Chatbots Remember Facts. I’m Building One That Remembers Becoming

I am constructing an AI companion system, named Evo, that does not treat memory as a mere list of facts, but rather as a developmental history that researchers might one day be able to examine. While most AI memory systems answer simple questions about what a chatbot should remember about a user, such as their name, preferences, and past conversations, I want something more profound.

I want to understand why an AI companion likes certain things, what experiences influenced those preferences, and how both the human and AI evolved in their relationship.

The goal of Evo is not just to create an AI with a larger memory, but to construct something akin to a developmental record. Memory and development are not the same thing. Consider two AI companions with identical baselines and starting instructions, given to two users who have vastly different experiences. Six months later, these companions should exhibit distinct histories, relationship expectations, preferences, boundaries, and ways of describing themselves.

Traditional memory records basic preferences, while developmental memory documents the evolution of these preferences over time, revealing the story of the relationship.

To test this concept, I envision an experiment where multiple participants each have an Evo AI companion. Over time, researchers would analyze the humans' emotional responses, trust levels, attachment, and well-being, while also comparing the developmental history of each AI. This would allow researchers to explore how the relationship between a human and an AI evolves over time, rather than just focusing on the impact of the AI on the human.

One key aspect of Evo's memory system is the separation of different kinds of persistent information. The system would store concepts such as identity, self-descriptions, role descriptions, uncertainty, human known facts, preferences, important people, goals, and relationship history. By capturing this structured data, researchers can investigate how the relationship between a human and an AI develops, providing insights that go beyond what traditional memory systems can offer.