM for Metagov’s use; it is an experiment in 1) whether a community’s priorities can be reflected in the rules, sorting and boundaries that define KOI, and 2) whether this produces a useful technology that can propagate the knowledge which flows through and from Metagov. We hope to eventually observe knowledge-sharing between BlockScience and Metagov through the peer-to-peer KOI architecture (for more on this see Zargham & Ben-Meir).
The technical team is currently working on the reference ID (RID) system that will provide our KOI with metadata on objects that live outside of it. Our expectation is that the RID system will eventually allow us to steer our KOI by weighting knowledge objects or specifying their relevance to certain uses (for instance, onboarding or for a specific team’s work). In the meantime, we are designing methods that will enable Metagov members to evaluate our KOI at different points in its development, including the information it has access to and the boundaries that are placed on that information via weightings, timestamps, confidentiality etc.
My hypothesis is that the KOI-Pond provides boundaries that are akin to a synthetic deme (a knowledge group). The more like Metagov’s natural deme it becomes, the more useful it will be to us, and to those seeking to interact with us.
Conclusion: Re-deme-ification
The KOI-Pond project is a first attempt at ‘re-deme-ification’, harnessing the dormant intelligence of large language models through the architecture of Metagov’s instance of a Knowledge Organisation Infrastructure called KOI-Pond. We are experimenting with how a community’s myths, knowledge, practices, rules and norms can be activated by creating a synthetic deme that mirrors its natural counterpart. By embedding Metagov’s distinct cultural DNA within the KOI-Pond, the experiment stands as a blueprint for integrating artificial intelligence into the epistemic processes of any collective, suggesting new pathways for innovation and interaction with knowledge groups.
About me
I am a Professor at RMIT University and an Australian Research Council-funded Future Fellow working on the project “Cooperation Through Code”. My research is examining permissionless systems (including public blockchains) using ethnographic methods. I am also a Research Director in Metagov.
[1] I worked for John Hartley in 2003–2006 and he introduced me to Jason Potts (now my husband and collaborator). Needless to say, I witnessed this intellectual journey first hand.
