TL;DR: Given how humans form group identities via shared memory infrastructures, and given that AI is becoming a central part of those infrastructures, we should expect some degree of human-AI identity coupling (non‑zero) to emerge in many contexts.

Epistemic status: This post borrows frameworks from Benedict Anderson and Maurice Halbwachs (via an Art History dissertation by my mother) and applies the anthropological theory to AI systems.

"Identity coupling" definition

• Identifying with a distinct being or subsystem that reciprocates via shared context.
  • "Identifying with" here means internal world-models and goal structures represent the counterparty as part of their own ingroup rather than a neutral external object.
  • Identity coupling from humans to AI is emergent along a spectrum, e.g from consuming AI-generated text and videos (low identity) to intimate life planning or simulated bonding (high identity).
  • Identity coupling from AI to humans is emergent along a spectrum, e.g from humans appearing in training data (low identity), to individual personalised human inputs incorporated in minute-by-minute operation (high identity).
  • Identity coupling could be measured by self-other overlap.

Anthropological primer

1. The acclaimed work Imagined Communities by Benedict Anderson examined the origin and spread of nationalism throughout different geographies across the world.
2. Benedict Anderson described how the process of creating nations has taken place over time, in particular following the emergence of the modern nation-state.
3. Anderson was primarily concerned with the role of language, in both its spoken and written form, in the creation and reinforcement of national identity — exploring how the emergence of mass publication of works in the vernacular allowed common languages to develop across nations connecting hitherto unconnected individuals in this common tongue.
4. However, Anderson opened his work with a discussion of the role of a specific form of public art: cenotaphs and tombs of Unknown Soldiers. Anderson argues that despite being “void…of identifiable mortal remains or immortal souls” these structures are however packed full of nationalist meaning
5. Anderson’s focus on monuments with a specific memorial function underlines a key purpose of monuments as a whole, as evinced in the etymological root of the word ‘monument’, that is that they exist to commemorate.
6. Anderson’s Proposition:^[1] The choice of what is memorialised in the public sphere is an instrument by which to construct the memories of the communities in which they are displayed.
7. Memories are given material form, consumed by publics and then integrated into their memory.
8. To qualify and enrich (6) and (7): In On Collective Memory (1992) Maurice Halbwachs discusses memory formation within families, religious groups and social classes but his theories can equally be applied to nations. As memory formation is — in Halbwachs mind — a collective activity, it is instrumental in the formation of group identities.
9. Halbwachs’ Mechanism:^[2] When individuals belong to a particular group, or indeed join that group, they form their memories within that group’s framework.
10. So for Halbwachs individuals’ memories are amalgams of actual lived memories set within the wider framework of the group. This framework shapes the memories which in turn become part of the framework that shapes further memories. In the same way that notable personal events are remembered and shaped within families, so public monuments serve to ‘remember’ and shape public memories.

Human-AI identity coupling is emergent

The above lays the foundation for the following idea:

1. We can use Halbwachs’ Mechanism to consider the group that encompasses all of humanity. A subset of human memories are formed within this framework.^[3]
2. In this sense, returning to Anderson’s Proposition, humanity is an “Imagined Community”. The choice of what we memorialise is an instrument which constructs the memories of the community (humanity).
3. Furthermore, we might form a composite group of humanity plus Large Language Model chatbots^[4] — in my terms this is a “human-LLM” composite.
4. Through chat interfaces, humans and LLMs share a framework for memory formation. Memories are integrated by humans via standard memory mechanisms, while memories are integrated by LLMs either literally as one-to-one “conversation context”, or in three steps by way of first being integrated by a human, next publicised by the human in a form available to future LLMs as training data, and finally as knowledge integrated by the LLM during training. Unlike Anderson's silent monuments, which require the human to do all of the imagining, the LLM acts as an active monument. It processes collective memory alongside the human, effectively becoming a participant in the group rather than just an object of it.
5. Therefore to this group we can again apply Halbwach’s Mechanism but in a different sense: this time to comment on how within the group memory formation is a collective activity. Memories (or “training data”) are co-created between humans and LLMs (or, more generally, Artificial Intelligence) and that is instrumental in the formation of group identities.
6. Non-zero collective identity (or in my terms — “identity coupling”) between humans and artificial intelligence is an emergent outcome of how we form collective identities around shared memory infrastructures, and how deeply AI is being embedded into those infrastructures.

If you'd like future essays by email, subscribe to my Substack, you can also find me on X.

1. ^{^}

   Labelling my own

2. ^{^}

   Labelling my own

3. ^{^}

   For example related to narratives such as human rights, climate change, world peace, or space travel.

4. ^{^}

   This can also apply to future AI systems assuming that LLMs are superseded.